Recent advancements in computer vision (CV) and large language models (LLMs) have spurred significant interest in multi-modal large language models (MLLMs), which aim to integrate visual and textual modalities for enhanced understanding and generation tasks. While much of the existing research focuses on optimizing projectors and LLMs to improve MLLM performance, a critical question remains underexplored: Has the full potential of visual features in MLLMs been realized? To address this question, we identify two key limitations in current MLLM architectures and propose vMLLM, a vision-enhanced MLLM designed to fully leverage the capabilities of visual features. vMLLM introduces two novel components: the Multi-level Aggregation Module (MAM) and the Intra- and inter-modal Enhancement Module (IEM). The MAM aggregates multi-layer features from the vision encoder, capturing both high-level semantic information and low-level spatial details, thereby enriching the visual representation. The IEM enhances visual features through intra- and inter-modal interactions, effectively suppressing irrelevant information while amplifying task-relevant features, leading to more robust multimodal understanding. We conduct extensive experiments on multiple benchmarks, evaluating vMLLM across diverse settings, including different vision encoders, training dataset scales, and varying sizes of LLMs. Our results demonstrate that vMLLM consistently achieves significant performance improvements, validating its effectiveness in harnessing the potential of visual features. These findings highlight the importance of optimizing visual feature extraction and interaction mechanisms in MLLMs, paving the way for more advanced multimodal AI systems..

This paper presents UniVST, a unified framework for localized video style transfer based on diffusion models. It operates without the need for training, offering a distinct advantage over existing diffusion methods that transfer style across entire videos. The endeavors of this paper comprise: (1) A point-matching mask propagation strategy that leverages the feature maps from the DDIM inversion. This streamlines the model's architecture by obviating the need for tracking models. (2) A training-free AdaIN-guided localized video stylization mechanism that operates at both the latent and attention levels. This balances content fidelity and style richness, mitigating the loss of localized details commonly associated with direct video stylization. (3) A sliding-window consistent smoothing scheme that harnesses optical flow within the pixel representation and refines predicted noise to update the latent space. This significantly enhances temporal consistency and diminishes artifacts in stylized video. Our proposed UniVST has been validated to be superior to existing methods in quantitative and qualitative metrics. It adeptly addresses the challenges of preserving the primary object's style while ensuring temporal consistency and detail preservation.

Interactive Segmentation (IS) segments specific objects or parts by deducing human intent from sparse input prompts. However, the sparse-to-dense mapping is ambiguous, making it challenging for users to obtain segmentations at the desired granularity and causing them to engage in trial-and-error cycles. Although existing multi-granularity IS models (e.g., SAM) alleviate the ambiguity of single-granularity methods by predicting multiple masks simultaneously, this approach has limited scalability and produces redundant results. To address this issue, we introduce a creative granularity-controllable IS paradigm that resolves ambiguity by enabling users to precisely control the segmentation granularity. Specifically, we propose a Unified Granularity Controller (UniGraCo) that supports multi-type optional granularity control signals to pursue unified control over diverse segmentation requirements, effectively overcoming the limitation of single-type control in adapting to different needs, thus boosting the system efficiency and practicality. To mitigate the excessive cost of annotating the multi-granularity masks and the corresponding granularity control signals for training UniGraCo, we construct an automated data engine capable of generating high-quality and granularity-abundant mask-granularity data pairs at low cost. To enable UniGraCo to learn unified granularity controllability in an efficient and stable manner, we further design a granularity-controllable learning strategy. This strategy leverages the generated data pairs to incrementally equip the pre-trained IS model with granularity controllability while preserving its segmentation capability. Extensive experiments on intricate scenarios at both instance and part level demonstrate that our UniGraCo has significant advantages over previous methods, highlighting its potential as a practical interactive tool.

Recent advances in Neural Architecture Search (NAS) are essentially attributed to Performance Estimation (PE), i.e., a method aims to effectively estimate an architecture. Meanwhile, Kendall's $\tau$τ is well recognized as the principled evaluation criteria for PE strategies in the literature. We argue that Kendall's $\tau$τ is not the optimal solution. Through extensive experiments and theoretical analysis, we take the initiative to reveal the problem behind the Kendall's $\tau$τ and propose a novel criterion named Minimum Keeping Ratio (MKR), which is closely connected to the final performance of NAS. It allows us to compare different PE approaches in a unified perspective, and use effective ablation studies to verify common beliefs and key differences of PE strategies. Based on the findings from MKR, we are able to derive a simple NAS method by integrating different PE strategies with random sampling. Such a method shows very strong performance in efficiency and effectiveness through extensive experiments on different challenging benchmarks. In particular, our simple random sampling NAS finds the optimal architecture in NASbenchMacro, NASbench201, and NASbench301. It is also well generalized to different search spaces (MobileNet) and tasks (semantic segmentation), finding an architecture surpasses the previous state-of-the-art architectures by 4.25 mIoU under $ 600M$600M FLOPs on ADE20K. Codes are available at https://anonymous.4open.science/r/Anonymization11264.

Albeit the scalable performance of vision transformers (ViTs), the dense computational costs undermine their position in industrial applications. Post-training quantization (PTQ), tuning ViTs with a tiny dataset and running in a low-bit format, well addresses the cost issue but unluckily bears more performance drops in lower-bit cases. In this paper, we introduce I&S-ViT, a novel method that regulates the PTQ of ViTs in an inclusive and stable fashion. I&S-ViT first identifies two issues in the PTQ of ViTs: (1) Quantization inefficiency in the prevalent log2 quantizer for post-Softmax activations; (2) Rugged and magnified loss landscape in coarse-grained quantization granularity for post-LayerNorm activations. Then, I&S-ViT addresses these issues by introducing: (1) A novel shift-uniform-log2 quantizer (SULQ) that incorporates a shift mechanism followed by uniform quantization to achieve both an inclusive domain representation and accurate distribution approximation; (2) A three-stage smooth optimization strategy (SOS) that amalgamates the strengths of channel-wise and layer-wise quantization to enable stable learning. Comprehensive evaluations across diverse vision tasks validate I&S-ViT's superiority over existing PTQ of ViTs methods, particularly in low-bit scenarios. For instance, I&S-ViT elevates the performance of W3A3 ViT-B by an impressive 50.68%.

Image deblurring is a challenging image task, which is regarded as a classical inverse problem. Deep primal-dual proximal network (DeepPDNet) is recently proposed which unrolls the Condat-Vũ primal-dual splitting algorithm as a feed-forward network and it has demonstrated excellent restoration performance. However, the feature patterns in the DeepPDNet are well manually designed and thus the network is not implemented in an efficient convolutional fashion. In this work, we revisit the DeepPDNet and extend it in three respects: i) the convolution and pooling operators as well as their associating adjoint operations are studied in the primal-dual algorithm, and then a deep convolutional primal-dual network (DeepConvPDNet) and its full variant with skips are proposed to preserve the optimization consistence of primal-dual Condat-Vũ algorithm; ii) two (cascade vs parallel) variants of the networks are designed according to the structure of convolutional kernels; iii) rather than that the blur kernels are given as prior knowledge, they can be encoded by a set of convolutional layers and deconvolutional layers for their conjugate, resulting to a full learnable deep convolutional primal-dual neural network.We investigate the proposed networks on the MNIST dataset, the grayscale and color version of BSD dataset and GoPro dataset for image deblurring. Extensive experiments are conducted to validate the performance of the proposed networks, and promising results in term of PSNR and SSIM are obtained in comparison with twelve methods including state-of-the-art methods (e.g. Restormer, DRUNet, and DeblurGAN), which validated its effectiveness.

We introduce DiffusionTrend for virtual fashion try-on, which forgoes the need for retraining diffusion models. Using advanced diffusion models, DiffusionTrend harnesses latent information rich in prior information to capture the nuances of garment details. Throughout the diffusion denoising process, these details are seamlessly integrated into the model image generation, expertly directed by a precise garment mask crafted by a lightweight and compact CNN. Although our DiffusionTrend model initially demonstrates suboptimal metric performance, our exploratory approach offers some important advantages: (1) It circumvents resource-intensive retraining of diffusion models on large datasets. (2) It eliminates the necessity for various complex and user-unfriendly model inputs. (3) It delivers a visually compelling try-on experience, underscoring the potential of training-free diffusion model. This initial foray into the application of untrained diffusion models in virtual try-on technology potentially paves the way for further exploration and refinement in this industrially and academically valuable field.

Multimodal large language models (MLLMs) demand considerable computations for inference due to the extensive parameters and the additional input tokens needed for visual information representation. Herein, we introduce Visual Tokens Withdrawal (VTW), a plug-and-play module to boost MLLMs for rapid inference. Our approach is inspired by two intriguing phenomena we have observed: (1) the attention sink phenomenon that is prevalent in LLMs also persists in MLLMs, suggesting that initial tokens and nearest tokens receive the majority of attention, while middle vision tokens garner minimal attention in deep layers; (2) the presence of information migration, which implies that visual information is transferred to subsequent text tokens within the first few layers of MLLMs. As per our findings, we conclude that vision tokens are unnecessary in the deep layers of MLLMs. Thus, we strategically withdraw them at a certain layer, enabling only text tokens to engage in subsequent layers. To pinpoint the ideal layer for VTW, we initially analyze a limited set of tiny datasets and choose the first layer that meets the Kullback-Leibler divergence criterion. Our VTW approach can cut computational overhead by over 40% across diverse multimodal tasks while maintaining performance.

Jailbreak attacks craft specific prompts or append adversarial suffixes to prompts, thereby inducing language models to generate harmful or unethical content and bypassing the model’s safety guardrails. With the recent blossom of large language models (LLMs), there’s a growing focus on jailbreak attacks to probe their safety. While current white-box attacks typically focus on meticulously identifying adversarial suffixes for specific models, their effectiveness and efficiency diminish when applied to different LLMs. In this paper, we propose a Monte Carlo Tree Search (MCTS) based Prompt Auto-generation (MPA) method to enhance the effectiveness and efficiency of attacks across various models. MPA automatically searches for and generates adversarial suffixes for valid jailbreak attacks. Specifically, we first identify a series of action candidates that could potentially trick LLMs into providing harmful responses. To streamline the exploration of adversarial suffixes, we design a prior confidence probability for each MCTS node. We then iteratively auto-generate adversarial prompts using the MCTS framework. Extensive experiments on multiple open-source models (like Llama, Gemma, and Mistral) and closed-source models (such as ChatGPT) show that our proposed MPA surpasses existing methods in search efficiency as well as attack effectiveness. The codes are available at https://github.com/KDEGroup/MPA.

Face manipulation techniques have achieved significant advances, presenting serious challenges to security and social trust. Recent works demonstrate that leveraging multimodal models can enhance the generalization and interpretability of face forgery detection. However, existing annotation approaches, whether through human labeling or direct Multimodal Large Language Model (MLLM) generation, often suffer from hallucination issues, leading to inaccurate text descriptions, especially for high-quality forgeries. To address this, we propose Face Forgery Text Generator (FFTG), a novel annotation pipeline that generates accurate text descriptions by leveraging forgery masks for initial region and type identification, followed by a comprehensive prompting strategy to guide MLLMs in reducing hallucination. We validate our approach through fine-tuning both CLIP with a three-branch training framework combining unimodal and multimodal objectives, and MLLMs with our structured annotations. Experimental results demonstrate that our method not only achieves more accurate annotations with higher region identification accuracy, but also leads to improvements in model performance across various forgery detection benchmarks. Our Codes are available in https://github.com/skJack/VLFFD.git.

Although diffusion-based image virtual try-on has made considerable progress, emerging approaches still struggle to effectively address the issue of hand occlusion (i.e., clothing regions occluded by the hand part), leading to a notable degradation of the try-on performance. To tackle this issue widely existing in real-world scenarios, we propose VTON-HandFit, leveraging the power of hand priors to reconstruct the appearance and structure for hand occlusion cases. Firstly, we tailor a Hand-Pose Aggregation Net using the ControlNet-based structure explicitly and adaptively encoding the global hand and pose priors. Besides, to fully exploit the hand-related structure and appearance information, we propose Hand-feature Disentanglement Embedding module to disentangle the hand priors into the hand structure-parametric and visual-appearance features, and customize a masked cross attention for further decoupled feature embedding. Lastly, we customize a hand-canny constraint loss to better learn the structure edge knowledge from the hand template of model image. VTON-HandFit outperforms the baselines in qualitative and quantitative evaluations on the public dataset and our self-collected hand-occlusion Handfit-3K dataset particularly for the arbitrary hand pose occlusion cases in real-world scenarios. Our project page is at: https://vton-handfit.github.io.

Despite a big leap forward in capability, multimodal large language models (MLLMs) tend to behave like a sloth in practical use, i.e., slow response and large latency. Recent efforts are devoted to building tiny MLLMs for better efficiency, but the plethora of visual tokens still used limit their actual speedup. In this paper, we propose a powerful and fast tiny MLLM called FlashSloth. Different from previous efforts, FlashSloth focuses on improving the descriptive power of visual tokens in the process of compressing their redundant semantics. In particular, FlashSloth introduces embedded visual compression designs to capture both visually salient and instruction-related image information, so as to achieving superior multimodal performance with fewer visual tokens. Extensive experiments are conducted to validate the proposed FlashSloth, and a bunch of tiny but strong MLLMs are also comprehensively compared, e.g., InternVL2, MiniCPM-V2 and Qwen2-VL. The experimental results show that compared with these advanced tiny MLLMs, our FlashSloth can greatly reduce the number of visual tokens, training memory and computation complexity while retaining high performance on various VL tasks. Our code is released at: https://github.com/codefanw/FlashSloth.

Face Restoration (FR) is a crucial area within image and video processing, focusing on reconstructing high-quality portraits from degraded inputs. Despite advancements in image FR, video FR remains relatively under-explored, primarily due to challenges related to temporal consistency, motion artifacts, and the limited availability of high-quality video data. Moreover, traditional face restoration typically prioritizes enhancing resolution and may not give as much consideration to related tasks such as facial colorization and inpainting. In this paper, we propose a novel approach for the Generalized Video Face Restoration (GVFR) task, which integrates video blind face restoration (BFR), inpainting, and colorization tasks that we empirically show to benefit each other. We present a unified framework, termed as stable video face restoration (SVFR), which leverages the generative and motion priors of Stable Video Diffusion (SVD) and incorporates task-specific information through a unified face restoration framework. A learnable task embedding is introduced to enhance task identification. Meanwhile, a novel Unified Latent Regularization (ULR) is employed to encourage the shared feature representation learning among different subtasks. To further enhance the restoration quality and temporal stability, we introduce the facial prior learning and the self-referred refinement as auxiliary strategies. The proposed framework effectively combines the complementary strengths of these tasks, enhancing temporal coherence and achieving superior restoration quality. This work advances the state-of-the-art in video FR and establishes a new paradigm for generalized video face restoration. Code and video demo are available at https://github.com/wangzhiyaoo/SVFR.git.

As one of the earliest ancient languages, Oracle Bone Script (OBS) encapsulates the cultural records and intellectual expressions of ancient civilizations. Despite the discovery of approximately 4,500 OBS characters, only about 1,600 have been deciphered. The remaining undeciphered ones, with their complex structure and abstract imagery, pose significant challenges for interpretation. To address these challenges, this paper proposes a novel two-stage semantic typography framework, named OracleFusion. In the first stage, this approach leverages the Multimodal Large Language Model (MLLM) with enhanced Spatial Awareness Reasoning (SAR) to analyze the glyph structure of the OBS character and perform visual localization of key components. In the second stage, we introduce Oracle Structural Vector Fusion (OSVF), incorporating glyph structure constraints and glyph maintenance constraints to ensure the accurate generation of semantically enriched vector fonts. This approach preserves the objective integrity of the glyph structure, offering visually enhanced representations that assist experts in deciphering OBS. Extensive qualitative and quantitative experiments demonstrate that OracleFusion outperforms state-of-the-art baseline models in terms of semantics, visual appeal, and glyph maintenance, significantly enhancing both readability and aesthetic quality. Furthermore, OracleFusion provides expert-like insights on unseen oracle characters, making it a valuable tool for advancing the decipherment of OBS.

The rapid development of AI-generated content (AIGC) technology has led to the misuse of highly realistic AI-generated images (AIGI) in spreading misinformation, posing a threat to public information security. Although existing AIGI detection techniques are generally effective, they face two issues: 1) a lack of human-verifiable explanations, and 2) a lack of generalization in the latest generation technology. To address these issues, we introduce a large-scale and comprehensive dataset, Holmes-Set, which includes the Holmes-SFTSet, an instruction-tuning dataset with explanations on whether images are AI-generated, and the Holmes-DPOSet, a human-aligned preference dataset. Our work introduces an efficient data annotation method called the Multi-Expert Jury, enhancing data generation through structured MLLM explanations and quality control via cross-model evaluation, expert defect filtering, and human preference modification. In addition, we propose Holmes Pipeline, a meticulously designed three-stage training framework comprising visual expert pre-training, supervised fine-tuning, and direct preference optimization. Holmes Pipeline adapts multimodal large language models (MLLMs) for AIGI detection while generating human-verifiable and human-aligned explanations, ultimately yielding our model AIGI-Holmes. During the inference stage, we introduce a collaborative decoding strategy that integrates the model perception of the visual expert with the semantic reasoning of MLLMs, further enhancing the generalization capabilities. Extensive experiments on three benchmarks validate the effectiveness of our AIGI-Holmes.

Recently, mixture of experts (MoE) has become a popular paradigm for achieving the trade-off between modal capacity and efficiency of multimodal large language models (MLLMs). Different from previous efforts, we are dedicated to exploring the dynamic experts in existing MLLMs and showing that a standard MLLM can also be a mixture of experts. However, achieving this target is still notoriously challenging. The well-trained MLLMs are more accustomed to the fixed pathway and a drastic change in its inference manner also greatly impedes its performance. To address these issues, we propose a novel dynamic expert routing method for existing MLLMs, termed Routing Experts (RoE), which can achieve example-dependent optimal path routing without obvious structure tweaks. Meanwhile, a new structure sparsity regularization is also introduced to force the well-trained MLLMs to learn more short-cut pathways. In addition, we also address the alignment of the training and inference of MLLMs in terms of network routing. To validate RoE, we apply it to a set of existing MLLMs, including LLaVA-1.5, LLaVA-HR and VILA, and conduct extensive experiments on a bunch of VL benchmarks. The experiment results not only show the effectiveness of our RoE in improving MLLMs' efficiency, but also yield obvious advantages over MoE-LLaVA in both performance and speed, e.g., an average performance gain of 3.3% on 5 benchmarks while being 1.61 times faster. Our code is anonymously released at https://github.com/DoubtedSteam/RoE

Despite the efficacy of network sparsity in alleviating the deployment strain of Large Language Models (LLMs), it endures significant performance degradation. Applying Low-Rank Adaptation (LoRA) to fine-tune the sparse LLMs offers an intuitive approach to counter this predicament, while it holds shortcomings include: 1) The inability to integrate LoRA weights into sparse LLMs post-training, and 2) Insufficient performance recovery at high sparsity ratios. In this paper, we introduces dynamic $\textbf{Lo}$w-rank $\textbf{S}$parse $\textbf{A}$daptation $\textbf{(LoSA)}$, a novel method that seamlessly integrates low-rank adaptation into LLM sparsity within a unified framework, thereby enhancing the performance of sparse LLMs without increasing the inference latency. In particular, LoSA dynamically sparsifies the LoRA outcomes based on the corresponding sparse weights during fine-tuning, thus guaranteeing that the LoRA module can be integrated into the sparse LLMs post-training. Besides, to achieve the optimal sparse model architecture, LoSA leverages Representation Mutual Information (RMI) as an indicator to determine the importance of layers, thereby dynamically determining the optimal layer-wise sparsity rates during fine-tuning. Predicated on this, LoSA adjusts the rank of the LoRA module based on the variability in layer-wise reconstruction errors, allocating an appropriate fine-tuning for each layer to reduce the output discrepancies between dense and sparse LLMs. Extensive experiments tell that LoSA can efficiently boost the efficacy of sparse LLMs within a few hours, without introducing any additional inferential burden. For example, LoSA reduced the perplexity of sparse LLaMA-2-7B by $\textbf{68.73}$$\downarrow$ and increased zero-shot accuracy by $\textbf{16.32}$%$\uparrow$, achieving a $\textbf{2.60$\times$}$ speedup on CPU and $\textbf{2.23$\times$}$ speedup on GPU, requiring only $\textbf{45 minutes}$ of fine-tuning on $\textbf{a single}$ NVIDIA A100 80GB GPU. Code is available at https://github.com/wzhuang-xmu/LoSA.

In existing multimodal large language models (MLLMs), image resolution plays a significant role for granular visual recognition. However, directly increasing image resolution leads to expensive computational cost for MLLMs. In this paper, we reveal that a combination of low- and high-resolution visual features can efficiently mitigate this shortcoming. Based on this principle, we propose a novel and efficient method for MLLMs, termed Mixture-of-Resolution Adaptation (MRA). In particular, MRA adopts two visual pathways for images of different resolutions, where high-resolution visual information is embedded into the low-resolution pathway via the novel mixture-of-resolution adapters (MR-Adapters). This design also greatly reduces the input sequence length of MLLMs. To validate MRA, we apply it to a recent MLLM called LLaVA, and term the new model LLaVA-HR. We conduct extensive experiments on 17 vision-language (VL) tasks, which show that LLaVA-HR outperforms existing MLLMs on 15 VL tasks, e.g., +5.2\% on TextVQA. More importantly, both training and inference of LLaVA-HR remain efficient with MRA, e.g., 20 training hours and faster inference speed than LLaVA-NeXT. Source codes are released at: https://github.com/luogen1996/LLaVA-HR.

The swift progress of Multi-modal Large Models (MLLMs) has showcased their impressive ability to tackle tasks blending vision and language. Yet, most current models and benchmarks cater to scenarios with a narrow scope of visual and textual contexts. These models often fall short when faced with complex comprehension tasks, which involve navigating through a plethora of irrelevant and potentially misleading information in both text and image forms. To bridge this gap, we introduce a new, more demanding task known as Interleaved Image-Text Comprehension (IITC). This task challenges models to discern and disregard superfluous elements in both images and text to accurately answer questions and to follow intricate instructions to pinpoint the relevant image. In support of this task, we further craft a new VEGA dataset, tailored for the IITC task on scientific content, and devised a subtask, Image-Text Association (ITA), to refine image-text correlation skills. Our evaluation of four leading closed-source models, as well as various open-source models using VEGA, underscores the rigorous nature of IITC. Even the most advanced models, such as Gemini-1.5-pro and GPT4V, only achieved modest success. By employing a multi-task, multi-scale post-training strategy, we have set a robust baseline for MLLMs on the IITC task, attaining an $85.8\%$ accuracy rate in image association and a $0.508$ Rouge score. These results validate the effectiveness of our dataset in improving MLLMs capabilities for nuanced image-text comprehension.

Recent advances in test-time adaptation (TTA) for Vision-Language Models (VLMs) have garnered increasing attention, particularly through the use of multiple augmented views of a single image to boost zero-shot generalization. Unfortunately, existing methods fail to strike a satisfactory balance between performance and efficiency, either due to excessive overhead of tuning text prompts or unstable benefits from handcrafted, training-free visual feature enhancement. In this paper, we present Global-Spatial Bias Learner (GS-Bias), an efficient and effective TTA paradigm that incorporates two learnable biases during TTA, unfolded as the global bias and spatial bias. Particularly, the global bias captures the global semantic features of a test image by learning consistency across augmented views, while spatial bias learns the semantic coherence between regions in the image’s spatial visual representation. It is worth highlighting that these two sets of biases are directly added to the logits outputed by the pretrained VLMs, which circumvent the full backpropagation through VLM that hinders the efficiency of existing TTA methods. This endows GS-Bias with extremely high efficiency while achieving state-of-the-art performance on 15 benchmark datasets. For example, it achieves a 2.23% improvement over TPT in cross-dataset generalization and a 2.72% improvement in domain generalization, while requiring only 6.5% of TPT’s memory usage on ImageNet.

In this paper, we address the challenge of determining the layer-wise sparsity rates of large language models (LLMs) through a theoretical perspective. Specifically, we identify a critical issue of **"reconstruction error explosion"** in existing LLMs sparsification methods. This refers to the cumulative effect of reconstruction errors throughout the sparsification process, where errors from earlier layers propagate and amplify in subsequent layers. As a result, the overall reconstruction error increases significantly, leading to a substantial degradation in model performance. Through theoretical analysis, we derive a simple yet effective approach to layer-wise sparsity allocation that mitigates this issue. Our method uses a monotonically increasing arithmetic progression, reducing the process of determining sparsity rates for multiple layers to the determination of a single common difference hyperparameter. Remarkably, this allows for the optimal layer-wise sparsity rates to be identified with just a few trials. Both our theoretical analysis and experimental results demonstrate that this sparsity allocation scheme is near optimal. Extensive experiments show that our method significantly improves the performance of sparse LLMs across various architectures, outperforming existing layer-wise sparsity methods. Furthermore, it enhances the performance of various compression techniques and is applicable to vision and multimodal models. Notably, our method achieves a reduction of 52.10 in perplexity for the 70% sparse LLaMA2-7B model obtained via Wanda, improves average zero-shot accuracy by 10.50%, and delivers speedups of 2.63$\times$ and 2.23$\times$ on CPU and GPU, respectively. Code is available at https://github.com/wzhuang-xmu/ATP.

In this paper, we aim to establish a simple, effective, and theoretically grounded benchmark for rigorously probing abstract reasoning in Large Language Models (LLMs). To achieve this, we first develop a mathematic framework that defines abstract reasoning as the ability to: (i) extract essential patterns independent of surface representations, and (ii) apply consistent rules to these abstract patterns. Based on this framework, we introduce two novel complementary metrics: Γ measures basic reasoning accuracy, while ∆ quantifies a model's reliance on specific symbols rather than underlying patterns - a key indicator of true abstraction versus mere memorization. To implement this measurement, we design a benchmark: systematic symbol remapping in rule-based tasks, which forces models to demonstrate genuine pattern recognition beyond superficial token matching. Extensive LLM evaluations using this benchmark (commercial API models, 7B-70B, multi-agent) reveal:1) critical limitations in non-decimal arithmetic and symbolic reasoning; 2) persistent abstraction gaps despite chain-of-thought prompting; and 3) ∆'s effectiveness in robustly measuring memory dependence by quantifying performance degradation under symbol remapping, particularly highlighting operand-specific memorization. These findings underscore that current LLMs, despite domain-specific strengths, still lack robust abstract reasoning, highlighting key areas for future improvement.

Vision-Language Models (VLMs) face two critical limitations in visual representation learning: degraded supervision due to information loss during gradient propagation, and the inherent semantic sparsity of textual supervision compared to visual data. We propose the Diffusion Supervision Vision-Language Model (DS-VLM), a plug-and-play framework that introduces diffusion-based direct supervision for vision-language alignment. By reconstructing input images through a diffusion model conditioned on outputs of the visual encoder and the connector, our method establishes a short-path gradient propagation channel from pixel space to visual features. This approach simultaneously preserves high-level semantic alignment through conventional text supervision while enhancing visual feature quality via pixel-level reconstruction constraints. Extensive experiments conducted across various visual encoders and LLMs of different scales demonstrate the effectiveness of our approach.

Multimodal person re-identification (Re-ID) aims to match pedestrian images across different modalities. However, most existing methods focus on limited cross-modal settings and fail to support arbitrary query-retrieval combinations, hindering practical deployment. We propose FlexiReID, a flexible framework that supports seven retrieval modes across four modalities: RGB, infrared, sketches, and text. FlexiReID introduces an adaptive mixture-of-experts (MoE) mechanism to dynamically integrate diverse modality features and a cross-modal query fusion module to enhance multimodal feature extraction. To facilitate comprehensive evaluation, we construct CIRS-PEDES, a unified dataset extending four popular Re-ID datasets to include all four modalities. Extensive experiments demonstrate that FlexiReID achieves state-of-the-art performance and offers strong generalization in complex scenarios.

Inference latency stands as a critical bottleneck in the large-scale deployment of Large Language Models (LLMs). Speculative decoding methods have recently shown promise in accelerating inference without compromising the output distribution. However, existing work typically relies on a dualistic draft-verify framework and lacks rigorous theoretical grounding. In this paper, we introduce a novel \emph{polybasic} speculative decoding framework, underpinned by a comprehensive theoretical analysis. Specifically, we prove a fundamental theorem that characterizes the optimal inference time for multi-model speculative decoding systems, shedding light on how to extend beyond the dualistic approach to a more general polybasic paradigm. Through our theoretical investigation of multi-model token generation, we expose and optimize the interplay between model capabilities, acceptance lengths, and overall computational cost. Our framework supports both standalone implementation and integration with existing speculative techniques, leading to accelerated performance in practice. Experimental results across multiple model families demonstrate that our approach yields speedup ratios ranging from $3.31\times$ to $4.01\times$ for LLaMA2-Chat 7B, up to $3.87 \times$ for LLaMA3-8B, up to $4.43 \times$ for Vicuna-7B and up to $3.85 \times$ for Qwen2-7B---all while preserving the original output distribution. We release our theoretical proofs and implementation code to facilitate further investigation into polybasic speculative decoding.

N:M sparsity stands as a progressively important tool for DNN compression, achieving practical speedups by stipulating at most N non-zero components within M sequential weights. Unfortunately, most existing works identify the N:M sparse mask through dense backward propagation to update all weights, which incurs exorbitant training costs. In this paper, we introduce BAME, a method that maintains consistent sparsity throughout the N:M sparse training process. BAME perpetually keeps both sparse forward and backward propagation, while iteratively performing weight pruning-and-regrowing within designated weight blocks to tailor the N:M mask. These blocks are selected through a joint assessment based on accumulated mask oscillation frequency and expected loss reduction of mask adaptation, thereby ensuring stable and efficient identification of the optimal N:M mask. Our empirical results substantiate the effectiveness of BAME, illustrating it performs comparably to or better than previous works that fully maintaining dense backward propagation during training. For instance, BAME attains a 72.0% top-1 accuracy while training a 1:16 sparse ResNet-50 on ImageNet, eclipsing SR-STE by 0.5%, despite achieving 2.37 training FLOPs reduction. Code is released at https://github.com/BAME-xmu/BAME

This paper introduces EasyInv, an easy yet novel approach that significantly advances the field of DDIM Inversion by addressing the inherent inefficiencies and performance limitations of traditional iterative optimization methods. At the core of our EasyInv is a refined strategy for approximating inversion noise, which is pivotal for enhancing the accuracy and reliability of the inversion process. By prioritizing the initial latent state, which encapsulates rich information about the original images, EasyInv steers clear of the iterative refinement of noise items. Instead, we introduce a methodical aggregation of the latent state from the preceding time step with the current state, effectively increasing the influence of the initial latent state and mitigating the impact of noise. We illustrate that EasyInv is capable of delivering results that are either on par with or exceed those of the conventional DDIM Inversion approach, especially under conditions where the model’s precision is limited or computational resources are scarce. Concurrently, our EasyInv offers an approximate threefold enhancement regarding inference efficiency over off-the-shelf iterative optimization techniques. It can be easily combined with most existing inversion methods by only four lines of code. See code at https://github.com/potato-kitty/EasyInv.

Micromanipulation using magnetic microswarms has garnered significant attention in recent years due to their potential in microscale cargo delivery tasks. While existing studies have demonstrated the capabilities of microswarms in basic manipulation tasks, they often lack the autonomy required to handle more complex specifications, particularly temporal logic tasks. In this paper, we propose a novel formal planning strategy for magnetic microswarms that enables cargo delivery in complex environments while satisfying finite linear temporal logic (LTLf) specifications. Our approach consists of two key components. First, we develop a high-level path planner based on a bidirectional temporal logic rapid-explore random tree star (BTL-RRT*) algorithm, which facilitates efficient planning while ensuring compliance with the given task specifications. Second, we employ an automaton to manage the manipulation modes of the microswarm, enabling real-time control over the capture and release of cargoes. In addition, we implement the planning strategy on microswarms actuated by a visual feedback magnetic tweezers system. Extensive simulations and experimental results demonstrate the effectiveness of the proposed planning strategy. The results indicate that, using the proposed approach, microswarms can autonomously select and deliver multiple microbeads to designated regions in both static and dynamic environments, adhering to the LTLfspecifications.

In this study, we introduce a novel method called group-wise VI sual token Selection and Aggregation (VISA) to address the issue of inefficient inference stemming from excessive visual tokens in multimoal large language models (MLLMs). Compared with previous token pruning approaches, our method can preserve more visual information while compressing visual tokens. We first propose a graph-based visual token aggregation (VTA) module. VTA treats each visual token as a node, forming a graph based on semantic similarity among visual tokens. It then aggregates information from removed tokens into kept tokens based on this graph, producing a more compact visual token representation. Additionally, we introduce a group-wise token selection strategy (GTS) to divide visual tokens into kept and removed ones, guided by text tokens from the final layers of each group. This strategy progressively aggregates visual information, enhancing the stability of the visual information extraction process. We conduct comprehensive experiments on LLaVA-1.5, LLaVA-NeXT, and Video-LLaVA across various benchmarks to validate the efficacy of VISA. Our method consistently outperforms previous methods, achieving a superior trade-off between model performance and inference speed.

Despite growing interest in hallucination in Multimodal Large Language Models, existing studies primarily focus on single-image settings, leaving hallucination in multi-image scenarios largely unexplored. To address this gap, we conduct the first systematic study of hallucinations in multi-image MLLMs and propose MIHBench, a benchmark specifically tailored for evaluating object-related hallucinations across multiple images. MIHBench comprises three core tasks: Multi-Image Object Existence Hallucination, Multi-Image Object Count Hallucination, and Object Identity Consistency Hallucination, targeting semantic understanding across object existence, quantity reasoning, and cross-view identity consistency. Through extensive evaluation, we identify key factors associated with the occurrence of multi-image hallucinations, including: a progressive relationship between the number of image inputs and the likelihood of hallucination occurrences; a strong correlation between single-image hallucination tendencies and those observed in multi-image contexts; and the influence of same-object image ratios and the positional placement of negative samples within image sequences on the occurrence of object identity consistency hallucination. To address these challenges, we propose a Dynamic Attention Balancing mechanism that adjusts inter-image attention distributions while preserving the overall visual attention proportion. Experiments across multiple state-of-the-art MLLMs demonstrate that our method effectively reduces hallucination occurrences and enhances semantic integration and reasoning stability in multi-image scenarios.

Open vocabulary image segmentation tackles the challenge of recognizing dynamically adjustable, predefined novel categories at inference time by leveraging vision-language alignment. However, existing paradigms typically perform class-agnostic region segmentation followed by category matching, which deviates from the human visual system's process of recognizing objects based on semantic concepts, leading to poor alignment between region segmentation and object concepts. To bridge this gap, we propose a novel Cognition-Inspired Framework for open vocabulary image segmentation that emulates the human visual recognition process: first forming a conceptual understanding of an object, then perceiving its spatial extent. The framework consists of three core components: (1) A Generative Vision-Language Model (G-VLM) that mimics human cognition by generating object concepts to provide semantic guidance for region segmentation. (2) A Concept-Aware Visual Enhancer module that fuses textual concept features with global visual representations, enabling adaptive visual perception based on object concepts. (3) A Cognition-Inspired Mask Decoder that integrates local instance features with G-VLM-provided semantic cues, allowing selective classification over a subset of relevant categories. Extensive experiments demonstrate that our framework achieves significant improvements, reaching 27.2 PQ, 17.0 mAP, and 35.3 mIoU on A-150. It further attains 56.2, 28.2, 15.4, 59.2, 18.7, and 95.8 mIoU on Cityscapes, Mapillary Vistas, A-847, PC-59, PC-459, and PAS-20, respectively. In addition, our framework supports vocabulary-free image segmentation, offering enhanced flexibility in recognizing unseen categories.

The reasoning segmentation task involves segmenting objects within an image by interpreting implicit user instructions. Despite significant advancements made by existing approaches, they remain constrained by low perceptual resolution, as visual encoders are typically pre-trained at lower resolutions. Furthermore, simply interpolating the positional embeddings of visual encoders to enhance perceptual resolution yields only marginal performance improvements while incurring substantial computational costs. To address this, we propose HRSeg, an efficient model with high-resolution fine-grained perception. It features two key innovations: High-Resolution Perception (HRP) and High-Resolution Enhancement (HRE). The HRP module processes high-resolution images through cropping, integrating local and global features for multi-granularity quality. The HRE module enhances mask features by integrating fine-grained information from high-resolution images, refining their alignment with text features for precise segmentation. Extensive ablation studies validate the effectiveness of our modules, while comprehensive experiments on multiple benchmark datasets demonstrate HRSeg's superior performance. Code will be available at https://github.com/WeihuangLin/HRSeg.

Anomaly inspection plays a vital role in industrial manufacturing, but the scarcity of anomaly samples significantly limits the effectiveness of existing methods in tasks such as localization and classification. While several anomaly synthesis approaches have been introduced for data augmentation, they often struggle with low realism, inaccurate mask alignment, and poor generalization. To overcome these limitations, we propose Generate Aligned Anomaly (GAA), a region-guided, few-shot anomaly image-mask pair generation framework. GAA leverages the strong priors of a pretrained latent diffusion model to generate realistic, diverse, and semantically aligned anomalies using only a small number of samples. The framework first employs Localized Concept Decomposition to jointly model the semantic features and spatial information of anomalies, enabling flexible control over the type and location of anomalies. It then utilizes Adaptive Multi-Round Anomaly Clustering to perform fine-grained semantic clustering of anomaly concepts, thereby enhancing the consistency of anomaly representations. Subsequently, a region-guided mask generation strategy ensures precise alignment between anomalies and their corresponding masks, while a low-quality sample filtering module is introduced to further improve the overall quality of the generated samples. Extensive experiments on the MVTec AD and LOCO datasets demonstrate that GAA achieves superior performance in both anomaly synthesis quality and downstream tasks such as localization and classification.

Text-based Person Retrieval (TBPR) is a challenging task that aims to retrieve pedestrian images according to natural language descriptions. Existing works mainly focus on discriminative feature learning via exploring cross-modal matching methods, while the overfitting issues caused by insufficient labeled data and the absence of well-designed auxiliary tasks are often overlooked. Motivated by the recent progress of large language models (LLMs), we propose a novel method named GPT-ReID for TBPR, which aims to leverage the strong comprehension of LLMs to alleviate the overfitting risk. Specifically, based on the great power of GPT, GPT-ReID first introduces an adversarial text generation scheme called GPTGAN, which aims to generate comprehensive strong positive captions and deceptive hard negative captions through the original captions for a single image. Furthermore, a joint auxiliary learning strategy is also proposed which contains Multi-Relation Aware (MRA), Keywords Masked Language Model (KMLM), and Keywords Replacement Detection (KRD), to facilitate global- and token-level optimization, enhancing cross-modal granular representation alignment. Extensive experiments on a set of highly competitive benchmark datasets validate the merits of the proposed GPT-ReID against a flurry of state-of-the-art methods, with Rank-1 accuracy reaching 78.42%, 69.43%, and 70.06% on CUHK-PEDS, ICFG-PEDS, and RSTPReid, respectively.

Yihan Zhang, Jie Fu, Rongrong Ji, Jie Chen. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Multimodal Large Language Model (MLLM) relies on the powerful LLM to perform multimodal tasks, showing amazing emergent abilities in recent studies, such as writing poems based on an image. However, it is difficult for these case studies to fully reflect the performance of MLLM, lacking a comprehensive evaluation. In this paper, we fill in this blank, presenting the first comprehensive MLLM Evaluation benchmark MME. It measures both perception and cognition abilities on a total of 14 subtasks. In order to avoid data leakage that may arise from direct use of public datasets for evaluation, the annotations of instruction-answer pairs are all manually designed. The concise instruction design allows us to fairly compare MLLMs, instead of struggling in prompt engineering. Besides, with such an instruction, we can also easily carry out quantitative statistics. A total of 30 advanced MLLMs are comprehensively evaluated on our MME, which not only suggests that existing MLLMs still have a large room for improvement, but also reveals the potential directions for the subsequent model optimization. The data are released at the project page: https://github.com/BradyFU/Awesome-Multimodal-Large-Language-Models/tree/Evaluation.

Recent Multimodal Large Language Models (MLLMs) have typically focused on integrating visual and textual modalities, with less emphasis placed on the role of speech in enhancing interaction. However, speech plays a crucial role in multimodal dialogue systems, and implementing high-performance in both vision and speech tasks remains a challenge due to the fundamental modality differences. In this paper, we propose a carefully designed multi-stage training methodology that progressively trains LLM to understand both visual and speech information, ultimately enabling fluent vision and speech interaction. Our approach not only preserves strong vision-language capacity, but also enables efficient speech-to-speech dialogue capabilities without separate ASR and TTS modules, significantly accelerating multimodal end-to-end response speed. By comparing against state-of-the-art counterparts across benchmarks for image, video, and speech, we demonstrate that our omni model is equipped with both strong visual and speech capabilities, making omni understanding and interaction.

Mixture-of-Experts (MoE) architectures enable efficient scaling of large language models but face prohibitive memory demands due to massive parameterization. Existing pruning methods rely on heuristic metrics or impractical enumeration of expert subsets, leading to suboptimal performance or scalability. In this paper, we propose Shapley-MoE, an efficient pruning method for MoE models inspired by cooperative game theory. By quantifying each expert’s contribution via Shapley value, our method identifies important experts without exhaustive combination evaluations. To overcome the NP-hard complexity of exact Shapley computation, we introduce a Monte Carlo sampling strategy for efficient approximation that reduces complexity to quadratic time. However, vanilla Monte Carlo sampling still faces issues of insufficient estimation accuracy and low sampling efficiency. To address these issues, we further propose two novel methods to improve sampling accuracy and efficiency: (1) Early Truncation, which early terminates unstable sampling steps caused by overly small expert subsets, and (2) Router-Guided Importance Sampling, which prioritize sampling important expert subsets using gating activation probabilities. Both theoretical and experimental analyses show that both methods can accelerate Shapley value estimation and improve accuracy. Extensive empirical evaluations demonstrate that our pruned MoE models outperform existing expert pruning methods. Notably, when applied to the Qwen2-57B-A14B model, our method reduces the number of experts by 25% with only a 0.92 increase in perplexity and over 96.4% of the average zero-shot accuracy is maintained.

State space models (SSMs) have recently garnered significant attention in computer vision. However, due to the unique characteristics of image data, adapting SSMs from natural language processing to computer vision has not outperformed the state-of-the-art convolutional neural networks (CNNs) and Vision Transformers (ViTs). Existing vision SSMs primarily leverage manually designed scans to flatten image patches into sequences locally or globally. This approach disrupts the original semantic spatial adjacency of the image and lacks flexibility, making it difficult to capture complex image structures. To address this limitation, we propose Dynamic Adaptive Scan (DAS), a data-driven method that adaptively allocates scanning orders and regions. This enables more flexible modeling capabilities while maintaining linear computational complexity and global modeling capacity. Based on DAS, we further propose the vision backbone DAMamba, which significantly outperforms popular vision Mamba models in vision tasks such as image classification, object detection, instance segmentation, and semantic segmentation. Notably, it surpasses some of the latest state-of-the-art CNNs and ViTs.

Multi-modal Large Language Models (MLLMs) excel at single-image tasks but struggle with multi-image understanding due to cross-modal misalignment, leading to hallucinations (context omission, conflation, and misinterpretation). Existing methods using Direct Preference Optimization (DPO) constrain optimization to a solitary image reference within the input sequence, neglecting holistic context modeling. To address this, we propose Context-to-Cue Direct Preference Optimization (CcDPO), a multi-level preference optimization framework that enhances per-image perception in multi-image settings by zooming into visual clues—from sequential context to local details. Our approach features two sequentially dependent components: (i) Context-Level Optimization: By introducing low-cost sequence preference pairs, we optimize the model to distinguish between complete and disrupted multi-image contexts, thereby correcting cognitive biases in MLLMs’ multi-image understanding. (ii) Needle-Level Optimization: By integrating region-specific visual prompts with multimodal preference supervision, we direct the model’s attention to critical visual details, effectively suppressing perceptual biases toward fine-grained visual information. To support scalable optimization, we also construct MultiScope-42k, an automatically generated multi-image dataset with hierarchical preference pairs. Experiments show that CcDPO significantly reduces hallucinations and yields consistent performance gains across general single- and multi-image tasks. Codes are available at https://github.com/LXDxmu/CcDPO.

Reducing the key-value (KV) cache burden in Large Language Models (LLMs) significantly accelerates inference. Dynamically selecting critical KV caches during decoding helps maintain performance. Existing methods use random linear hashing to identify important tokens, but this approach is inefficient due to the orthogonal distribution of queries and keys within two narrow cones in LLMs. We introduce Spotlight Attention, a novel method that employs non-linear hashing functions to optimize the embedding distribution of queries and keys, enhancing coding efficiency and robustness. We also developed a lightweight, stable training framework using a Bradley-Terry ranking-based loss, enabling optimization of the non-linear hashing module on GPUs with 16GB memory in 8 hours. Experimental results show that Spotlight Attention drastically improves retrieval precision while shortening the length of the hash code at least 5$\times$ compared to traditional linear hashing. Finally, we exploit the computational advantages of bitwise operations by implementing specialized CUDA kernels, achieving hashing retrieval for 512K tokens in under 100$\mu$s on a single A100 GPU, with end-to-end throughput up to 3$\times$ higher than vanilla decoding.

This paper introduces Completion Pruning Policy Optimization (CPPO) to accelerate the training of reasoning models based on Group Relative Policy Optimization (GRPO). GRPO, while effective, incurs high training costs due to the need to sample multiple completions for each question. Our experiment and theoretical analysis reveal that the number of completions impacts model accuracy yet increases training time multiplicatively, and not all completions contribute equally to policy training---their contribution depends on their relative advantage. To address these issues, we propose CPPO, which prunes completions with low absolute advantages, significantly reducing the number needed for gradient calculation and updates. Additionally, we introduce a dynamic completion allocation strategy to maximize GPU utilization by incorporating additional questions, further enhancing training efficiency. Experiments show that CPPO achieves up to $7.98\times$ speedup on GSM8K and $3.48\times$ on Math while preserving or even enhancing the accuracy compared to the original GRPO. We release our code at https://github.com/lzhxmu/CPPO.

Current evaluation paradigms for large language models (LLMs) represent a critical blind spot in AI research—relying on opaque numerical metrics that conceal fundamental limitations in spatial reasoning while providing no intuitive understanding of model capabilities. This deficiency creates a dangerous disconnect between reported performance and practical abilities, particularly for applications requiring physical world understanding. We introduce LTD-Bench, a breakthrough benchmark that transforms LLM evaluation from abstract scores to directly observable visual outputs by requiring models to generate drawings through dot matrices or executable code. This approach makes spatial reasoning limitations immediately apparent even to non-experts, bridging the fundamental gap between statistical performance and intuitive assessment. LTD-Bench implements a comprehensive methodology with complementary generation tasks (testing spatial imagination) and recognition tasks (assessing spatial perception) across three progressively challenging difficulty levels, methodically evaluating both directions of the critical language-spatial mapping. Our extensive experiments with state-of-the-art models expose an alarming capability gap: even LLMs achieving impressive results on traditional benchmarks demonstrate profound deficiencies in establishing bidirectional mappings between language and spatial concepts—a fundamental limitation that undermines their potential as genuine world models. Furthermore, LTD-Bench's visual outputs enable powerful diagnostic analysis, offering a potential approach to investigate model similarity. Our dataset and codes are available at https://github.com/walktaster/LTD-Bench.

With the growing requirement for natural human-computer interaction, speech-based systems receive increasing attention as speech is one of the most common forms of daily communication. However, the existing speech models still experience high latency when generating the first audio token during streaming, which poses a significant bottleneck for deployment. To address this issue, we propose VITA-Audio, an end-to-end large speech model with fast audio-text token generation. Specifically, we introduce a lightweight Multiple Cross-modal Token Prediction (MCTP) module that efficiently generates multiple audio tokens within a single model forward pass, which not only accelerates the inference but also significantly reduces the latency for generating the first audio in streaming scenarios. In addition, a four-stage progressive training strategy is explored to achieve model acceleration with minimal loss of speech quality. To our knowledge, VITA-Audio is the first multi-modal large language model capable of generating audio output during the first forward pass, enabling real-time conversational capabilities with minimal latency. VITA-Audio is fully reproducible and is trained on open-source data only. Experimental results demonstrate that our model achieves an inference speedup of 3~5x at the 7B parameter scale, but also significantly outperforms open-source models of similar model size on multiple benchmarks for automatic speech recognition (ASR), text-to-speech (TTS), and spoken question answering (SQA) tasks.

Existing large video-language models (LVLMs) struggle to comprehend long videos correctly due to limited context. To address this problem, fine-tuning long-context LVLMs and employing GPT-based agents have emerged as promising solutions. However, fine-tuning LVLMs would require extensive high-quality data and substantial GPU resources, while GPT-based agents would rely on proprietary models (e.g., GPT-4o). In this paper, we propose Video Retrieval-Augmented Generation (Video-RAG), a training-free and cost-effective pipeline that employs visually-aligned auxiliary texts to help facilitate cross-modality alignment while providing additional information beyond the visual content. Specifically, we leverage open-source external tools to extract visually-aligned information from pure video data (e.g., audio, optical character, and object detection), and incorporate the extracted information into an existing LVLM as auxiliary texts, alongside video frames and queries, in a plug-and-play manner. Our Video-RAG offers several key advantages: (i) lightweight with low computing overhead due to single-turn retrieval; (ii) easy implementation and compatibility with any LVLM; and (iii) significant, consistent performance gains across long video understanding benchmarks, including Video-MME, MLVU, and LongVideoBench. Notably, our model demonstrates superior performance over proprietary models like Gemini-1.5-Pro and GPT-4o when utilized with a 72B model.

In this paper, we study the visual redundancy problem of multimodal large language models (MLLMs) from the perspective of attention behaviors. Via extensive empirical experiments, we observe and conclude three main inference stages of MLLMs: (i) Early fusion between tokens is first accomplished quickly. (ii) Intra-modality modeling then comes to play. (iii) Multimodal reasoning resumes and lasts until the end of inference. In particular, we reveal that visual tokens will stop contributing to reasoning when the text tokens receive enough image information. Based on this observation, we propose an effective method to improve the efficiency of MLLMs, termed dynamic visual-token exit (DyVTE), which is orthogonal but collaborative to previous token-wise visual compression methods. To validate the efficacy of DyVTE, we apply it to a set of MLLMs, including LLaVA, VILA, EAGLE and InternVL. The experimental results not only show the effectiveness of our DyVTE in improving MLLMs' efficiency, e.g., DyVTE reduces the computation overhead of LLaVA-1.5 by up to 45.7% without performance drop, but also reveal a general pattern across multiple MLLMs, well facilitating the in-depth analysis of MLLMs. Our code is anonymously released at https://anonymous.4open.science/r/AnonymousDyVTE-26AB/.

Weakly supervised object localization (WSOL) aims to localize objects using only image-level labels as supervision. Despite recent advancements incorporating transformers into WSOL have resulted in improvements, these methods often rely on category-agnostic attention maps, leading to suboptimal object localization. This paper presents a novel CLIP-Driven TRansformer (CDTR) that learns category-aware representations for accurate object localization. Specifically, we initially propose a Category-aware Stimulation Module (CSM) that embeds learnable category biases into self-attention maps, enhancing the learning process with auxiliary supervision. Additionally, an Object Constraint Module (OCM) is designed to refine object regions in a self-supervised manner, leveraging the discriminative potential of the self-attention maps provided by CSM. To create a synergistic connection between CSM and OCM, we further develop a Semantic Kernel Integrator (SKI), which generates a semantic kernel for self-attention maps. Meanwhile, we explore the CLIP model and design a Semantic Boost Adapter (SBA) to enrich object representations by integrating semantic-specific image and text representations into self-attention maps. Extensive experimental evaluations on benchmark datasets, such as CUB-200-2011 and ILSVRC highlight the superior performance of our CDTR framework.

We introduce Hyper-YOLO, a new object detection method that integrates hypergraph computations to capture the complex high-order correlations among visual features. Traditional YOLO models, while powerful, have limitations in their neck designs that restrict the integration of cross-level features and the exploitation of high-order feature interrelationships. To address these challenges, we propose the Hypergraph Computation Empowered Semantic Collecting and Scattering (HGC-SCS) framework, which transposes visual feature maps into a semantic space and constructs a hypergraph for high-order message propagation. This enables the model to acquire both semantic and structural information, advancing beyond conventional feature-focused learning. Hyper-YOLO incorporates the proposed Mixed Aggregation Network (MANet) in its backbone for enhanced feature extraction and introduces the Hypergraph-Based Cross-Level and Cross-Position Representation Network (HyperC2Net) in its neck. HyperC2Net operates across five scales and breaks free from traditional grid structures, allowing for sophisticated high-order interactions across levels and positions. This synergy of components positions Hyper-YOLO as a state-of-the-art architecture in various scale models, as evidenced by its superior performance on the COCO dataset. Specifically, Hyper-YOLO-N significantly outperforms the advanced YOLOv8-N and YOLOv9-T with 12% and 9% improvements.

In this paper, we present UISE, a unified image segmentation framework that achieves efficient performance across various segmentation tasks, eliminating the need for multiple specialized pipelines. UISE employs dynamic convolutions between universal segmentation kernels and image feature maps, enabling a single pipeline for different tasks such as panoptic, instance, semantic, and video instance segmentation. To address computational requirements, we introduce a feature pyramid aggregator for image feature extraction and a separable dynamic decoder for generating segmentation kernels. The aggregator re-parameterizes interpolation-first modules in a convolution-first manner, resulting in a significant acceleration of the pipeline without incurring additional costs. The decoder incorporates multi-head cross-attention through separable dynamic convolution, enhancing both efficiency and accuracy. Extensive experiments are conducted to validate UISE's performance across different segmentation tasks. To the best of our knowledge, UISE is the first universal segmentation framework that delivers competitive performance in terms of both speed and accuracy when compared to current state-of-the-art models.

The rising importance of 3D representation learning, pivotal in computer vision, autonomous driving, and robotics, is evident. However, a prevailing trend, which straightforwardly resorted to transferring 2D alignment strategies to the 3D domain, encounters three distinct challenges: (1) Information Degradation: This arises from the alignment of 3D data with mere single-view 2D images and generic texts, neglecting the need for multi-view images and detailed subcategory texts. (2) Insufficient Synergy: These strategies align 3D representations to image and text features individually, hampering the overall optimization for 3D models. (3) Underutilization: The fine-grained information inherent in the learned representations is often not fully exploited, indicating a potential loss in detail. To address these issues, we introduce JM3D, a comprehensive approach integrating point cloud, text, and image. Key contributions include the Structured Multimodal Organizer (SMO), enriching vision-language representation with multiple views and hierarchical text, and the Joint Multi-modal Alignment (JMA), combining language understanding with visual representation. Our advanced model, JM3D-LLM, marries 3D representation with large language models via efficient fine-tuning. Evaluations on ModelNet40 and ScanObjectNN establish JM3D's superiority. The superior performance of JM3D-LLM further underscores the effectiveness of our representation transfer approach.

Diffusion models suffer severe object repetition and local distortion when the inference resolution differs from its pre-trained resolution. We propose AccDiffusion v2, an accurate method for patch-wise higher-resolution diffusion extrapolation without training. Our in-depth analysis in this paper shows that using an identical text prompt for different patches leads to repetitive generation, while the absence of a prompt undermines image details. In response, our AccDiffusion v2 novelly decouples the vanilla image-content-aware prompt into a set of patch-content-aware prompts, each of which serves as a more precise description of a patch. Further analysis reveals that local distortion arises from inaccurate descriptions in prompts about the local structure of higher-resolution images. To address this issue, AccDiffusion v2, for the first time, introduces an auxiliary local structural information through ControlNet during higher-resolution diffusion extrapolation aiming to mitigate the local distortions. Finally, our analysis indicates that global semantic information is conducive to suppressing both repetitive generation and local distortion. Hence, our AccDiffusion v2 further proposes dilated sampling with window interaction for better global semantic information during higher-resolution diffusion extrapolation. We conduct extensive experiments, including both quantitative and qualitative comparisons, to demonstrate the efficacy of our AccDiffusion v2. The quantitative comparison shows that AccDiffusion v2 achieves state-of-the-art performance in image generation extrapolation without training. The qualitative comparison intuitively illustrates that AccDiffusion v2 effectively suppresses the issues of repetitive generation and local distortion in image generation extrapolation.

This paper focuses on semi-supervised crowd counting, where only a small portion of the training data are labeled. We formulate the pixel-wise density value to regress as a probability distribution, instead of a single deterministic value. On this basis, we propose a semi-supervised crowd counting model. First, we design a pixel-wise distribution matching loss to measure the differences in the pixel-wise density distributions between the prediction and the ground-truth; Second, we enhance the transformer decoder by using density tokens to specialize the forwards of decoders w.r.t. different density intervals; Third, we design the interleaving consistency self-supervised learning mechanism to learn from unlabeled data efficiently. Extensive experiments on four datasets are performed to show that our method clearly outperforms the competitors by a large margin under various labeled ratio settings.

Pre-training and fine-tuning have been the de-facto paradigm in vision-language domains. Along with the rapid growth of model sizes, fully fine-tuning these large-scale vision-language pre-training (VLP) models requires prohibitively expensive storage costs. To address this issue, recent advances in NLP offer a promising and efficient adaptation approach called LoRA, which aims to approximate the fine-tuning of large pre-trained model by updating low-rank parameters. Despite its effectiveness, we identify that LoRA suffers a large approximation error on VLP models and its optimization is also inefficient, which greatly limits its performance upper bound. In this paper, we mathematically prove that the approximation error of low-rank adaptation can be optimized by a new optimization objective, i.e., the weight distance between LoRA and fine-tuning. Based on this finding, we propose a novel PETL method for VLP models, namely momentum imitation learning (MoIL). Specifically, MoIL formulates PETL as a weight imitation learning process and directly optimize the approximation error bound of the low-rank adaptation. Based on this training scheme, we also explore a new hybrid approximation function to reduce the learning difficulty of low-rank adaptations. With these two novel designs, MoIL can greatly improve the optimization efficiency of the low-rank parameters on VLP models. We validate MoIL on three VLP models ranging from end-to-end network to two-stage network, and conduct extensive experiments on four VL tasks. Experimental results demonstrate superior performance and optimization efficiency of MoIL than existing PETL methods. For instance, by updating only 6.23% parameters, MoIL can even outperform full tuning by +2.3% on image-text matching task. Meanwhile, its inference efficiency and generalization ability is also validated by multiple VLP models, e.g., VLMO and VinVL.

Deep neural networks often suffer from poor generalization due to complex and non-convex loss landscapes. Sharpness-Aware Minimization (SAM) is a popular solution that smooths the loss landscape by minimizing the maximized change of training loss when adding a perturbation to the weight. However, indiscriminate perturbation of SAM on all parameters is suboptimal and results in excessive computation, double the overhead of common optimizers like Stochastic Gradient Descent (SGD). In this paper, we propose Sparse SAM (SSAM), an efficient and effective training scheme that achieves sparse perturbation by a binary mask. To obtain the sparse mask, we provide two solutions based on Fisher information and dynamic sparse training, respectively. We investigate the impact of different masks, including unstructured, structured, and $N$N:$M$M structured patterns, as well as explicit and implicit forms of implementing sparse perturbation. We theoretically prove that SSAM can converge at the same rate as SAM, i.e., $O(\log T/\sqrt{T})$O(logT/T) . Sparse SAM has the potential to accelerate training and smooth the loss landscape effectively. Extensive experimental results on CIFAR and ImageNet-1K confirm that our method is superior to SAM in terms of efficiency, and the performance is preserved or even improved with a perturbation of merely 50% sparsity.

Camouflaged Object Detection (COD) poses a significant challenge in computer vision, playing a critical role in applications. Existing COD methods often exhibit challenges in accurately predicting nuanced boundaries with high-confidence predictions. In this work, we introduce CamoDiffusion, a new learning method that employs a conditional diffusion model to generate masks that progressively refine the boundaries of camouflaged objects. In particular, we first design an adaptive transformer conditional network, specifically designed for integration into a Denoising Network, which facilitates iterative refinement of the saliency masks. Second, based on the classical diffusion model training, we investigate a variance noise schedule and a structure corruption strategy, which aim to enhance the accuracy of our denoising model by effectively handling uncertain input. Third, we introduce a Consensus Time Ensemble technique, which integrates intermediate predictions using a sampling mechanism, thus reducing overconfidence and incorrect predictions. Finally, we conduct extensive experiments on three benchmark datasets that show that: 1) the efficacy and universality of our method is demonstrated in both camouflaged and salient object detection tasks. 2) compared to existing state-of-the-art methods, CamoDiffusion demonstrates superior performance 3) CamoDiffusion offers flexible enhancements, such as an accelerated version based on the VQ-VAE model and a skip approach.

Panoptic Narrative Detection (PND) and Segmentation (PNS) are two challenging tasks that involve identifying and locating multiple targets in an image according to a long narrative description. In this paper, we propose a unified and effective framework called NICE that can jointly learn these two panoptic narrative recognition tasks. Existing visual grounding tasks use a two-branch paradigm, but applying this directly to PND and PNS can result in prediction conflict due to their intrinsic many-to-many alignment property. To address this, we introduce two cascading modules based on the barycenter of the mask, which are Coordinate Guided Aggregation (CGA) and Barycenter Driven Localization (BDL), responsible for segmentation and detection, respectively. By linking PNS and PND in series with the barycenter of segmentation as the anchor, our approach naturally aligns the two tasks and allows them to complement each other for improved performance. Specifically, CGA provides the barycenter as a reference for detection, reducing BDL's reliance on a large number of candidate boxes. BDL leverages its excellent properties to distinguish different instances, which improves the performance of CGA for segmentation. Extensive experiments demonstrate that NICE surpasses all existing methods by a large margin, achieving 4.1% for PND and 2.9% for PNS over the state-of-the-art. These results validate the effectiveness of our proposed collaborative learning strategy.

This paper presents a Spatial Re-parameterization (SpRe) method for the N:M sparsity. SpRe stems from an observation regarding the restricted variety in spatial sparsity of convolution kernels presented in N:M sparsity compared with unstructured sparsity. Particularly, N:M sparsity exhibits a fixed sparsity rate within the spatial domains due to its distinctive pattern that mandates N non-zero components among M successive weights in the input channel dimension of convolution filters. On the contrary, we observe that conventional unstructured sparsity displays a substantial divergence in sparsity across the spatial domains, which we experimentally verify to be very crucial for its robust performance retention compared with N:M sparsity. Therefore, SpRe employs the spatial-sparsity distribution of unstructured sparsity by assigning an extra branch in conjunction with the original N:M branch at training time, which allows the N:M sparse network to sustain a similar distribution of spatial sparsity with unstructured sparsity. During inference, the extra branch can be further re-parameterized into the main N:M branch, without exerting any distortion on the sparse pattern or additional computation costs. SpRe has achieved a commendable feat by matching the performance of N:M sparsity methods with state-of-the-art unstructured sparsity methods across various benchmarks. Our project is available at https://github.com/zyxxmu/SpRE.

We propose a universal video-level modality-awareness tracking model with online dense temporal token learning (called UM-ODTrack). It is designed to support various tracking tasks, including RGB, RGB+Thermal, RGB+Depth, and RGB+Event, utilizing the same model architecture and parameters. Specifically, our model is designed with three core goals: Video-level Sampling. We expand the model's inputs to a video sequence level, aiming to see a richer video context from an near-global perspective. Video-level Association. Furthermore, we introduce two simple yet effective online dense temporal token association mechanisms to propagate the appearance and motion trajectory information of target via a video stream manner. Modality Scalable. We propose two novel gated perceivers that adaptively learn cross-modal representations via a gated attention mechanism, and subsequently compress them into the same set of model parameters via a one-shot training manner for multi-task inference. This new solution brings the following benefits: (i) The purified token sequences can serve as temporal prompts for the inference in the next video frames, whereby previous information is leveraged to guide future inference. (ii) Unlike multi-modal trackers that require independent training, our one-shot training scheme not only alleviates the training burden, but also improves model representation. Extensive experiments on visible and multi-modal benchmarks show that our UM-ODTrack achieves a new SOTA performance.

Post-training quantization (PTQ) for vision transformers (ViTs) has received increasing attention from both academic and industrial communities due to its minimal data needs and high time efficiency. However, many current methods fail to account for the complex interactions between quantized weights and activations, resulting in significant quantization errors and suboptimal performance. This paper presents ERQ, an innovative two-step PTQ method specifically crafted to reduce quantization errors arising from activation and weight quantization sequentially. The first step, Activation quantization error reduction (Aqer), first applies Reparameterization Initialization aimed at mitigating initial quantization errors in high-variance activations. Then, it further mitigates the errors by formulating a Ridge Regression problem, which updates the weights maintained at full-precision using a closed-form solution. The second step, Weight quantization error reduction (Wqer), first applies Dual Uniform Quantization to handle weights with numerous outliers, which arise from adjustments made during Reparameterization Initialization, thereby reducing initial weight quantization errors. Then, it employs an iterative approach to further tackle the errors. In each iteration, it adopts Rounding Refinement that uses an empirically derived, efficient proxy to refine the rounding directions of quantized weights, complemented by a Ridge Regression solver to reduce the errors. Comprehensive experimental results demonstrate ERQ's superior performance across various ViTs variants and tasks. For example, ERQ surpasses the state-of-the-art GPTQ by a notable 36.81% in accuracy for W3A4 ViT-S.

Despite the great success, most models in image captioning (IC) are still stuck in the dilemma of generating simple and non-discriminative captions. In this paper, we study this problem from the perspective of data augmentation and propose a novel method called Multi-modal Mixup ( M 3 ixup). Compared with the original Mixup strategy designed for image classification , the proposed M 3 ixup has three novel designs to mix IC samples from the aspects of visual features, sentence embeddings and loss values, respectively. In practice, M 3 ixup can not only enrich the diversity of IC training data , but also enforce the model to focus more on visual information for captioning, thereby alleviating the negative effect of dataset bias and addressing the issue of simple captioning. To validate M 3 ixup, we apply it to three baseline models and conduct extensive experiments on MS COCO. The experimental results demonstrate that our proposed M 3 ixup can not only improve the discriminability and quality of generated captions, but also help the baseline models obtain obvious performance gains, i.e. , improving the CIDEr scores of the state-of-the-art model from 133.8 to 135.3 on off-line testing and 135.4 to 137.1 on online testing.

All-in-one image restoration methods, as opposed to conventional image restoration methods, reconstruct images impaired by various degradations within a unified model, eliminating the need for separate network parameters for each task. However, current all-in-one image restoration approaches tackle various types of image degradation using an identical underlying model, neglecting the inherent variability in complexity across different image restoration tasks, resulting in inefficient allocation of computational resources. To address this limitation, this article introduces the arbitrary routing framework (ARF), designed to effectively assess the difficulty of image restoration tasks and identify the most suitable network structure based on these complexities. This framework can be integrated with existing all-in-one image restoration models, enabling efficient inference by activating various proportions of the entire network, that is subnetworks, based on their task-specific complexities. More specifically, the ARF comprises two principal components: 1) the arbitrary routing backbone (ARB) and 2) a task-specific neural architecture search (T-NAS). The ARB incorporates a routing layer between consecutive convolutional groups, offering a wide array of potential subnetwork configurations while adding only negligible extra parameters. Concurrently, T-NAS autonomously identifies the most effective subnetworks for each image restoration task, optimizing both performance and efficiency through an efficiency-aware reward function. Comprehensive experiments across various image restoration tasks demonstrate that the ARF significantly improves performance metrics, that is, an increase of 0.31 in reconstruction PSNR, while also achieving a notable reduction in computational demands by 37.1% compared with the benchmark AirNet method. The code has been made available in the supplementary materials.

The focus of Face Anti-Spoofing (FAS) is shifting toward improving generalization performance in unseen scenarios. Traditional methods employing adversarial learning and meta-learning aim to extract or decouple generalizable features to address these challenges. However, enhancing performance solely through facial features remains challenging without additional informative inputs. To address this, Vision-Language Models (VLMs) with robust generalization capabilities have recently been introduced to FAS. Despite their potential, these VLMs typically adopt a late alignment strategy, relying only on encoder output features for modality alignment, which largely neglects mutual guidance between modalities. To bridge this gap, inspired by recent advancements in prompt learning, we employ learnable prompts and masking as intermediaries to enhance interaction between text and visual modalities, enabling the extraction of more generalizable features. Specifically, we propose ME-FAS, a Modality-Enhanced cross-domain FAS model integrating Prompt Fusion Transfer (PFT) and Text-guided Image Masking (TIM). PFT facilitates the integration of text features with visual information, improving domain adaptability in alignment with the textual context. Meanwhile, TIM leverages text features to mask image patches, directing visual features toward critical generalizable facial information, such as the eyes and mouth. Comprehensive evaluations across multiple benchmarks and various visualizations demonstrate significant performance gains, validating the effectiveness of our proposed approach. Our code and models are available at https://github.com/clpbc/ME-FAS.

In 3D point cloud-based object detection, attention mechanism in Group-Free [1] learns direct relationships between proposals and all seed points, providing each proposal with a global context in the form of a cross-attention map. However, our analysis and experimental comparison show that the attention mechanism assigns inappropriately large attention weights to certain seed points far from a proposal, which is not conducive to detecting objects correctly. In this work, we alleviate the above problem by proposing a mask method. For an initial proposal, our method first calculates a spatial distance-based mask, which measures the spatial relationship between all seed points and the proposal. Then, we fuse the mask into cross-attention layers in stacked attention modules and get a refined cross-attention map. In essence, our mask gives each proposal a local context; after it is fused with the global context given by the attention mechanism, the refined cross-attention map could suppress the negative impact of some distant seed points on a proposal. We present two alternative strategies to compute the mask, a hard mask, and a soft mask. Experimental results demonstrate that the soft mask brings better performance. In the soft mask, for each initial proposal's 3D-box shape, we use a parametric approximate ellipsoid as the basis of the mask's calculation, which has only two learnable parameters. Experimental results show our work could outperform Group-Free 0.7 [email protected] at the cost of increasing inference time by less than 1%. The performance of our algorithm on the public dataset SUN RGB-D is 63.7 [email protected] and 45.5 [email protected], which is the best performance among algorithms that preserve the irregular of seed points.

The diagnosis of colon polyps is important for the prevention of colorectal cancer. Polyp segmentation, however, is still a challenging problem given that recent medical computer-aided equipment suffers from situations of polyp variations in terms of size, color, texture, and poor illuminations in endoscopy videos. These obstacles hinder the prediction of polyp boundaries. Inspired by the observation that the values of pixels on the border region change more sharply than others, we propose the oriented-derivative (OD) representation to capture the relationship between pixels and the boundary region given distance and orientation. To adaptively use the proposed representation in arbitrary frameworks, we design plug-in modules to learn the representation and aggregate features to improve the accuracy of boundary predictions in the polyp segmentation task, which can be implemented in frameworks including the encoder-decoder and top-down architectures. Extensive experimental results show the improvement from the proposed oriented-derivative representation for the polyp segmentation task and the extendibility of our proposed modules in different architectures. Our methods achieved an improvement ranging from 0.3% to 2.5% (mDice) compared with the baseline on five publicly available datasets, includingKvasir, CVC-ClinicDB, EndoScene, CVC-ColonDB, andETIS.

Semi-supervised semantic segmentation pursues a holistic pixel-wise understanding of unseen images with limited annotation. To this end, existing methods focus on regularizing per-pixel prediction consistency within unlabeled data, while rarely modeling contextual relationships. But in fact, contextual semantics can provide valuable clues for scene understanding like inner-object continuity and spatial relationships' causality. Thus, in this paper, we propose a Dual-level Masked Semantics Inference (DMSI) that takes the initiative to explicitly learn contextual relationships via enforcing our model to infer the semantics of a pixel according to its surrounding contexts. This allows our model to exhaust accurate semantics by incorporating inter-pixel context clues, further leading to comprehensive segmentation. Specifically, DMSI comprises two main components. 1) Dual-level mask consistency regularization (DMCR) that learns the ability of semantics inference by aligning the predictions of masked views with the prediction of the complete view. The masked views here come from both the image level and feature level, where our model captures low-level attributes and high-level representations respectively. 2) AdaMask that provides a proper mask position and ratio for each image, guiding our model to focus on semantic-rich regions while providing balanced training between hard and easy samples. Through learning the ability of semantic inferring, DMSI remarkably enhances the interaction between pixels, further progressively intensifying the understanding of semantics. Extensive experiments under various settings on Cityscapes and Pascal VOC 2012 show that DMSI achieves new state-of-the-art performances. Furthermore, analysis indicates that our method has superiority in mining inter-pixel semantic relationships and improving robustness facing noise corruption.

Weakly supervised object localization (WSOL) stands as a pivotal endeavor within the realm of computer vision, entailing the location of objects utilizing merely image-level labels. Contemporary approaches in WSOL have leveraged FPMs, yielding commendable outcomes. However, these existing FPM-based techniques are predominantly confined to rudimentary strategies of either augmenting the foreground or diminishing the background presence. We argue for the exploration and exploitation of the intricate interplay between the object's foreground and its background to achieve efficient object localization. In this manuscript, we introduce an innovative framework, termed adaptive zone learning (AZL), which operates on a coarse-to-fine basis to refine FPMs through a triad of adaptive zone mechanisms. First, an adversarial learning mechanism (ALM) is employed, orchestrating an interplay between the foreground and background regions. This mechanism accentuates coarse-grained object regions in a mutually adversarial manner. Subsequently, an oriented learning mechanism (OLM) is unveiled, which harnesses local insights from both foreground and background in a fine-grained manner. This mechanism is instrumental in delineating object regions with greater granularity, thereby generating better FPMs. Furthermore, we propose a reinforced learning mechanism (RLM) as the compensatory mechanism for adversarial design, by which the undesirable foreground maps are refined again. Extensive experiments on CUB-200-2011 and ILSVRC datasets demonstrate that AZL achieves significant and consistent performance improvements over other state-of-the-art WSOL methods.

This article explores a novel dynamic network for vision and language (V&L) tasks, where the inferring structure is customized on the fly for different inputs. Most previous state-of-the-art (SOTA) approaches are static and handcrafted networks, which not only heavily rely on expert knowledge but also ignore the semantic diversity of input samples, therefore resulting in suboptimal performance. To address these issues, we propose a novel Dynamic Transformer Network (DTNet) for image captioning, which dynamically assigns customized paths to different samples, leading to discriminative yet accurate captions. Specifically, to build a rich routing space and improve routing efficiency, we introduce five types of basic cells and group them into two separate routing spaces according to their operating domains, i.e., spatial and channel. Then, we design a Spatial-Channel Joint Router (SCJR), which endows the model with the capability of path customization based on both spatial and channel information of the input sample. To validate the effectiveness of our proposed DTNet, we conduct extensive experiments on the MS-COCO dataset and achieve new SOTA performance on both the Karpathy split and the online test server. The source code is publicly available at https://github.com/xmu-xiaoma666/DTNet.

Unsupervised domain adaptation (UDA) person reidentification (Re-ID) aims to identify pedestrian images within an unlabeled target domain with an auxiliary labeled source-domain dataset. Many existing works attempt to recover reliable identity information by considering multiple homogeneous networks. And take these generated labels to train the model in the target domain. However, these homogeneous networks identify people in approximate subspaces and equally exchange their knowledge with others or their mean net to improve their ability, inevitably limiting the scope of available knowledge and putting them into the same mistake. This article proposes a dual-level asymmetric mutual learning (DAML) method to learn discriminative representations from a broader knowledge scope with diverse embedding spaces. Specifically, two heterogeneous networks mutually learn knowledge from asymmetric subspaces through the pseudo label generation in a hard distillation manner. The knowledge transfer between two networks is based on an asymmetric mutual learning (AML) manner. The teacher network learns to identify both the target and source domain while adapting to the target domain distribution based on the knowledge of the student. Meanwhile, the student network is trained on the target dataset and employs the ground-truth label through the knowledge of the teacher. Extensive experiments in Market-1501, CUHK-SYSU, and MSMT17 public datasets verified the superiority of DAML over state-of-the-arts (SOTA).

Visible-infrared person re-identification (VI-ReID) is the task of matching the same individuals across the visible and infrared modalities. Its main challenge lies in the modality gap caused by the cameras operating on different spectra. Existing VI-ReID methods mainly focus on learning general features across modalities, often at the expense of feature discriminability. To address this issue, we present a novel cycle-construction-based network for neutral yet discriminative feature learning, termed CycleTrans. Specifically, CycleTrans uses a lightweight knowledge capturing module (KCM) to capture rich semantics from the modality-relevant feature maps according to pseudo anchors. Afterward, a discrepancy modeling module (DMM) is deployed to transform these features into neutral ones according to the modality-irrelevant prototypes. To ensure feature discriminability, another two KCMs are further deployed for feature cycle constructions. With cycle construction, our method can learn effective neutral features for visible and infrared images while preserving their salient semantics. Extensive experiments on SYSU-MM01 and RegDB datasets validate the merits of CycleTrans against a flurry of state-of-the-art (SOTA) methods, on rank-1 in SYSU-MM01 and on rank-1 in RegDB. Our code is available at https://github.com/DoubtedSteam/CycleTrans.

While the deep convolutional neural network (DCNN) has achieved overwhelming success in various vision tasks, its heavy computational and storage overhead hinders the practical use of resource-constrained devices. Recently, compressing DCNN models has attracted increasing attention, where binarization-based schemes have generated great research popularity due to their high compression rate. In this article, we propose modulated convolutional networks (MCNs) to obtain binarized DCNNs with high performance. We lead a new architecture in MCNs to efficiently fuse the multiple features and achieve a similar performance as the full-precision model. The calculation of MCNs is theoretically reformulated as a discrete optimization problem to build binarized DCNNs, for the first time, which jointly consider the filter loss, center loss, and softmax loss in a unified framework. Our MCNs are generic and can decompose full-precision filters in DCNNs, e.g., conventional DCNNs, VGG, AlexNet, ResNets, or Wide-ResNets, into a compact set of binarized filters which are optimized based on a projection function and a new updated rule during the backpropagation. Moreover, we propose modulation filters (M-Filters) to recover filters from binarized ones, which lead to a specific architecture to calculate the network model. Our proposed MCNs substantially reduce the storage cost of convolutional filters by a factor of 32 with a comparable performance to the full-precision counterparts, achieving much better performance than other state-of-the-art binarized models.

Although In-Context Learning (ICL) brings remarkable performance gains to Large Language Models (LLMs), the improvements remain lower than fine-tuning on downstream tasks. This paper introduces Multi-Modal In-Context Tuning (MMICT), a novel multi-modal fine-tuning paradigm that boosts multi-modal fine-tuning by fully leveraging the promising ICL capability of Multi-Modal LLMs (MM-LLMs). We propose the Multi-Modal Hub (M-Hub), a unified module that captures various multi-modal features according to different inputs and objectives. Based on M-Hub, MMICT enables MM-LLMs to learn from in-context visual-guided textual features and subsequently generate outputs conditioned on the textual-guided visual features. Moreover, leveraging the flexibility of M-Hub, we design a variety of in-context demonstrations. Extensive experiments on a diverse range of downstream multi-modal tasks demonstrate that MMICT significantly outperforms traditional fine-tuning strategy and the vanilla ICT method that directly takes the concatenation of all information from different modalities as input. Our implementation is available at: https://github.com/KDEGroup/MMICT .

In recent times, automatic text-to-3D content creation has made significant progress, driven by the development of pretrained 2D diffusion models. Existing text-to-3D methods typically optimize the 3D representation to ensure that the rendered image aligns well with the given text, as evaluated by the pretrained 2D diffusion model. Nevertheless, a substantial domain gap exists between 2D images and 3D assets, primarily attributed to variations in camera-related attributes and the exclusive presence of foreground objects. Consequently, employing 2D diffusion models directly for optimizing 3D representations may lead to suboptimal outcomes. To address this issue, we present X-Dreamer, a novel approach for high-quality text-to-3D content creation that effectively bridges the gap between text-to-2D and text-to-3D synthesis. The key components of X-Dreamer are two innovative designs: Camera-Guided Low-Rank Adaptation (CG-LoRA) and Attention-Mask Alignment (AMA) Loss. CG-LoRA dynamically incorporates camera information into the pretrained diffusion models by employing camera-dependent generation for trainable parameters. This integration makes the 2D diffusion model camera-sensitive. AMA loss guides the attention map of the pretrained diffusion model using the binary mask of the 3D object, prioritizing the creation of the foreground object. This module ensures that the model focuses on generating accurate and detailed foreground objects. Extensive evaluations demonstrate the effectiveness of our proposed method compared to existing text-to-3D approaches. Our project webpage: https://anonymous-11111.github.io/ . Our code is available at https://github.com/xmu-xiaoma666/X-Dreamer .

This work is oriented toward the task of open-set Human Object Interaction (HOI) detection. The challenge lies in identifying completely new, out-of-domain relationships, as opposed to in-domain ones which have seen improvements in zero-shot HOI detection. To address this challenge, we introduce a simple Disentangled HOI Detection (DHD) model for detecting novel relationships by integrating an open-set object detector with a Visual Language Model (VLM). We utilize a disentangled image-text contrastive learning metric for training and connect the bottom-up visual features to text embeddings through lightweight unary and pair-wise adapters. Our model can benefit from the open-set object detector and the VLM to detect novel action categories and combine actions with novel object categories. We further present the VG-HOI dataset, a comprehensive benchmark with over 17k HOI relationships for open-set scenarios. Experimental results show that our model can detect unknown action classes and combine unknown object classes. Furthermore, it can generalize to over 17k HOI classes while being trained on just 600 HOI classes.

This paper focuses on addressing the issue of image demoiréing. Unlike the large volume of existing studies that rely on learning from paired real data, we attempt to learn a demoiréing model from unpaired real data, i.e., moiré images associated with irrelevant clean images. The proposed method, referred to as Unpaired Demoiréing(UnDeM), synthesizes pseudo moiré images from unpaired datasets, generating pairs with clean images for training demoiréing models. To achieve this, we divide real moiré images into patches and group them in compliance with their moiré complexity. We introduce a novel moiré generation framework to synthesize moiré images with diverse moiré features, resembling real moiré patches, and details akin to real moiré-free images. Additionally, we introduce an adaptive denoise method to eliminate the low-quality pseudo moiré images that adversely impact the learning of demoiréing models. We conduct extensive experiments on the commonly-used FHDMi and UHDM datasets. Results manifest that our UnDeM performs better than existing methods when using existing demoiréing models such as MBCNN and ESDNet-L. Code: https://github.com/zysxmu/UnDeM.

Given the long textual product information and the product image, Multi-modal Product Summarization (MPS) aims to increase customers’ desire to purchase by highlighting product characteristics with a short textual summary. Existing MPS methods can produce promising results. Nevertheless, they still 1) lack end-to-end product summarization, 2) lack multi-grained multi-modal modeling, and 3) lack multi-modal attribute modeling. To improve MPS, we propose an end-to-end multi-grained multi-modal attribute-aware product summarization method (MMAPS) for generating high-quality product summaries in e-commerce. MMAPS jointly models product attributes and generates product summaries. We design several multi-grained multi-modal tasks to better guide the multi-modal learning of MMAPS. Furthermore, we model product attributes based on both text and image modalities so that multi-modal product characteristics can be manifested in the generated summaries. Extensive experiments on a real large-scale Chinese e-commence dataset demonstrate that our model outperforms state-of-the-art product summarization methods w.r.t. several summarization metrics. Our code is publicly available at: https://github.com/KDEGroup/MMAPS.

The Segment Anything Model (SAM) marks a notable milestone in segmentation models, highlighted by its robust zero-shot capabilities and ability to handle diverse prompts. SAM follows a pipeline that separates interactive segmentation into image preprocessing through a large encoder and interactive inference via a lightweight decoder, ensuring efficient real-time performance. However, SAM faces stability issues in challenging samples upon this pipeline. These issues arise from two main factors. Firstly, the image preprocessing disables SAM to dynamically use image-level zoom-in strategies to refocus on the target object during interaction. Secondly, the lightweight decoder struggles to sufficiently integrate interactive information with image embeddings. To address these two limitations, we propose FocSAM with a pipeline redesigned on two pivotal aspects. First, we propose Dynamic Window Multi-head Self-Attention (Dwin-MSA) to dynamically refocus SAM's image embeddings on the target object. Dwin-MSA localizes attention computations around the target object, enhancing object-related embeddings with minimal computational overhead. Second, we propose Pixel-wise Dynamic ReLU (P-DyReLU) to enable sufficient integration of interactive information from a few initial clicks that have significant impacts on the overall segmentation results. Experimentally, FocSAM augments SAM's interactive segmentation performance to match the existing state-of-the-art method in segmentation quality, requiring only about 5.6% of this method's inference time on CPUs. Code is available at https://github.com/YouHuang67/focsam.

Referring Remote Sensing Image Segmentation (RRSIS) is a new challenge that combines computer vision and natu-ral language processing. Traditional Referring Image Seg-mentation (RIS) approaches have been impeded by the com-plex spatial scales and orientations found in aerial imagery, leading to suboptimal segmentation results. To address these challenges, we introduce the Rotated Multi-Scale In-teraction Network (RMSIN), an innovative approach de-signed for the unique demands of RRSIS. RMSIN incorpo-rates an Intra-scale Interaction Module (IIM) to effectively address the fine-grained detail required at multiple scales and a Cross-scale Interaction Module (CIM) for integrating these details coherently across the network. Furthermore, RMSIN employs an Adaptive Rotated Convolution (ARC) to account for the diverse orientations of objects, a novel contribution that significantly enhances segmentation accu-racy. To assess the efficacy of RMSIN, we have curated an expansive dataset comprising 17,402 image-caption-mask triplets, which is unparalleled in terms of scale and vari-ety. This dataset not only presents the model with a wide range of spatial and rotational scenarios but also estab-lishes a stringent benchmark for the RRSIS task, ensuring a rigorous evaluation of performance. Experimental eval-uations demonstrate the exceptional performance of RM-SIN, surpassing existing state-of-the-art models by a signif-icant margin. Datasets and code are available at https://github.com/Lsan2401/RMSIN.

Recent advancements in personalized image generation using diffusion models have been noteworthy. However, existing methods suffer from inefficiencies due to the requirement for subject-specific fine-tuning. This computationally intensive process hinders efficient deployment, limiting practical usability. Moreover, these methods often grapple with identity distortion and limited expression diversity. In light of these challenges, we propose Portrait-Booth, an innovative approach designed for high efficiency, robust identity preservation, and expression-editable text-to-image generation, without the need for fine-tuning. Por-traitBooth leverages subject embeddingsfrom aface recognition model for personalized image generation without fine-tuning. It eliminates computational overhead and mitigates identity distortion. The introduced dynamic identity preservation strategy further ensures close resemblance to the original image identity. Moreover, PortraitBooth incorporates emotion-aware cross-attention control for diverse facial expressions in generated images, supporting text-driven expression editing. Its scalability enables efficient and high-quality image creation, including multi-subject generation. Extensive results demonstrate superior performance over other state-of-the-art methods in both single and multiple image generation scenarios. Our project page is at https://portraitbooth.github.io.

Vision foundation models have been explored recently to build general-purpose vision systems. However, predomi-nant paradigms, driven by casting instance-level tasks as an object-word alignment, bring heavy cross-modality in-teraction, which is not effective in prompting object detection and visual grounding. Another line of work that fo-cuses on pixel-level tasks often encounters a large annotation gap of things and stuff, and suffers from mutual inter-ference between foreground-object and background-class segmentation. In stark contrast to the prevailing methods, we present APE, a universal visual perception model for aligning and prompting everything all at once in an image to perform diverse tasks, i.e., detection, segmentation, and grounding, as an instance-level sentence-object matching paradigm. Specifically, APE advances the convergence of detection and grounding by reformulating language-guided grounding as open-vocabulary detection, which efficiently scales up model prompting to thousands of category vocab-ularies and region descriptions while maintaining the ef-fectiveness of cross-modality fusion. To bridge the granu-larity gap of different pixel-level tasks, APE equalizes se-mantic and panoptic segmentation to proxy instance learning by considering any isolated regions as individual in-stances. APE aligns vision and language representation on broad data with natural and challenging characteristics all at once without task-specific fine-tuning. The extensive ex-periments on over 160 datasets demonstrate that, with only one-suit of weights, APE outperforms (or is on par with) the state-of-the-art models, proving that an effective yet univer-sal perception for anything aligning and prompting is in-deed feasible. Codes and trained models are released at https://github.com/shenyunhang/APE.

Post-training Sparsity (PTS) is a recently emerged av-enue that chases efficient network sparsity with limited data in need. Existing PTS methods, however, undergo significant performance degradation compared with traditional methods that retrain the sparse networks via the whole dataset, especially at high sparsity ratios. In this paper, we attempt to reconcile this disparity by transposing three cardinal factors that profoundly alter the performance of conventional sparsity into the context of PTS. Our endeav-ors particularly comprise (1) A base-decayed sparsity objective that promotes efficient knowledge transferring from dense network to the sparse counterpart. (2) A reducing-regrowing search algorithm designed to ascertain the op-timal sparsity distribution while circumventing overfitting to the small calibration set in PTS. (3) The employment of dynamic sparse training predicated on the preceding as-pects, aimed at comprehensively optimizing the sparsity structure while ensuring training stability. Our proposed framework, termed UniPTS, is validated to be much superior to existing PTS methods across extensive benchmarks. As an illustration, it amplifies the performance of POT, a recently proposed recipe, from 3.9% to 68.6% when pruning ResNet-50 at 90% sparsity ratio on ImageNet. We re-lease the code of our paper at https://github.com/xjjxmu/UniPTS.

The rich spatio-temporal information is crucial to capture the complicated target appearance variations in visual tracking. However, most top-performing tracking algorithms rely on many hand-crafted components for spatio-temporal information aggregation. Consequently, the spatio-temporal information is far away from being fully explored. To alleviate this issue, we propose an adaptive tracker with spatio-temporal transformers (named AQA-Track), which adopts simple autoregressive queries to effectively learn spatio-temporal information without many hand-designed components. Firstly, we introduce a set of learnable and autoregressive queries to capture the instantaneous target appearance changes in a sliding window fashion. Then, we design a novel attention mechanism for the interaction of existing queries to generate a new query in current frame. Finally, based on the initial target template and learnt autoregressive queries, a spatio-temporal information fusion module (STM) is designed for spatiotemporal formation aggregation to locate a target object. Benefiting from the STM, we can effectively combine the static appearance and instantaneous changes to guide robust tracking. Extensive experiments show that our method significantly improves the tracker's performance on six popular tracking benchmarks: LaSOT, LaSOText, TrackingNet, GOT-10k, TNL2K, and UAV123. Code and models will be https://github.com/orgs/GXNU-ZhongLab.

Interactive Segmentation (IS) segments specific objects or parts in the image according to user input. Current IS pipelines fall into two categories: single-granularity out-put and multi-granularity output. The latter aims to allevi-ate the spatial ambiguity present in the former. However, the multi-granularity output pipeline suffers from limited interaction flexibility and produces redundant results. In this work, we introduce Granularity-Controllable Interactive Segmentation (GraCo), a novel approach that allows precise control of prediction granularity by introducing ad-ditional parameters to input. This enhances the customization of the interactive system and eliminates redundancy while resolving ambiguity. Nevertheless, the exorbitant cost of annotating multi-granularity masks and the lack of avail-able datasets with granularity annotations make it difficult for models to acquire the necessary guidance to control out-put granularity. To address this problem, we design an any-granularity mask generator that exploits the semantic property of the pre-trained IS model to automatically gen-erate abundant mask-granularity pairs without requiring additional manual annotation. Based on these pairs, we propose a granularity-controllable learning strategy that efficiently imparts the granularity controllability to the IS model. Extensive experiments on intricate scenarios at ob-ject and part levels demonstrate that our GraCo has signifi-cant advantages over previous methods. This highlights the potential of GraCo to be a flexible annotation tool, capable of adapting to diverse segmentation scenarios. The project page: https://zhao-yian.github.io/GraCo.

Text-to-image (T2I) generative models have attracted significant attention and found extensive applications within and beyond academic research. For example, the Civitai community, a platform for T2I innovation, currently hosts an impressive array of 74,492 distinct models. However, this diversity presents a formidable challenge in selecting the most appropriate model and parameters, a process that typically requires numerous trials. Drawing inspiration from the tool usage research of large language models (LLMs), we introduce DiffAgent, an LLM agent designed to screen the accurate selection in seconds via API calls. DiffAgent leverages a novel two-stage training framework, SFTA, enabling it to accurately align T2I API responses with user input in accordance with human preferences. To train and evaluate DiffAgent's capabilities, we present DABench, a comprehensive dataset encompassing an extensive range of T2I APIs from the community. Our evaluations reveal that DiffAgent not only excels in identifying the appropriate T2I API but also underscores the effectiveness of the SFTA training framework. Codes are available at https://github.com/OpenGVLab/DiffAgent.

Existing multi-label knowledge distillation methods simply use regression or single-label classification methods without fully exploiting the essence of multi-label classification, resulting in student models’ inadequate performance and poor functional similarity to teacher models. In this paper, we reinterpret multi-label classification as multiple intra-class ranking tasks, with each class corresponding to a ranking task. Furthermore, we define the knowledge of multi-label classification models as the ranking of intra-class samples. On the one hand, we propose to evaluate the functional similarity between multi-label classification models with Kendall’s tau and rank-biased overlap, which are common metrics for evaluating ranking similarity. On the other hand, we propose a new functionally similar multi-label knowledge distillation method called FSD, which enables student models to learn the ranking of intra-class samples from teacher models. Finally, experimental results validate that FSD outperforms existing methods, especially for functional similarity. Specifically, we achieve a mAP of 73.38% and a mKDT of 0.686 on COCO, which are 2.22% and 0.19 better than existing methods, respectively.

The ever-increasing large language models (LLMs), though opening a potential path for the upcoming artificial general intelligence, sadly drops a daunting obstacle on the way towards their on-device deployment. As one of the most well-established pre-LLMs approaches in reducing model complexity, network pruning appears to lag behind in the era of LLMs, due mostly to its costly fine-tuning (or re-training) necessity under the massive volumes of model parameter and training data. To close this industry-academia gap, we introduce Dynamic Sparse No Training ($\texttt{DSNT}$), a training-free fine-tuning approach that slightly updates sparse LLMs without the expensive backpropagation and any weight updates. Inspired by the Dynamic Sparse Training, $\texttt{DSNT}$ minimizes the reconstruction error between the dense and sparse LLMs, in the fashion of performing iterative weight pruning-and-growing on top of sparse LLMs. To accomplish this purpose, $\texttt{DSNT}$ particularly takes into account the anticipated reduction in reconstruction error for pruning and growing, as well as the variance w.r.t. different input data for growing each weight. This practice can be executed efficiently in linear time since its obviates the need of backpropagation for fine-tuning LLMs. Extensive experiments on LLaMA-V1/V2, Vicuna, and OPT across various benchmarks demonstrate the effectiveness of $\texttt{DSNT}$ in enhancing the performance of sparse LLMs, especially at high sparsity levels. For instance, $\texttt{DSNT}$ is able to outperform the state-of-the-art Wanda by 26.79 perplexity at 70% sparsity with LLaMA-7B. Our paper offers fresh insights into how to fine-tune sparse LLMs in an efficient training-free manner and open new venues to scale the great potential of sparsity to LLMs. Codes are available at https://github.com/zyxxmu/DSnoT.

Masked autoencoders have become popular training paradigms for self-supervised visual representation learning. These models randomly mask a portion of the input and reconstruct the masked portion according to assigned target representations. In this paper, we show that a careful choice of the target representation is unnecessary for learning good visual representation since different targets tend to derive similarly behaved models. Driven by this observation, we propose a multi-stage masked distillation pipeline and use a randomly initialized model as the teacher, enabling us to effectively train high-capacity models without any effort to carefully design the target representation. On various downstream tasks, the proposed method to perform masked knowledge distillation with bootstrapped teachers (dbot) outperforms previous self-supervised methods by nontrivial margins. We hope our findings, as well as the proposed method, could motivate people to rethink the roles of target representations in pre-training masked autoencoders.

The significant resource requirements associated with Large-scale Language Models (LLMs) have generated considerable interest in the development of techniques aimed at compressing and accelerating neural networks. Among these techniques, Post-Training Quantization (PTQ) has emerged as a subject of considerable interest due to its noteworthy compression efficiency and cost-effectiveness in the context of training. Existing PTQ methods for LLMs limit the optimization scope to scaling transformations between pre- and post-quantization weights. This constraint results in significant errors after quantization, particularly in low-bit configurations. In this paper, we advocate for the direct optimization using equivalent Affine transformations in PTQ (AffineQuant). This approach extends the optimization scope and thus significantly minimizing quantization errors. Additionally, by employing the corresponding inverse matrix, we can ensure equivalence between the pre- and post-quantization outputs of PTQ, thereby maintaining its efficiency and generalization capabilities. To ensure the invertibility of the transformation during optimization, we further introduce a gradual mask optimization method. This method initially focuses on optimizing the diagonal elements and gradually extends to the other elements. Such an approach aligns with the Levy-Desplanques theorem, theoretically ensuring invertibility of the transformation. As a result, significant performance improvements are evident across different LLMs on diverse datasets. Notably, these improvements are most pronounced when using very low-bit quantization, enabling the deployment of large models on edge devices. To illustrate, we attain a C4 perplexity of $15.76$ (2.26$\downarrow$ vs $18.02$ in OmniQuant) on the LLaMA2-$7$B model of W$4$A$4$ quantization without overhead. On zero-shot tasks, AffineQuant achieves an average of $58.61\%$ accuracy ( $1.98\%\uparrow$ vs $56.63$ in OmniQuant) when using $4$/$4$-bit quantization for LLaMA-$30$B, which setting a new state-of-the-art benchmark for PTQ in LLMs. Codes are available at: https://github.com/bytedance/AffineQuant.

Despite the exceptional performance of Large Language Models (LLMs), the substantial volume of key-value (KV) pairs cached during inference presents a barrier to their efficient deployment. To ameliorate this, recent works have aimed to selectively eliminate these caches, informed by the attention scores of associated tokens. However, such cache eviction invariably leads to output perturbation, regardless of the token choice. This perturbation escalates with the compression ratio, which can precipitate a marked deterioration in LLM inference performance. This paper introduces Cache Merging (CaM) as a solution to mitigate this challenge. CaM adaptively merges to-be-evicted caches into the remaining ones, employing a novel sampling strategy governed by the prominence of attention scores within discarded locations. In this manner, CaM enables memory-efficient LLMs to preserve critical token information, even obviating the need to maintain their corresponding caches. Extensive experiments utilizing LLaMA, OPT, and GPT-NeoX across various benchmarks corroborate CaM’s proficiency in bolstering the performance of memory-efficient LLMs. Code is released at https://github.com/zyxxmu/cam.

Text-to-3D-aware face (T3D Face) generation and manipulation is an emerging research hot spot in machine learning, which still suffers from low efficiency and poor quality. In this paper, we propose an ***E**nd-to-End **E**fficient and **E**ffective* network for fast and accurate T3D face generation and manipulation, termed $E^3$-FaceNet. Different from existing complex generation paradigms, $E^3$-FaceNet resorts to a direct mapping from text instructions to 3D-aware visual space. We introduce a novel *Style Code Enhancer* to enhance cross-modal semantic alignment, alongside an innovative *Geometric Regularization* objective to maintain consistency across multi-view generations. Extensive experiments on three benchmark datasets demonstrate that $E^3$-FaceNet can not only achieve picture-like 3D face generation and manipulation, but also improve inference speed by orders of magnitudes. For instance, compared with Latent3D, $E^3$-FaceNet speeds up the five-view generations by almost 470 times, while still exceeding in generation quality. Our code is released at <https://github.com/Aria-Zhangjl/E3-FaceNet>.

The current state-of-the-art No-Reference Image Quality Assessment (NR-IQA) methods typically rely on feature extraction from upstream semantic backbone networks, assuming that all extracted features are relevant. However, we make a key observation that not all features are beneficial, and some may even be harmful, necessitating careful selection. Empirically, we find that many image pairs with small feature spatial distances can have vastly different quality scores, indicating that the extracted features may contain quality-irrelevant noise. To address this issue, we propose a Quality-Aware Feature Matching IQA Metric (QFM-IQM) that employs an adversarial perspective to remove harmful semantic noise features from the upstream task. Specifically, QFM-IQM enhances the semantic noise distinguish capabilities by matching image pairs with similar quality scores but varying semantic features as adversarial semantic noise and adaptively adjusting the upstream task’s features by reducing sensitivity to adversarial noise perturbation. Furthermore, we utilize a distillation framework to expand the dataset and improve the model’s generalization ability. Extensive experiments conducted on eight standard IQA datasets have demonstrated the effectiveness of our proposed QFM-IQM.

Blind Image Quality Assessment (BIQA) mirrors subjective made by human observers. Generally, humans favor comparing relative qualities over predicting absolute qualities directly. However, current BIQA models focus on mining the "local" context, i.e., the relationship between information among individual images and the absolute quality of the image, ignoring the "global" context of the relative quality contrast among different images in the training data. In this paper, we present the Perceptual Context and Sensitivity BIQA (CSIQA), a novel contrastive learning paradigm that seamlessly integrates "global” and "local” perspectives into the BIQA. Specifically, the CSIQA comprises two primary components: 1) A Quality Context Contrastive Learning module, which is equipped with different contrastive learning strategies to effectively capture potential quality correlations in the global context of the dataset. 2) A Quality-aware Mask Attention Module, which employs the random mask to ensure the consistency with visual local sensitivity, thereby improving the model’s perception of local distortions. Extensive experiments on eight standard BIQA datasets demonstrate the superior performance to the state-of-the-art BIQA methods.

Recent advancements in automatic 3D avatar generation guided by text have made significant progress. However, existing methods have limitations such as oversaturation and low-quality output. To address these challenges, we propose X-Oscar, a progressive framework for generating high-quality animatable avatars from text prompts. It follows a sequential "Geometry→Texture→Animation" paradigm, simplifying optimization through step-by-step generation. To tackle oversaturation, we introduce Adaptive Variational Parameter (AVP), representing avatars as an adaptive distribution during training. Additionally, we present Avatar-aware Score Distillation Sampling (ASDS), a novel technique that incorporates avatar-aware noise into rendered images for improved generation quality during optimization. Extensive evaluations confirm the superiority of X-Oscar over existing text-to-3D and text-to-avatar approaches. Our anonymous project page: https://anonymous1440.github.io/.

Post-Training Quantization (PTQ) is a vital technique for network compression and acceleration, gaining prominence as model sizes increase. This paper addresses a critical challenge in PTQ: the severe impact of outliers on the accuracy of quantized transformer architectures. Specifically, we introduce the concept of ‘reconstruction granularity’ as a novel solution to this issue, which has been overlooked in previous works. Our work provides theoretical insights into the role of reconstruction granularity in mitigating the outlier problem in transformer models. This theoretical framework is supported by empirical analysis, demonstrating that varying reconstruction granularities significantly influence quantization performance. Our findings indicate that different architectural designs necessitate distinct optimal reconstruction granularities. For instance, the multi-stage Swin Transformer architecture benefits from finer granularity, a deviation from the trends observed in ViT and DeiT models. We further develop an algorithm for determining the optimal reconstruction granularity for various ViT models, achieving state-of-the-art (SOTA) performance in PTQ. For example, applying our method to $4$-bit quantization, the Swin-Base model achieves a Top-1 accuracy of $82.24%$ on the ImageNet classification task. This result surpasses the RepQ-ViT by $3.92%$ ($82.24%$ VS $78.32%$). Similarly, our approach elevates the ViT-Small to a Top-1 accuracy of $80.50%$, outperforming NoisyQuant by $3.64%$ ($80.50%$ VS $76.86%$). Codes are available in Supplementary Materials.

The issue of hallucinations is a prevalent concern in existing Large Vision-Language Models (LVLMs). Previous efforts have primarily focused on investigating object hallucinations, which can be easily alleviated by introducing object detectors. However, these efforts neglect hallucinations in inter-object relationships, which is essential for visual comprehension. In this work, we introduce R-Bench, a novel benchmark for evaluating Vision Relationship Hallucination. R-Bench features image-level questions that focus on the existence of relationships and instance-level questions that assess local visual comprehension. We identify three types of relationship co-occurrences that lead to hallucinations: relationship-relationship, subject-relationship, and relationship-object. The visual instruction tuning dataset's long-tail distribution significantly impacts LVLMs' understanding of visual relationships. Additionally, our analysis reveals that current LVLMs tend to overlook visual content, overly rely on the common sense knowledge of Large Language Models (LLMs), and struggle with spatial relationship reasoning based on contextual information.

In this paper, we introduce SemiRES, a semi-supervised framework that effectively leverages a combination of labeled and unlabeled data to perform RES. A significant hurdle in applying semi-supervised techniques to RES is the prevalence of noisy pseudo-labels, particularly at the boundaries of objects. SemiRES incorporates the Segment Anything Model (SAM), renowned for its precise boundary demarcation, to improve the accuracy of these pseudo-labels. Within SemiRES, we offer two alternative matching strategies: IoU-based Optimal Matching (IOM) and Composite Parts Integration (CPI). These strategies are designed to extract the most accurate masks from SAM’s output, thus guiding the training of the student model with enhanced precision. In instances where a precise mask cannot be matched from the available candidates, we develop the Pixel-Wise Adjustment (PWA) strategy, guiding the student model’s training directly by the pseudo-labels. Extensive experiments on three RES benchmarks—RefCOCO, RefCOCO+, and G-Ref reveal its superior performance compared to fully supervised methods, especially in low-data scenarios. Remarkably, with only 1% labeled data, our SemiRES outperforms the supervised baseline by a large margin, e.g. +18.64% gains on RefCOCO val set.

Post-training quantization (PTQ) for vision transformers (ViTs) has garnered significant attention due to its efficiency in compressing models. However, existing methods typically overlook the intricate interdependence between quantized weight and activation, leading to considerable quantization error. In this paper, we propose ERQ, a two-step PTQ approach meticulously crafted to sequentially reduce the quantization error arising from activation and weight quantization. ERQ first introduces Activation quantization error reduction (Aqer) that strategically formulates the minimization of activation quantization error as a Ridge Regression problem, tackling it by updating weights with full-precision. Subsequently, ERQ introduces Weight quantization error reduction (Wqer) that adopts an iterative approach to mitigate the quantization error induced by weight quantization. In each iteration, an empirically derived, efficient proxy is employed to refine the rounding directions of quantized weights, coupled with a Ridge Regression solver to curtail weight quantization error. Experimental results attest to the effectiveness of our approach. Notably, ERQ surpasses the state-of-the-art GPTQ by 22.36% in accuracy for W3A4 ViT-S.

Visual grounding is a task of locating the object referred by a natural language description. To reduce annotation costs, recent researchers are devoted into one-stage weakly supervised methods for visual grounding, which typically adopt the anchor-text matching paradigm. Despite the efficiency, we identify that anchor representations are often noisy and insufficient to describe object information, which inevitably hinders the vision-language alignments. In this paper, we propose a novel query-based one-stage framework for weakly supervised visual grounding, namely QueryMatch. Different from previous work, QueryMatch represents candidate objects with a set of query features, which inherently establish accurate one-to-one associations with visual objects. In this case, QueryMatch re-formulates weakly supervised visual grounding as a query-text matching problem, which can be optimized via the query-based contrastive learning. Based on QueryMatch, we further propose an innovative strategy for effective weakly supervised learning, namely Active Query Selection (AQS). In particular, AQS aims to enhance the effectiveness of query-based contrastive learning by actively selecting high-quality query features. Through this strategy, AQS can greatly benefit the weakly supervised learning of QueryMatch. To validate our approach, we conduct extensive experiments on three benchmark datasets of two grounding tasks, i.e., referring expression comprehension (REC) and segmentation (RES). Experimental results not only show the state-of-art performance of QueryMatch in two tasks, e.g., over +5% [email protected] on RefCOCO in REC and over +20% mIOU on RefCOCO in RES, but also confirm the effectiveness of AQS in weakly supervised learning. Source codes are available at https://github.com/TensorThinker/QueryMatch.

With the advent of large language models(LLMs) enhanced by the chain-of-thought(CoT) methodology, the visual reasoning problem is usually decomposed into manageable sub-tasks and tackled sequentially with various external tools. However, such a paradigm faces the challenge of the potential "determining hallucinations" in decision generation due to insufficient visual information and the limitation of low-level perception tools that fail to provide abstract summaries necessary for comprehensive reasoning. We argue that converging visual context acquisition and logical reasoning is pivotal for tackling visual reasoning tasks. This paper delves into the realm of multimodal CoT to solve intricate visual reasoning tasks with multimodal large language models(MLLMs) and their cognitive capability. To this end, we propose an innovative multimodal CoT framework, termed Cantor, characterized by a perception-decision architecture. Cantor first acts as a decision generator and integrates visual inputs to analyze the image and problem, ensuring a closer alignment with the actual context. Furthermore, Cantor leverages the advanced cognitive functions of MLLMs to perform as multifaceted experts for deriving higher-level information, enhancing the CoT generation process. Our extensive experiments demonstrate the efficacy of the proposed framework, showing significant improvements in multimodal CoT performance across two complex visual reasoning datasets, without necessitating fine-tuning or ground-truth rationales. Project Page: https://ggg0919.github.io/cantor/.

Foundational segmentation models, predominantly trained on scenes typical of natural environments, struggle to generalize across varied image domains. Traditional "training-to-adapt'' methods rely heavily on extensive data retraining and model architectures modifications. This significantly limits the models' generalization capabilities and efficiency in deployment. In this study, we propose a novel adaptation paradigm, termed "prompting-to-adapt'', to tackle the above issue by introducing an innovative image prompter. This prompter generates domain-specific prompts through few-shot image-mask pairs, incorporating diverse image processing techniques to enhance adaptability. To tackle the inherent non-differentiability of image prompts, we further devise an information-estimation-based gradient descent strategy that leverages the information entropy of image processing combinations to optimize the prompter, ensuring effective adaptation. Through extensive experiments across nine datasets spanning seven image domains (i.e., depth, thermal, camouflage, endoscopic, ultrasound, grayscale, and natural) and four scenarios (i.e., common scenes, camouflage objects, medical images, and industrial data), we demonstrate that our approach significant improves the foundational models' adaptation capabilities. Moreover, the interpretability of the generated prompts provides insightful revelations into their image processing mechanisms. Source code is available at: \urlgithub.com/yuema1303/Prompting-to-Adapt-FSM.

Recently, Segment Anything Model (SAM) has become a research hotspot in the fields of multimedia and computer vision, which exhibits powerful yet versatile capabilities on various (un) conditional image segmentation tasks. Although SAM can support different types of segmentation prompts, we note that, compared to point- and box-guided segmentations, it performs much worse on text-instructed tasks, e.g., referring image segmentation (RIS). In this paper, we argue that deep text instruction tuning is key to mitigate such shortcoming caused by the shallow fusion scheme in its default light-weight mask decoder. To address this issue, we propose two simple yet effective deep instruction tuning (DIT) methods for SAM, one is end-to-end and the other is layer-wise. With minimal modifications, DITs can directly transform the image encoder of SAM as a stand-alone vision-language learner in contrast to building another deep fusion branch, maximizing the benefit of its superior segmentation capability. Extensive experiments on three highly competitive benchmark datasets of RIS show that a simple end-to-end DIT can improve SAM by a large margin, while the layer-wise DIT can further boost the performance to state-of-the-art with much less data and training expenditures. Our code is released at: https://github.com/wysnzzzz/DIT.

The integration of large language models into open-world detection frameworks significantly improves versatility in new environments. Prompt representations derived from these models help establish classification boundaries for both base and novel categories within open-world detectors. However, we are the first to discover that directly fine-tuning language models in detection systems results in redundant attention patterns and leads to suboptimal prompt representations. In order to fully leverage the capabilities of large language models and augment prompt encoding for detection, this study introduces a redundancy assessment metric to identify uniform attention patterns. Furthermore, in areas with high redundancy, we incorporate multimodal inplace prompt tuning (MIPT) to enrich the text prompt with visual clues. Experimental results validate the efficacy of our MIPT framework, achieving a notable increase across benchmarks, e.g. elevating GLIP-L from 22.6% to 25.0% on ODinW-35, and 9.0% improvement on LVIS.

3D Referring Expression Segmentation (3D-RES) is dedicated to segmenting a specific instance within a 3D space based on a natural language description. However, current approaches are limited to segmenting a single target, restricting the versatility of the task. To overcome this limitation, we introduce Generalized 3D Referring Expression Segmentation (3D-GRES), which extends the capability to segment any number of instances based on natural language instructions. In addressing this broader task, we propose the Multi-Query Decoupled Interaction Network (MDIN), designed to break down multi-object segmentation tasks into simpler, individual segmentations. MDIN comprises two fundamental components: Text-driven Sparse Queries (TSQ) and Multi-object Decoupling Optimization (MDO). TSQ generates sparse point cloud features distributed over key targets as the initialization for queries. Meanwhile, MDO is tasked with assigning each target in multi-object scenarios to different queries while maintaining their semantic consistency. To adapt to this new task, we build a new dataset, namely Multi3DRes. Our comprehensive evaluations on this dataset demonstrate substantial enhancements over existing models, thus charting a new path for intricate multi-object 3D scene comprehension. The benchmark and code are available at https://github.com/sosppxo/MDIN.

This paper presents the first study to explore the potential of parameter quantization for multimodal large language models to alleviate the significant resource constraint encountered during vision-language instruction tuning. We introduce a Quantization-aware Scale LeArning method based on multimodal Warmup, termed QSLAW. This method is grounded in two key innovations: (1) The learning of group-wise scale factors for quantized LLM weights to mitigate the quantization error arising from activation outliers and achieve more effective vision-language instruction tuning; (2) The implementation of a multimodal warmup that progressively integrates linguistic and multimodal training samples, thereby preventing overfitting of the quantized model to multimodal data while ensuring stable adaptation of multimodal large language models to downstream vision-language tasks. Extensive experiments demonstrate that models quantized by QSLAW perform on par with, or even surpass, their full-precision counterparts, while facilitating up to 1.4 times reduction in VL tuning time and GPU consumption.

Referring image segmentation (RIS) aims to segment a particular region based on a specific expression. Existing one-stage methods have explored various fusion strategies, yet they encounter two significant issues. Primarily, most methods rely on manually selected visual features from the visual encoder layers. Moreover, the direct fusion of word-level features into coarse aligned features disrupts the established vision-language alignment. In this paper, we introduce an innovative framework for RIS that seeks to overcome these challenges with adaptive alignment of vision and language features, termed the Adaptive Selection with Dual Alignment (ASDA). ASDA innovates in two aspects. Firstly, we design an Adaptive Feature Selection and Fusion (AFSF) module to dynamically select visual features focusing on different regions related to various descriptions. AFSF is equipped with scale-wise feature aggregator to provide hierarchically coarse features that preserve crucial low-level details. Secondly, a Word Guided Dual-Branch Aligner (WGDA) is leveraged to integrate coarse features with linguistic cues by word-guided attention, which effectively addresses the common issue of vision-language misalignment. Extensive experimental results demonstrate that our ASDA framework surpasses state-of-the-art methods on RefCOCO, RefCOCO+ and G-Ref benchmark.

The rapid progress in generative models has given rise to the critical task of AI-Generated Content Stealth (AIGC-S), which aims to create AI-generated images that can evade both forensic detectors and human inspection. This task is crucial for understanding the vulnerabilities of existing detection methods and developing more robust techniques. However, current adversarial attacks often introduce visible noise, have poor transferability, and fail to address spectral differences between AI-generated and genuine images. To address this, we propose StealthDiffusion, a framework based on stable diffusion that modifies AI-generated images into high-quality, imperceptible adversarial examples capable of evading state-of-the-art forensic detectors. StealthDiffusion comprises two main components: Latent Adversarial Optimization, which generates adversarial perturbations in the latent space of stable diffusion, and Control-VAE, a module that reduces spectral differences between the generated adversarial images and genuine images without affecting the original diffusion model's generation process. Extensive experiments show that StealthDiffusion is effective in both white-box and black-box settings, transforming AI-generated images into high-quality adversarial forgeries with frequency spectra similar to genuine images. These forgeries are classified as genuine by advanced forensic classifiers and are difficult for humans to distinguish.

The rapid progress of Deepfake technology has made face swapping highly realistic, raising concerns about the malicious use of fabricated facial content. Existing methods often struggle to generalize to unseen domains due to the diverse nature of facial manipulations. In this paper, we revisit the generation process and identify a universal principle: Deepfake images inherently contain information from both source and target identities, while genuine faces maintain a consistent identity. Building upon this insight, we introduce DiffusionFake, a novel plug-and-play framework that reverses the generative process of face forgeries to enhance the generalization of detection models. DiffusionFake achieves this by injecting the features extracted by the detection model into a frozen pre-trained Stable Diffusion model, compelling it to reconstruct the corresponding target and source images. This guided reconstruction process constrains the detection network to capture the source and target related features to facilitate the reconstruction, thereby learning rich and disentangled representations that are more resilient to unseen forgeries. Extensive experiments demonstrate that DiffusionFake significantly improves cross-domain generalization of various detector architectures without introducing additional parameters during inference. The code are available in https://github.com/skJack/DiffusionFake.git.

In recent years, there has been significant research focusing on addressing security concerns in single-modal person re-identification (ReID) systems that are based on RGB images. However, the safety of cross-modality scenarios, which are more commonly encountered in practical applications involving images captured by infrared cameras, has not received adequate attention. The main challenge in cross-modality ReID lies in effectively dealing with visual differences between different modalities. For instance, infrared images are typically grayscale, unlike visible images that contain color information. Existing attack methods have primarily focused on the characteristics of the visible image modality, overlooking the features of other modalities and the variations in data distribution among different modalities. This oversight can potentially undermine the effectiveness of these methods in image retrieval across diverse modalities. This study represents the first exploration into the security of cross-modality ReID models and proposes a universal perturbation attack specifically designed for cross-modality ReID. This attack optimizes perturbations by leveraging gradients from diverse modality data, thereby disrupting the discriminator and reinforcing the differences between modalities. We conducted experiments on three widely used cross-modality datasets, namely RegDB, SYSU, and LLCM. The results not only demonstrate the effectiveness of our method but also provide insights for future improvements in the robustness of cross-modality ReID systems.

Recent advancements in 3D generation have leveraged synthetic datasets with ground truth 3D assets and predefined camera trajectories. However, the potential of adopting real-world datasets, which can produce significantly more realistic 3D scenes, remains largely unexplored. In this work, we delve into the key challenge of the complex and scene-specific camera trajectories found in real-world captures. We introduce Director3D, a robust open-world text-to-3D generation framework, designed to generate both real-world 3D scenes and adaptive camera trajectories. To achieve this, (1) we first utilize a Trajectory Diffusion Transformer, acting as the \emph{Cinematographer}, to model the distribution of camera trajectories based on textual descriptions. Next, a Gaussian-driven Multi-view Latent Diffusion Model serves as the \emph{Decorator}, modeling the image sequence distribution given the camera trajectories and texts. This model, fine-tuned from a 2D diffusion model, directly generates pixel-aligned 3D Gaussians as an immediate 3D scene representation for consistent denoising. Lastly, the 3D Gaussians are further refined by a novel SDS++ loss as the \emph{Detailer}, which incorporates the prior of the 2D diffusion model. Extensive experiments demonstrate that Director3D outperforms existing methods, offering superior performance in real-world 3D generation.

Significant progress has been made in the field of Instruction-based Image Editing (IIE). However, evaluating these models poses a significant challenge. A crucial requirement in this field is the establishment of a comprehensive evaluation benchmark for accurately assessing editing results and providing valuable insights for its further development. In response to this need, we propose I2EBench, a comprehensive benchmark designed to automatically evaluate the quality of edited images produced by IIE models from multiple dimensions. I2EBench consists of 2,000+ images for editing, along with 4,000+ corresponding original and diverse instructions. It offers three distinctive characteristics: 1) Comprehensive Evaluation Dimensions: I2EBench comprises 16 evaluation dimensions that cover both high-level and low-level aspects, providing a comprehensive assessment of each IIE model. 2) Human Perception Alignment: To ensure the alignment of our benchmark with human perception, we conducted an extensive user study for each evaluation dimension. 3) Valuable Research Insights: By analyzing the advantages and disadvantages of existing IIE models across the 16 dimensions, we offer valuable research insights to guide future development in the field. We will open-source I2EBench, including all instructions, input images, human annotations, edited images from all evaluated methods, and a simple script for evaluating the results from new IIE models. The code, dataset, and generated images from all IIE models are provided in GitHub: https://github.com/cocoshe/I2EBench.

This paper makes a step towards modeling the modality discrepancy in the cross-spectral re-identification task. Based on the Lambertain model, we observe that the non-linear modality discrepancy mainly comes from diverse linear transformations acting on the surface of different materials. From this view, we unify all data augmentation strategies for cross-spectral re-identification as mimicking such local linear transformations and categorize them into moderate transformation and radical transformation. By extending the observation, we propose a Random Linear Enhancement (RLE) strategy which includes Moderate Random Linear Enhancement (MRLE) and Radical Random Linear Enhancement (RRLE) to push the boundaries of both types of transformation. Moderate Random Linear Enhancement is designed to provide diverse image transformations that satisfy the original linear correlations under constrained conditions, whereas Radical Random Linear Enhancement seeks to generate local linear transformations directly without relying on external information. The experimental results not only demonstrate the superiority and effectiveness of RLE but also confirm its great potential as a general-purpose data augmentation for cross-spectral re-identification.

3D Referring Expression Segmentation (3D-RES) aims to segment 3D objects by correlating referring expressions with point clouds. However, traditional approaches frequently encounter issues like over-segmentation or mis-segmentation, due to insufficient emphasis on spatial information of instances. In this paper, we introduce a Rule-Guided Spatial Awareness Network (RG-SAN) by utilizing solely the spatial information of the target instance for supervision. This approach enables the network to accurately depict the spatial relationships among all entities described in the text, thus enhancing the reasoning capabilities. The RG-SAN consists of the Text-driven Localization Module (TLM) and the Rule-guided Weak Supervision (RWS) strategy. The TLM initially locates all mentioned instances and iteratively refines their positional information. The RWS strategy, acknowledging that only target objects have supervised positional information, employs dependency tree rules to precisely guide the core instance’s positioning. Extensive testing on the ScanRefer benchmark has shown that RG-SAN not only establishes new performance benchmarks, with an mIoU increase of 5.1 points, but also exhibits significant improvements in robustness when processing descriptions with spatial ambiguity. All codes are available at https://github.com/sosppxo/RG-SAN.

In this work, we propose a training-free method to inject visual prompts into Multimodal Large Language Models (MLLMs) through learnable latent variable optimization. We observe that attention, as the core module of MLLMs, connects text prompt tokens and visual tokens, ultimately determining the final results. Our approach involves adjusting visual tokens from the MLP output during inference, controlling the attention response to ensure text prompt tokens attend to visual tokens in referring regions. We optimize a learnable latent variable based on an energy function, enhancing the strength of referring regions in the attention map. This enables detailed region description and reasoning without the need for substantial training costs or model retraining. Our method offers a promising direction for integrating referring abilities into MLLMs, and supports referring with box, mask, scribble and point. The results demonstrate that our method exhibits out-of-domain generalization and interpretability.

PSNR-oriented models are a critical class of super-resolution models with applications across various fields. However, these models tend to generate over-smoothed images, a problem that has been analyzed previously from the perspectives of models or loss functions, but without taking into account the impact of data properties. In this paper, we present a novel phenomenon that we term the center-oriented optimization (COO) problem, where a model's output converges towards the center point of similar high-resolution images, rather than towards the ground truth. We demonstrate that the strength of this problem is related to the uncertainty of data, which we quantify using entropy. We prove that as the entropy of high-resolution images increases, their center point will move further away from the clean image distribution, and the model will generate over-smoothed images. Implicitly optimizing the COO problem, perceptual-driven approaches such as perceptual loss, model structure optimization, or GAN-based methods can be viewed. We propose an explicit solution to the COO problem, called Detail Enhanced Contrastive Loss (DECLoss). DECLoss utilizes the clustering property of contrastive learning to directly reduce the variance of the potential high-resolution distribution and thereby decrease the entropy. We evaluate DECLoss on multiple super-resolution benchmarks and demonstrate that it improves the perceptual quality of PSNR-oriented models. Moreover, when applied to GAN-based methods, such as RaGAN, DECLoss helps to achieve state-of-the-art performance, such as 0.093 LPIPS with 24.51 PSNR on 4× downsampled Urban100, validating the effectiveness and generalization of our approach.

Transformers have shown remarkable performance, however, their architecture design is a time-consuming process that demands expertise and trial-and-error. Thus, it is worthwhile to investigate efficient methods for automatically searching high-performance Transformers via Transformer Architecture Search (TAS). In order to improve the search efficiency, training-free proxy based methods have been widely adopted in Neural Architecture Search (NAS). Whereas, these proxies have been found to be inadequate in generalizing well to Transformer search spaces, as confirmed by several studies and our own experiments. This paper presents an effective scheme for TAS called TRansformer Architecture search with ZerO-cost pRoxy guided evolution (T-Razor) that achieves exceptional efficiency. First, through theoretical analysis, we discover that the synaptic diversity of multi-head self-attention (MSA) and the saliency of multi-layer perceptron (MLP) are correlated with the performance of corresponding Transformers. The properties of synaptic diversity and synaptic saliency motivate us to introduce the ranks of synaptic diversity and saliency that denoted as DSS++ for evaluating and ranking Transformers. DSS++ incorporates correlation information among sampled Transformers to provide unified scores for both synaptic diversity and synaptic saliency. We then propose a block-wise evolution search guided by DSS++ to find optimal Transformers. DSS++ determines the positions for mutation and crossover, enhancing the exploration ability. Experimental results demonstrate that our T-Razor performs competitively against the state-of-the-art manually or automatically designed Transformer architectures across four popular Transformer search spaces. Significantly, T-Razor improves the searching efficiency across different Transformer search spaces, e.g., reducing required GPU days from more than 24 to less than 0.4 and outperforming existing zero-cost approaches. We also apply T-Razor to the BERT search space and find that the searched Transformers achieve competitive GLUE results on several Neural Language Processing (NLP) datasets. This work provides insights into training-free TAS, revealing the usefulness of evaluating Transformers based on the properties of their different blocks.

Transformer has achieved impressive progress in visual tracking due to their capability of global modeling, which enables them to learn low-frequency features(i.e., high-level semantic information). However, it seems to overlook the high-frequency features(i.e., low-level texture and edge information) which are crucial to identify different intra-class object instances in the tracking task. To address this issue, we propose a transformer based tracker via frequency fusion perspective that investigated whether high-frequency and low-frequency features can be effectively combined to achieve robust tracking. Specifically, we design a simple yet effective two-stage fusion strategy and use an appropriate frequency fusion strategy in tracking process of each stage so as to make full use of frequency domain information. In the feature extraction stage, we use wavelet decomposition of high-frequency subbands to solve the performance loss caused by the transformer’s catastrophic forgetting of high-frequency information. In the prediction head stage, we use a variety of wavelet decomposition subbands to model the multi-frequency information. The two-stage fusion strategy makes our model extract more balanced and beneficial multi-frequency information, enabling it to effectively capture target texture information and local edge information while also being sensitive to global information. Extensive experiments on six challenging benchmarks (i.e., LaSOT$_{ext}$, UAV123, TNL2K, LaSOT, TrackingNet, and GOT-10k) demonstrates the superior performance of our tracker.

Face swapping aims to transfer the identity of a source face to a target face image while preserving the target attributes (e.g., facial expression, head pose, illumination, and background). Most existing methods use a face recognition model to extract global features from the source face and directly fuse them with the target to generate a swapping result. However, identity-irrelevant attributes (e.g., hairstyle and facial appearances) contribute a lot to the recognition task, and thus swapping this task-specific feature inevitably interfuses source attributes with target ones. In this paper, we propose an identity-aware variational autoencoder (ID-VAE) based face swapping framework, dubbed VAFSwap, which learns disentangled identity and attribute representations for high-fidelity face swapping. In particular, we overcome the unpaired training barrier of VAE and impose a proxy identity on the latent space by exploiting the weak supervision from an auxiliary image set whose identity is averaged from multiple collected face images. To explicitly guide the identity fusion, we further devise an identity-associated matrix that corresponds different face regions with their identity representations to perform identity-related feature interactions. Finally, we incorporate spatial dimensions into the latent space and exploit the generative priors of a pre-trained face generator, allowing the effective elimination of noticeable swapping artifacts. Extensive experiments on the FaceForensics++ and CelebA-HQ datasets demonstrate that our method outperforms the state-of-the-art significantly.

Most of the existing bounding box-based trackers rely on a classification subnetwork and a regression subnetwork to predict the location and scale of the bounding box. They learn the classification subnetwork by processing each sample individually and applying the suggested classification confidence to produce the final prediction. They typically involve heuristic positive sample configurations, which inevitably introduce mislabelled training samples and therefore deteriorate their tracking performance. Moreover, the parallel prediction of the bounding box position and scale may lead to misalignment of classification and regression. To address these issues,we propose a simple yet effective soft constraint-based tracking framework without positive samples (named SoftCT). SoftCT adaptively senses the target’s pixel position through a soft constraint mechanism, which eliminates potential performance gaps caused by artificially marking the target’s pixel position. In addition, SoftCT computes the state of the bounding box by aggregating such positional information, thereby allowing the tracker to avoid misalignment in classification and regression due to uninformed communication. Specifically, SoftCT directly senses the position of the target pixel and fuses this information into the bounding box prediction, rather than requiring explicit annotation or regression of the target pixel. Extensive experiments on six tracking benchmarks including GOT-10k, TrackingNet, LaSOT, UAV123, LaSOText and TNL2K demonstrate that our tracker achieves state-of-the-art performance, confirming its effectiveness and efficiency.

In this paper, we present a simple, flexible and effective vision-language (VL) tracking pipeline, termed MMTrack, which casts VL tracking as a token generation task. Traditional paradigms address VL tracking task indirectly with sophisticated prior designs, making them over-specialize on the features of specific architectures or mechanisms. In contrast, our proposed framework serializes language description and bounding box into a sequence of discrete tokens. In this new design paradigm, all token queries are required to perceive the desired target and directly predict spatial coordinates of the target in an auto-regressive manner. The design without other prior modules avoids multiple sub-tasks learning and hand-designed loss functions, significantly reducing the complexity of VL tracking modeling and allowing our tracker to use a simple cross-entropy loss as unified optimization objective for VL tracking task. Extensive experiments on TNL2K, LaSOT, LaSOT$_{\mathrm{ext}}$and OTB99-Lang benchmarks show that our approach achieves promising results, compared to other state-of-the-arts.

Recent works have indicated that deep neural networks (DNNs) are vulnerable to adversarial attacks, wherein an attacker perturbs an input example with human-imperceptible noise that can easily fool the DNNs, resulting in incorrect predictions. This severely limits the application of deep learning in security-critical scenarios, such as face authentication. Adversarial training (AT) is one of the most practical approaches to strengthening the robustness of DNNs. However, existing AT-based methods treat each training sample independently, thereby ignoring the underlying topological structure in the training data. To this end, in this paper, we take full advantage of the topology information and introduce a Topology Aligning Adversarial Training (TAAT) algorithm. TAAT aims to encourage the trained model to maintain consistency in the topological structure within the feature space of both natural and adversarial examples. To ensure the stability and efficiency of topology alignment, we further introduce a novel Knowledge-Guided (KG) training scheme. This scheme explicitly aligns local logit outputs with global topological structures, leveraging a robust auxiliary model to enhance the target model’s performance. To verify the effectiveness of the proposed method, we conduct extensive experiments on popular benchmark datasets (e.g., CIFAR and ImageNet) and evaluate the robustness against state-of-the-art adversarial attacks (e.g., PGD-attack and AutoAttack). The experimental results demonstrate that the proposed method has superior robustness over the previous state-of-the-art methods. Our code and pre-trained models are available at https://github.com/SkyKuang/TAAT.

Depth information opens up new opportunities for video object segmentation (VOS) to be more accurate and robust in complex scenes. However, the RGBD VOS task is largely unexplored due to the expensive collection of RGBD data and time-consuming annotation of segmentation. In this work, we first introduce a new benchmark for RGBD VOS, named DepthVOS, which contains 350 videos (over 55k frames in total) annotated with masks and bounding boxes. We futher propose a novel, strong baseline model - Fused Color-Depth Network (FusedCDNet), which can be trained solely under the supervision of bounding boxes, while being used to generate masks with a bounding box guideline only in the first frame. Thereby, the model possesses three major advantages: a weakly-supervised training strategy to overcome the high-cost annotation, a cross-modal fusion module to handle complex scenes, and weakly-supervised inference to promote ease of use. Extensive experiments demonstrate that our proposed method performs on par with top fully-supervised algorithms. We will open-source our project on https://github.com/yjybuaa/depthvos/ to facilitate the development of RGBD VOS.

In an effort to improve the efficiency and scalability of single-image super-resolution (SISR) applications, we introduce AnySR, to rebuild existing arbitrary-scale SR methods into any-scale, any-resource implementation. As a contrast to off-the-shelf methods that solve SR tasks across various scales with the same computing costs, our AnySR innovates in: 1) building arbitrary-scale tasks as any-resource implementation, reducing resource requirements for smaller scales without additional parameters; 2) enhancing any-scale performance in a feature-interweaving fashion, inserting scale pairs into features at regular intervals and ensuring correct feature/scale processing. The efficacy of our AnySR is fully demonstrated by rebuilding most existing arbitrary-scale SISR methods and validating on five popular SISR test datasets. The results show that our AnySR implements SISR tasks in a computing-more-efficient fashion, and performs on par with existing arbitrary-scale SISR methods. For the first time, we realize SISR tasks as not only any-scale in literature, but also as any-resource. Our code is available at https://github.com/CrispyFeSo4/AnySR.

Referring image segmentation aims to segment objects that are described by natural language expressions. Although remarkable advancements have been made to align natural language expressions with visual representations for better performance, the interaction between image-level and text-level information is still not formulated properly. Most of the previous works focus on building correlations between vision and language, ignoring the variety of objects. The target objects with unique appearances may not be correctly located or completely segmented. In this article, we propose a novel Bilateral Knowledge Interaction Network, termed BKINet, which reformulates the image-text interaction in a bilateral manner to adapt concrete knowledge of the target object in the image. BKINet contains two key components: a knowledge learning module (KLM) and a knowledge applying module (KAM). In the KLM, the abstract knowledge from text features is replenished with concrete knowledge from visual features to adapt to the target objects in the input images, which generates the knowledge interaction kernels (KI kernels) containing abundant referring information. With the referring information of KI kernels, the KAM is designed to highlight the most relevant visual features for predicting the accurate segmentation mask. Extensive experiments on three widely-used datasets,i.e.RefCOCO, RefCOCO+, and G-ref, demonstrate the superiority of BKINet over the state-of-the-art.

The task of Human-Object Interaction (HOI) detection is to detect humans and their interactions with surrounding objects, where transformer-based methods show dominant advances currently. However, these methods ignore the relationship among humans, objects, and interactions: 1) human features are more contributive than object ones to interaction prediction; 2) interactive information disturbs the detection of objects but helps human detection. In this article, we propose aHuman and Object Disentangling Network(HODN) to model the HOI relationships explicitly, where humans and objects are first detected by two disentangling decoders independently and then processed by an interaction decoder. Considering that human features are more contributive to interaction, we propose aHuman-Guide Linkingmethod to make sure the interaction decoder focuses on the human-centric regions with human features as the positional embeddings. To handle the opposite influences of interactions on humans and objects, we propose aStop-Gradient Mechanismto stop interaction gradients from optimizing the object detection but to allow them to optimize the human detection. Our proposed method achieves competitive performance on both the V-COCO and the HICO-Det datasets. It can be combined with existing methods easily for state-of-the-art results.

One-stage Referring Expression Comprehension (REC) is a task that requires accurate alignment between text descriptions and visual content. In recent years, numerous efforts have been devoted to cross-modal learning for REC, while the influence of other factors in this task still lacks a systematic study. To fill this gap, we conduct an empirical study in this article. Concretely, we ablate 42 candidate designs/settings based on a common REC framework, and these candidates cover the entire process of one-stage REC from network design to model training. Afterwards, we conduct over 100 experimental trials on three REC benchmark datasets. The extensive experimental results reveal the key factors that affect REC performance in addition to multi-modal fusion, e.g., multi-scale features and data augmentation. Based on these findings, we further propose a simple yet strong model called SimREC, which achieves new state-of-the-art performance on these benchmarks. In addition to these progresses, we also find that with much less training overhead and parameters, SimREC can achieve better performance than a set of large-scale pre-trained models, e.g., UNITER and VILLA, portraying the special role of REC in existing V&L research.

Vision Transformers (ViT) have made many breakthroughs in computer vision tasks. However, considerable redundancy arises in the spatial dimension of an input image, leading to massive computational costs. Therefore, We propose a coarse-to-fine vision transformer (CF-ViT) to relieve computational burden while retaining performance in this paper. Our proposed CF-ViT is motivated by two important observations in modern ViT models: (1) The coarse-grained patch splitting can locate informative regions of an input image. (2) Most images can be well recognized by a ViT model in a small-length token sequence. Therefore, our CF-ViT implements network inference in a two-stage manner. At coarse inference stage, an input image is split into a small-length patch sequence for a computationally economical classification. If not well recognized, the informative patches are identified and further re-split in a fine-grained granularity. Extensive experiments demonstrate the efficacy of our CF-ViT. For example, without any compromise on performance, CF-ViT reduces 53% FLOPs of LV-ViT, and also achieves 2.01x throughput. Code of this project is at https://github.com/ChenMnZ/CF-V

To bridge the ever-increasing gap between deep neural networks' complexity and hardware capability, network quantization has attracted more and more research attention. The latest trend of mixed precision quantization takes advantage of hardware's multiple bit-width arithmetic operations to unleash the full potential of network quantization. However, existing approaches rely heavily on an extremely time-consuming search process and various relaxations when seeking the optimal bit configuration. To address this issue, we propose to optimize a proxy metric of network orthogonality that can be efficiently solved with linear programming, which proves to be highly correlated with quantized model accuracy and bit-width. Our approach significantly reduces the search time and the required data amount by orders of magnitude, but without a compromise on quantization accuracy. Specifically, we achieve 72.08% Top-1 accuracy on ResNet-18 with 6.7Mb parameters, which does not require any searching iterations. Given the high efficiency and low data dependency of our algorithm, we use it for the post-training quantization, which achieves 71.27% Top-1 accuracy on MobileNetV2 with only 1.5Mb parameters.

In this paper, we propose YOSO, a real-time panoptic segmentation framework. YOSO predicts masks via dynamic convolutions between panoptic kernels and image feature maps, in which you only need to segment once for both instance and semantic segmentation tasks. To reduce the computational overhead, we design a feature pyramid aggregator for the feature map extraction, and a separable dynamic decoder for the panoptic kernel generation. The aggregator re-parameterizes interpolation-first modules in a convolution-first way, which significantly speeds up the pipeline without any additional costs. The decoder performs multi-head cross-attention via separable dynamic convolution for better efficiency and accuracy. To the best of our knowledge, YOSO is the first real-time panoptic segmentation framework that delivers competitive performance compared to state-of-the-art models. Specifically, YOSO achieves 46.4 PQ, 45.6 FPS on COCO; 52.5 PQ, 22.6 FPS on Cityscapes; 38.0 PQ, 35.4 FPS on ADE20K; and 34.1 PQ, 7.1 FPS on Mapillary Vistas. Code is available at https://github.com/hujiecpp/YOSO.

Despite excellent performance in image generation, Generative Adversarial Networks (GANs) are notorious for its requirements of enormous storage and intensive computation. As an awesome “performance maker”, knowledge distillation is demonstrated to be particularly efficacious in exploring low-priced GANs. In this paper, we investigate the irreplaceability of teacher discriminator and present an inventive discriminator-cooperated distillation, abbreviated as DCD, towards refining better feature maps from the generator. In contrast to conventional pixel-to-pixel match methods in feature map distillation, our DCD utilizes teacher discriminator as a transformation to drive intermediate results of the student generator to be perceptually close to corresponding outputs of the teacher generator. Furthermore, in order to mitigate mode collapse in GAN compression, we construct a collaborative adversarial training paradigm where the teacher discriminator is from scratch established to co-train with student generator in company with our DCD. Our DCD shows superior results compared with existing GAN compression methods. For instance, after reducing over$40\times MACs$and$80\times$parameters of CycleGAN, we well decrease FID metric from 61.53 to 48.24 while the current SoTA method merely has 51.92. This work's source code has been made accessible at https://github.com/poopit/DCD-official.

Building a universal Video-Language model for solving various video understanding tasks (e.g., text-video retrieval, video question answering) is an open challenge to the machine learning field. Towards this goal, most recent works build the model by stacking uni-modal and cross-modal feature encoders and train it with pair-wise contrastive pre-text tasks. Though offering attractive generality, the resulted models have to compromise between efficiency and performance. They mostly adopt different architectures to deal with different downstream tasks. We find this is because the pair-wise training cannot well align and fuse features from different modalities. We then introduce Clover-a Correlated Video-Language pre-training method-towards a universal Video-Language modelfor solving multiple video understanding tasks with neither performance nor efficiency compromise. It improves cross-modal feature alignment and fusion via a novel tri-modal alignment pre-training task. Additionally, we propose to enhance the tri-modal alignment via incorporating learning from semantic masked samples and a new pair-wise ranking loss. Clover establishes new state-of-the-arts on multiple downstream tasks, including three retrieval tasks for both zero-shot and fine-tuning settings, and eight video question answering tasks. Codes and pre-trained models will be released at https://github.com/LeeYN-43/Clover.

Referring Expression Comprehension (REC) is a task of grounding the referent based on an expression, and its development is greatly limited by expensive instance-level annotations. Most existing weakly supervised methods are built based on two-stage detection networks, which are computationally expensive. In this paper, we resort to the efficient one-stage detector and propose a novel weakly supervised model called RefCLIP.Specifically, RefCLIP redefines weakly supervised REC as an anchor-text matching problem, which can avoid the complex post-processing in existing methods. To achieve weakly supervised learning, we introduce anchor-based contrastive loss to optimize Re-fCLlP via numerous anchor-text pairs. Based on RefCLIP, we further propose the first model-agnostic weakly supervised training scheme for existing REC models, where RefCLIP acts as a mature teacher to generate pseudo-labels for teaching common REC models. With our careful designs, this scheme can even help existing REC models achieve better weakly supervised performance than RefCLIP, e.g., TransVg and SimREC. To validate our approaches, we conduct extensive experiments on four REC benchmarks, i.e., RefCOCO, RefCOCO+, RefCOCOg and ReferItGame. Experimental results not only report our significant performance gains over existing weakly supervised models, e.g., +24.87% on RefCOCO, but also show the 5x faster inference speed. Project: https://RefCLIP.github.io.

Recent advances in 3D point cloud analysis bring a diverse set of network architectures to the field. However, the lack of a unified framework to interpret those networks makes any systematic comparison, contrast, or analysis challenging, and practically limits healthy development of the field. In this paper, we take the initiative to explore and propose a unified framework called PointMeta, to which the popular 3D point cloud analysis approaches could fit. This brings three benefits. First, it allows us to compare different approaches in a fair manner, and use quick experiments to verify any empirical observations or assumptions summarized from the comparison. Second, the big picture brought by PointMeta enables us to think across different components, and revisit common beliefs and key design decisions made by the popular approaches. Third, based on the learnings from the previous two analyses, by doing simple tweaks on the existing approaches, we are able to derive a basic building block, termed PointMetaBase. It shows very strong performance in efficiency and effectiveness through extensive experiments on challenging benchmarks, and thus verifies the necessity and benefits of high-level interpretation, contrast, and comparison like PointMeta. In particular, PointMetaBase surpasses the previous state-of-the-art method by 0.7%/1.4/%2.1% mIoU with only 2%/11%/13% of the computation cost on the S3DIS datasets. The code and models are available at https://github.com/linhaojia13/PointMetaBase.

Post-training quantization (PTQ) is widely regarded as one of the most efficient compression methods practically, benefitting from its data privacy and low computation costs. We argue that an overlooked problem of oscillation is in the PTQ methods. In this paper, we take the initiative to explore and present a theoretical proof to explain why such a problem is essential in PTQ. And then, we try to solve this problem by introducing a principled and generalized frame-work theoretically. In particular, we first formulate the oscillation in PTQ and prove the problem is caused by the difference in module capacity. To this end, we define the module capacity (ModCap) under data-dependent and data-free scenarios, where the differentials between adjacent modules are used to measure the degree of oscillation. The problem is then solved by selecting top-k differentials, in which the corresponding modules are jointly optimized and quantized. Extensive experiments demonstrate that our method successfully reduces the performance drop and is generalized to different neural networks and PTQ methods. For example, with 2/4 bit ResNet-50 quantization, our method surpasses the previous state-of-the-art method by 1.9%. It becomes more significant on small model quantization, e.g. surpasses BRECQ method by 6.61% on MobileNetV2 × 0.5.

Referring expression comprehension (REC) often requires a large number of instance-level annotations for fully supervised learning, which are laborious and expensive. In this paper, we present the first attempt of semi-supervised learning for REC and propose a strong baseline method called RefTeacher. Inspired by the recent progress in computer vision, RefTeacher adopts a teacher-student learning paradigm, where the teacher REC network predicts pseudolabels for optimizing the student one. This paradigm allows REC models to exploit massive unlabeled data based on a small fraction of labeled. In particular, we also identify two key challenges in semi-supervised REC, namely, sparse supervision signals and worse pseudo-label noise. To address these issues, we equip RefTeacher with two novel designs called Attention-based Imitation Learning (AIL) and Adaptive Pseudo-label Weighting (APW). AIL can help the student network imitate the recognition behaviors of the teacher, thereby obtaining sufficient supervision signals. APW can help the model adaptively adjust the contributions of pseudo-labels with varying qualities, thus avoiding confirmation bias. To validate RefTeacher, we conduct extensive experiments on three REC benchmark datasets. Experimental results show that RefTeacher obtains obvious gains over the fully supervised methods. More importantly, using only 10% labeled data, our approach allows the model to achieve near 100% fully supervised performance, e.g., only −2.78% on RefCoco. Project: https://refteacher.github.io/.

In the field of human pose estimation, regression-based methods have been dominated in terms of speed, while heatmap-based methods are far ahead in terms of performance. How to take advantage of both schemes remains a challenging problem. In this paper, we propose a novel human pose estimation framework termed DistilPose, which bridges the gaps between heatmap-based and regression-based methods. Specifically, DistilPose maximizes the transfer of knowledge from the teacher model (heatmap-based) to the student model (regression-based) through Token-distilling Encoder (TDE) and Simulated Heatmaps. TDE aligns the feature spaces of heatmap-based and regression-based models by introducing tokenization, while Simulated Heatmaps transfer explicit guidance (distribution and confidence) from teacher heatmaps into student models. Extensive experiments show that the proposed DistilPose can significantly improve the performance of the regression-based models while maintaining efficiency. Specifically, on the MSCOCO validation dataset, DistilPose-S obtains 71.6% mAP with 5.36M parameters, 2.38 GFLOPs, and 40.2 FPS, which saves 12.95×, 7.16× computational cost and is 4.9× faster than its teacher model with only 0.9 points performance drop. Furthermore, DistilPose-L obtains 74.4% mAP on MSCOCO validation dataset, achieving a new state-of-the-art among predominant regression-based models. Code will be available at https://github.com/yshMars/DistilPose.

Recently, deep learning-based facial landmark detection has achieved significant improvement. However, the semantic ambiguity problem degrades detection performance. Specifically, the semantic ambiguity causes inconsistent annotation and negatively affects the model's convergence, leading to worse accuracy and instability prediction. To solve this problem, we propose a Self-adapTive Ambiguity Reduction (STAR) loss by exploiting the properties of se-mantic ambiguity. We find that semantic ambiguity results in the anisotropic predicted distribution, which inspires us to use predicted distribution to represent semantic ambiguity. Based on this, we design the STAR loss that measures the anisotropism of the predicted distribution. Compared with the standard regression loss, STAR loss is encouraged to be small when the predicted distribution is anisotropic and thus adaptively mitigates the impact of semantic ambiguity. Moreover, we propose two kinds of eigen-value restriction methods that could avoid both distribution's abnormal change and the model's premature convergence. Finally, the comprehensive experiments demonstrate that STAR loss outperforms the state-of-the-art methods on three benchmarks, i.e., COFW, 300W, and WFLW, with negligible computation overhead. Code is at https://github.com/ZhenglinZhou/STAR

Most existing neural network pruning methods hand-crafted their importance criteria and structures to prune. This constructs heavy and unintended dependencies on heuristics and expert experience for both the objective and the parameters of the pruning approach. In this paper, we try to solve this problem by introducing a principled and unified framework based on Information Bottleneck (IB) theory, which further guides us to an automatic pruning approach. Specifically, we first formulate the channel pruning problem from an IB perspective, and then implement the IB principle by solving a Hilbert-Schmidt Independence Criterion (HSIC) Lasso problem under certain conditions. Based on the theoretical guidance, we then provide an automatic pruning scheme by searching for global penalty coefficients. Verified by extensive experiments, our method yields state-of-the-art performance on various benchmark networks and datasets. For example, with VGG-16, we achieve a 60%-FLOPs reduction by removing 76% of the parameters, with an improvement of 0.40% in top-1 accuracy on CIFAR-10. With ResNet-50, we achieve a 56%-FLOPs reduction by removing 50% of the parameters, with a small loss of 0.08% in the top-1 accuracy on ImageNet. The code is available at https://github.com/sunggo/APIB.

Weakly supervised object localization (WSOL) aims to localize objects based on only image-level labels as supervision. Recently, transformers have been introduced into WSOL, yielding impressive results. The self-attention mechanism and multilayer perceptron structure in transformers preserve long-range feature dependency, facilitating complete localization of the full object extent. However, current transformer-based methods predict bounding boxes using category-agnostic attention maps, which may lead to confused and noisy object localization. To address this issue, we propose a novel Category-aware Allocation TRansformer (CATR) that learns category-aware representations for specific objects and produces corresponding category-aware attention maps for object localization. First, we introduce a Category-aware Stimulation Module (CSM) to induce learnable category biases for self-attention maps, providing auxiliary supervision to guide the learning of more effective transformer representations. Second, we design an Object Constraint Module (OCM) to refine the object regions for the category-aware attention maps in a self-supervised manner. Extensive experiments on the CUB-200-2011 and ILSVRC datasets demonstrate that the proposed CATR achieves significant and consistent performance improvements over competing approaches.

CutMix is a vital augmentation strategy that determines the performance and generalization ability of vision transformers (ViTs). However, the inconsistency between the mixed images and the corresponding labels harms its efficacy. Existing CutMix variants tackle this problem by generating more consistent mixed images or more precise mixed labels, but inevitably introduce heavy training overhead or require extra information, undermining ease of use. To this end, we propose an novel and effective Self-Motivated image Mixing method (SMMix), which motivates both image and label enhancement by the model under training itself. Specifically, we propose a max-min attention region mixing approach that enriches the attention-focused objects in the mixed images. Then, we introduce a fine-grained label assignment technique that co-trains the output tokens of mixed images with fine-grained supervision. Moreover, we devise a novel feature consistency constraint to align features from mixed and unmixed images. Due to the subtle designs of the self-motivated paradigm, our SMMix is significant in its smaller training overhead and better performance than other CutMix variants. In particular, SMMix improves the accuracy of DeiT-T/S/B, CaiT-XXS-24/36, and PVT-T/S/M/L by more than +1% on ImageNet-1k. The generalization capability of our method is also demonstrated on downstream tasks and out-of-distribution datasets. Our project is available at https://github.com/ChenMnZ/SMMix.

Token compression aims to speed up large-scale vision transformers (e.g. ViTs) by pruning (dropping) or merging tokens. It is an important but challenging task. Although recent advanced approaches achieved great success, they need to carefully handcraft a compression rate (i.e. number of tokens to remove), which is tedious and leads to sub-optimal performance. To tackle this problem, we propose Differentiable Compression Rate (DiffRate), a novel token compression method that has several appealing properties prior arts do not have. First, DiffRate enables propagating the loss function’s gradient onto the compression ratio, which is considered as a non-differentiable hyperparameter in previous work. In this case, different layers can automatically learn different compression rates layer-wisely without extra overhead. Second, token pruning and merging can be naturally performed simultaneously in DiffRate, while they were isolated in previous works. Third, extensive experiments demonstrate that DiffRate achieves state-of-the-art performance. For example, by applying the learned layer-wise compression rates to an off-the-shelf ViT-H (MAE) model, we achieve a 40% FLOPs reduction and a 1.5× throughput improvement, with a minor accuracy drop of 0.16% on ImageNet without fine-tuning, even outperforming previous methods with fine-tuning. Codes and models are available at https://github.com/OpenGVLab/DiffRate.

Pseudo-labeling is significant for semi-supervised instance segmentation, which generates instance masks and classes from unannotated images for subsequent training. However, in existing pipelines, pseudo-labels that contain valuable information may be directly filtered out due to mismatches in class and mask quality. To address this issue, we propose a novel framework, called pseudo-label aligning instance segmentation (PAIS), in this paper. In PAIS, we devise a dynamic aligning loss (DALoss) that adjusts the weights of semi-supervised loss terms with varying class and mask score pairs. Through extensive experiments conducted on the COCO and Cityscapes datasets, we demonstrate that PAIS is a promising framework for semi-supervised instance segmentation, particularly in cases where labeled data is severely limited. Notably, with just 1% labeled data, PAIS achieves 21.2 mAP (based on MaskRCNN) and 19.9 mAP (based on K-Net) on the COCO dataset, outperforming the current state-of-the-art model, i.e., NoisyBoundary with 7.7 mAP, by a margin of over 12 points. Code is available at: https://github.com/hujiecpp/PAIS.

Interactive image segmentation enables annotators to efficiently perform pixel-level annotation for segmentation tasks. However, the existing interactive segmentation pipeline suffers from inefficient computations of interactive models because of the following two issues. First, annotators’ later click is based on models’ feedback of annotators’ former click. This serial interaction is unable to utilize model’s parallelism capabilities. Second, in each interaction step, the model handles the invariant image along with the sparse variable clicks, resulting in a process that’s highly repetitive and redundant. For efficient computations, we propose a method named InterFormer that follows a new pipeline to address these issues. In-terFormer extracts and preprocesses the computationally time-consuming part i.e. image processing from the existing process. Specifically, InterFormer employs a large vision transformer (ViT) on high-performance devices to prepro-cess images in parallel, and then uses a lightweight module called interactive multi-head self attention (I-MSA) for interactive segmentation. Furthermore, the I-MSA module’s deployment on low-power devices extends the practical application of interactive segmentation. The I-MSA module utilizes the preprocessed features to efficiently response to the annotator inputs in real-time. The experiments on several datasets demonstrate the effectiveness of Inter-Former, which outperforms previous interactive segmentation models in terms of computational efficiency and segmentation quality, achieve real-time high-quality interactive segmentation on CPU-only devices. The code is available at https://github.com/YouHuang67/InterFormer.

Diffusion models are emerging expressive generative models, in which a large number of time steps (inference steps) are required for a single image generation. To accelerate such tedious process, reducing steps uniformly is considered as an undisputed principle of diffusion models. We consider that such a uniform assumption is not the optimal solution in practice; i.e., we can find different optimal time steps for different models. Therefore, we propose to search the optimal time steps sequence and compressed model architecture in a unified framework to achieve effective image generation for diffusion models without any further training. Specifically, we first design a unified search space that consists of all possible time steps and various architectures. Then, a two stage evolutionary algorithm is introduced to find the optimal solution in the designed search space. To further accelerate the search process, we employ FID score between generated and real samples to estimate the performance of the sampled examples. As a result, the proposed method is (i).training-free, obtaining the optimal time steps and model architecture without any training process; (ii). orthogonal to most advanced diffusion samplers and can be integrated to gain better sample quality. (iii). generalized, where the searched time steps and architectures can be directly applied on different diffusion models with the same guidance scale. Experimental results show that our method achieves excellent performance by using only a few time steps, e.g. 17.86 FID score on ImageNet 64 × 64 with only four steps, compared to 138.66 with DDIM.

Text-driven 3D stylization is a complex and crucial task in the fields of computer vision (CV) and computer graphics (CG), aimed at transforming a bare mesh to fit a tar-get text. Prior methods adopt text-independent multilayer perceptrons (MLPs) to predict the attributes of the target mesh with the supervision of CLIP loss. However, such text-independent architecture lacks textual guidance during predicting attributes, thus leading to unsatisfactory stylization and slow convergence. To address these limitations, we present X-Mesh, an innovative text-driven 3D stylization framework that incorporates a novel Text-guided Dynamic Attention Module (TDAM). The TDAM dynamically integrates the guidance of the target text by utilizing text-relevant spatial and channel-wise attentions during vertex feature extraction, resulting in more accurate attribute prediction and faster convergence speed. Furthermore, existing works lack standard benchmarks and automated metrics for evaluation, often relying on subjective and non-reproducible user studies to assess the quality of stylized 3D assets. To overcome this limitation, we introduce a new standard text-mesh benchmark, namely MIT-30, and two automated metrics, which will enable future research to achieve fair and objective comparisons. Our extensive qualitative and quantitative experiments demonstrate that X-Mesh outperforms previous state-of-the-art methods. Our codes and results are available at our project webpage: https://xmu-xiaoma666.github.io/Projects/X-Mesh/

We focus on addressing the dense backward propagation issue for training efficiency of N:M fine-grained sparsity that preserves at most N out of M consecutive weights and achieves practical speedups supported by the N:M sparse tensor core. Therefore, we present a novel method of Bi-directional Masks (Bi-Mask) with its two central innovations in: 1) Separate sparse masks in the two directions of forward and backward propagation to obtain training acceleration. It disentangles the forward and backward weight sparsity and overcomes the very dense gradient computation. 2) An efficient weight row permutation method to maintain performance. It picks up the permutation candidate with the most eligible N:M weight blocks in the backward to minimize the gradient gap between traditional unidirectional masks and our bi-directional masks. Compared with existing uni-directional scenario that applies a transposable mask and enables backward acceleration, our Bi-Mask is experimentally demonstrated to be more superior in performance. Also, our Bi-Mask performs on par with or even better than methods that fail to achieve backward acceleration. Project of this paper is available at https://github.com/zyxxmu/Bi-Mask.

Object placement aims to insert a foreground object into a background image with a suitable location and size to create a natural composition. To predict a diverse distribution of placements, existing methods usually establish a one-to-one mapping from random vectors to the placements. However, these random vectors are not interpretable, which prevents users from interacting with the object placement process. To address this problem, we propose an Interactive Object Placement method with Reinforcement Learning, dubbed IOPRE, to make sequential decisions for producing a reasonable placement given an initial location and size of the foreground. We first design a novel action space to flexibly and stably adjust the location and size of the foreground while preserving its aspect ratio. Then, we propose a multi-factor state representation learning method, which integrates composition image features and sinusoidal positional embeddings of the foreground to make decisions for selecting actions. Finally, we design a hybrid reward function that combines placement assessment and the number of steps to ensure that the agent learns to place objects in the most visually pleasing and semantically appropriate location. Experimental results on the OPA dataset demonstrate that the proposed method achieves state-of-the-art performance in terms of plausibility and diversity.

Refactoring is an indispensable practice of improving the quality and maintainability of source code in software evolution. Rename refactoring is the most frequently performed refactoring that suggests a new name for an identifier to enhance readability when the identifier is poorly named. However, most existing works only identify renaming activities between two versions of source code, while few works express concern about how to suggest a new name. In this paper, we study automatic rename refactoring on variable names, which is considered more challenging than other rename refactoring activities. We first point out the connections between rename refactoring and various prevalent learning paradigms and the difference between rename refactoring and general text generation in natural language processing. Based on our observations, we propose RefBERT, a two-stage pre-trained framework for rename refactoring on variable names. RefBERT first predicts the number of sub-tokens in the new name and then generates sub-tokens accordingly. Several techniques, including constrained masked language modeling, contrastive learning, and the bag-of-tokens loss, are incorporated into RefBERT to tailor it for automatic rename refactoring on variable names. Through extensive experiments on our constructed refactoring datasets, we show that the generated variable names of RefBERT are more accurate and meaningful than those produced by the existing method. Our implementation and data are available at https://github.com/KDEGroup/RefBERT.

Issue-commit links, as a type of software traceability links, play a vital role in various software development and maintenance tasks. However, they are typically deficient, as developers often forget or fail to create tags when making commits. Existing studies have deployed deep learning techniques, including pretrained models, to improve automatic issue-commit link recovery. Despite their promising performance, we argue that previous approaches have four main problems, hindering them from recovering links in large software projects. To overcome these problems, we propose an efficient and accurate pre-trained framework called EALink for issue-commit link recovery. EALink requires much fewer model parameters than existing pre-trained methods, bringing efficient training and recovery. Moreover, we design various techniques to improve the recovery accuracy of EALink. We construct a large-scale dataset and conduct extensive experiments to demonstrate the power of EALink. Results show that EALink outperforms the state-of-the-art methods by a large margin (15.23%-408.65%) on various evaluation metrics. Meanwhile, its training and inference overhead is orders of magnitude lower than existing methods. We provide our implementation and data at https://github.com/KDEGroup/EALink.

Synthesizing vivid human portraits is a research hot spot in image generation with a wide scope of applications. In addition to fidelity, generation controllability is another key factor that has long plagued its development. To address this issue, existing solutions usually adopt either textual or visual conditions for the target face synthesis, e.g., descriptions or segmentation masks, which still cannot fully control the generation due to the intrinsic shortages of each condition. In this paper, we propose to make use of both types of prior information to facilitate controllable face generation. In particular, we hope to produce coarse-grained information about faces based on the segmentation masks, such as face shapes and poses, and the text description is used to render detailed face attributes, e.g., face color, makeup and gender. More importantly, we hope that the generation can be easily controlled via interactively editing both types of information, making face generation more applicable to real-world applications. To accomplish this target, we propose a novel face generation model termed PixelFace+. In PixelFace+, both the text and mask are encoded as pixel-wise priors, based on which the pixel synthesis process is conducted to produce the expected portraits. Meanwhile, the loss objectives are also carefully designed to make sure that the generated faces are semantically aligned with both text and mask inputs. To validate the proposed PixelFace+, we conducted a comprehensive set of experiments on the widely recognized benchmark called MMCelebA. We not only quantitatively compare PixelFace+ with a bunch of newly proposed Text-to-Face(T2F) generation methods, but also give plenty of qualitative analyses. The experimental results demonstrate that PixelFace+ not only outperforms existing generation methods in both image quality and conditional matching but also shows a much superior controllability of face generation. More importantly, PixelFace+ presents a convenient and interactive way of face generation and manipulation via editing the text and mask inputs. Our SOURCE CODE and DEMO are given in our supplementary materials.

Human-Object Interaction (HOI) detection aims to understand the interactions between humans and objects, which plays a curtail role in high-level semantic understanding tasks. However, most works pursue designing better architectures to learn overall features more efficiently, while ignoring the long-tail nature of interaction-object pair categories. In this paper, we propose to alleviate the impact of such an unbalanced distribution via Virtual Image Leaning (VIL). Firstly, a novel label-to-image approach, Multiple Steps Image Creation (MUSIC), is proposed to create a high-quality dataset that has a consistent distribution with real images. In this stage, virtual images are generated based on prompts with specific characterizations and selected by multi-filtering processes. Secondly, we use both virtual and real images to train the model with the teacher-student framework. Considering the initial labels of some virtual images are inaccurate and inadequate, we devise an Adaptive Matching-and-Filtering (AMF) module to construct pseudo-labels. Our method is independent of the internal structure of HOI detectors, so it can be combined with off-the-shelf methods by training merely 10 additional epochs. With the assistance of our method, multiple methods obtain significant improvements, and new state-of-the-art results are achieved on two benchmarks.

Person re-identification (Re-ID) has been extensively investigated in recent years. However, many existing paradigms rely on holistic person regions for matching, disregarding the challenges posed by occlusions in real-world scenarios. Recent methods have explored occlusion augmentation or external semantic cues. Nevertheless, these approaches tend to be coarse-grained, discarding valuable semantic information in local regions when determining them as occlusions. In this paper, we propose a Fine-grained Occlusion Disentanglement Network (FODN) that can extract more information from limited person regions. Specifically, we propose a fine-grained occlusion augmentation scheme to generate diverse occlusion data and employ bilinear interpolation and downsampling strategies to obtain fine-grained occlusion labels. We then design an occlusion feature disentanglement Module that decouples norm and angle from features and supervises the occlusion-aware task using the aforementioned occlusion labeling and person re-identification tasks, respectively, resulting in more robust features. Additionally, we propose a dynamic local weight controller to balance the relative importance of various human body parts, thereby improving the model's ability to mine more effective local features from limited human body regions after occlusion removal. Comprehensive experiments on various person Re-ID benchmarks demonstrate the superiority of FODN over state-of-the-art methods.

Text-based person retrieval (TPR) is a challenging task that involves retrieving a specific individual based on a textual description. Despite considerable efforts to bridge the gap between vision and language, the significant differences between these modalities continue to pose a challenge. Previous methods have attempted to align text and image samples in a modal-shared space, but they face uncertainties in optimization directions due to the movable features of both modalities and the failure to account for one-to-many relationships of image-text pairs in TPR datasets. To address this issue, we propose an effective bi-directional one-to-many embedding paradigm that offers a clear optimization direction for each sample, thus mitigating the optimization problem. Additionally, this embedding scheme generates multiple features for each sample without introducing trainable parameters, making it easier to align with several positive samples. Based on this paradigm, we propose a novel Bi-directional one-to-many Embedding Alignment (Beat) model to address the TPR task. Our experimental results demonstrate that the proposed Beat model achieves state-of-the-art performance on three popular TPR datasets, including CUHK-PEDES (65.61 R@1), ICFG-PEDES (58.25 R@1), and RSTPReID (48.10 R@1). Furthermore, additional experiments on MS-COCO, CUB, and Flowers datasets further demonstrate the potential of Beat to be applied to other image-text retrieval tasks.

In recent years, 3D representation learning has turned to 2D vision-language pre-trained models to overcome data scarcity challenges. However, existing methods simply transfer 2D alignment strategies, aligning 3D representations with single-view 2D images and coarse-grained parent category text. These approaches introduce information degradation and insufficient synergy issues, leading to performance loss. Information degradation arises from overlooking the fact that a 3D representation should be equivalent to a series of multi-view images and more fine-grained subcategory text. Insufficient synergy neglects the idea that a robust 3D representation should align with the joint vision-language space, rather than independently aligning with each modality. In this paper, we propose a multi-view joint modality modeling approach, termed JM3D, to obtain a unified representation for point cloud, text, and image. Specifically, a novel Structured Multimodal Organizer (SMO) is proposed to address the information degradation issue, which introduces contiguous multi-view images and hierarchical text to enrich the representation of vision and language modalities. A Joint Multi-modal Alignment (JMA) is designed to tackle the insufficient synergy problem, which models the joint modality by incorporating language knowledge into the visual modality. Extensive experiments on ModelNet40 and ScanObjectNN demonstrate the effectiveness of our proposed method, JM3D, which achieves state-of-the-art performance in zero-shot 3D classification. JM3D outperforms ULIP by approximately 4.3% on PointMLP and achieves an improvement of up to 6.5% accuracy on PointNet++ in top-1 accuracy for zero-shot 3D classification on ModelNet40. The source code and trained models for all our experiments are publicly available at https://github.com/Mr-Neko/JM3D.

Despite considerable progress, the advancement of Panoptic Narrative Grounding (PNG) remains hindered by costly annotations. In this paper, we introduce a novel Semi-Supervised Panoptic Narrative Grounding (SS-PNG) learning scheme, capitalizing on a smaller set of labeled image-text pairs and a larger set of unlabeled pairs to achieve competitive performance. Unlike visual segmentation tasks, PNG involves one pixel belonging to multiple open-ended nouns. As a result, existing multi-class based semi-supervised segmentation frameworks cannot be directly applied to this task. To address this challenge, we first develop a novel SS-PNG Network (SS-PNG-NW) tailored to the SS-PNG setting. We thoroughly investigate strategies such as Burn-In and data augmentation to determine the optimal generic configuration for the SS-PNG-NW. Additionally, to tackle the issue of imbalanced pseudo-label quality, we propose a Quality-Based Loss Adjustment (QLA) approach to adjust the semi-supervised objective, resulting in an enhanced SS-PNG-NW+. Employing our proposed QLA, we improve BCE Loss and Dice loss at pixel and mask levels, respectively. We conduct extensive experiments on PNG datasets, with our SS-PNG-NW+ demonstrating promising results comparable to fully-supervised models across all data ratios. Remarkably, our SS-PNG-NW+ outperforms fully-supervised models with only 30% and 50% supervision data, exceeding their performance by 0.8% and 1.1% respectively. This highlights the effectiveness of our proposed SS-PNG-NW+ in overcoming the challenges posed by limited annotations and enhancing the applicability of PNG tasks. The source code is available at https://github.com/nini0919/SSPNG.

Previous studies have shown that optimizing the information bottleneck can significantly improve the robustness of deep neural networks. Our study closely examines the information bottleneck principle and proposes an Information Bottleneck Distillation approach. This specially designed, robust distillation technique utilizes prior knowledge obtained from a robust pre-trained model to boost information bottlenecks. Specifically, we propose two distillation strategies that align with the two optimization processes of the information bottleneck. Firstly, we use a robust soft-label distillation method to increase the mutual information between latent features and output prediction. Secondly, we introduce an adaptive feature distillation method that automatically transfers relevant knowledge from the teacher model to the student model, thereby reducing the mutual information between the input and latent features. We conduct extensive experiments to evaluate our approach's robustness against state-of-the-art adversarial attackers such as PGD-attack and AutoAttack. Our experimental results demonstrate the effectiveness of our approach in significantly improving adversarial robustness. Our code is available at https://github.com/SkyKuang/IBD.

Recently, growing interest has been aroused in extending the multimodal capability of large language models (LLMs), e.g., vision-language (VL) learning, which is regarded as the next milestone of artificial general intelligence. However, existing solutions are prohibitively expensive, which not only need to optimize excessive parameters, but also require another large-scale pre-training before VL instruction tuning. In this paper, we propose a novel and affordable solution for the effective VL adaption of LLMs, called Mixture-of-Modality Adaptation (MMA). Instead of using large neural networks to connect the image encoder and LLM, MMA adopts lightweight modules, i.e., adapters, to bridge the gap between LLMs and VL tasks, which also enables the joint optimization of the image and language models. Meanwhile, MMA is also equipped with a routing algorithm to help LLMs achieve an automatic shift between single- and multi-modal instructions without compromising their ability of natural language understanding. To validate MMA, we apply it to a recent LLM called LLaMA and term this formed large vision-language instructed model as LaVIN. To validate MMA and LaVIN, we conduct extensive experiments under two setups, namely multimodal science question answering and multimodal dialogue. The experimental results not only demonstrate the competitive performance and the superior training efficiency of LaVIN than existing multimodal LLMs, but also confirm its great potential as a general-purpose chatbot. More importantly, the actual expenditure of LaVIN is extremely cheap, e.g., only 1.4 training hours with 3.8M trainable parameters, greatly confirming the effectiveness of MMA. Our code is anonymously released at: https://anonymous.4open.science/r/LaVIN--1067.

Webly supervised learning has attracted increasing attention for its effectiveness in exploring publicly accessible data at scale without manual annotation. However, most existing methods of learning with web datasets are faced with challenges from label noise, and they have limited assumptions on clean samples under various noise. For instance, web images retrieved with queries of ”tiger cat“ (a cat species) and ”drumstick“ (a musical instrument) are almost dominated by images of tigers and chickens, which exacerbates the challenge of fine-grained visual concept learning. In this case, exploiting both web images and their associated texts is a requisite solution to combat real-world noise. In this paper, we propose Cross-modality Aligned Prototypes (CAPro), a unified prototypical contrastive learning framework to learn visual representations with correct semantics. For one thing, we leverage textual prototypes, which stem from the distinct concept definition of classes, to select clean images by text matching and thus disambiguate the formation of visual prototypes. For another, to handle missing and mismatched noisy texts, we resort to the visual feature space to complete and enhance individual texts and thereafter improve text matching. Such semantically aligned visual prototypes are further polished up with high-quality samples, and engaged in both cluster regularization and noise removal. Besides, we propose collective bootstrapping to encourage smoother and wiser label reference from appearance-similar instances in a manner of dictionary look-up. Extensive experiments on WebVision1k and NUS-WIDE (Web) demonstrate that CAPro well handles realistic noise under both single-label and multi-label scenarios. CAPro achieves new state-of-the-art performance and exhibits robustness to open-set recognition. Codes are available at https://github.com/yuleiqin/capro.

Open-world Semi-Supervised Learning (OSSL) is a realistic and challenging task, aiming to classify unlabeled samples from both seen and novel classes using partially labeled samples from the seen classes. Previous works typically explore the relationship of samples as priors on the pre-defined single-granularity labels to help novel class recognition. In fact, classes follow a taxonomy and samples can be classified at multiple levels of granularity, which contains more underlying relationships for supervision. We thus argue that learning with single-granularity labels results in sub-optimal representation learning and inaccurate pseudo labels, especially with unknown classes. In this paper, we take the initiative to explore and propose a uniformed framework, called Taxonomic context prIors Discovering and Aligning (TIDA), which exploits the relationship of samples under various granularity. It allows us to discover multi-granularity semantic concepts as taxonomic context priors (i.e., sub-class, target-class, and super-class), and then collaboratively leverage them to enhance representation learning and improve the quality of pseudo labels. Specifically, TIDA comprises two components: i) A taxonomic context discovery module that constructs a set of hierarchical prototypes in the latent space to discover the underlying taxonomic context priors; ii) A taxonomic context-based prediction alignment module that enforces consistency across hierarchical predictions to build the reliable relationship between classes among various granularity and provide additions supervision. We demonstrate that these two components are mutually beneficial for an effective OSSL framework, which is theoretically explained from the perspective of the EM algorithm. Extensive experiments on seven commonly used datasets show that TIDA can significantly improve the performance and achieve a new state of the art. The source codes are publicly available at https://github.com/rain305f/TIDA.

With ever increasing parameters and computation, vision-language pre-trained (VLP) models exhibit prohibitive expenditure in downstream task adaption. Recent endeavors mainly focus on parameter efficient transfer learning (PETL) for VLP models by only updating a small number of parameters. However, excessive computational overhead still plagues the application of VLPs. In this paper, we aim at parameter and computation efficient transfer learning (PCETL) for VLP models. In particular, PCETL not only needs to limit the number of trainable parameters in VLP models, but also to reduce the computational redundancy during inference, thus enabling a more efficient transfer. To approach this target, we propose a novel dynamic architecture skipping (DAS) approach towards effective PCETL. Instead of directly optimizing the intrinsic architectures of VLP models, DAS first observes the significances of their modules to downstream tasks via a reinforcement learning (RL) based process, and then skips the redundant ones with lightweight networks, i.e. adapters, according to the obtained rewards. In this case, the VLP model can well maintain the scale of trainable parameters while speeding up its inference on downstream tasks. To validate DAS, we apply it to two representative VLP models, namely ViLT and METER, and conduct extensive experiments on a bunch of VL tasks. The experimental results not only show the great advantages of DAS in reducing computational complexity, e.g. -11.97% FLOPs of METER on VQA2.0, but also confirm its competitiveness against existing PETL methods in terms of parameter scale and performance. Our source code is given in our appendix.

Nowadays, Multi-purpose Messaging Mobile App (MMMA) has become increasingly prevalent. MMMAs attract fraudsters and some cybercriminals provide support for frauds via black market accounts (BMAs). Compared to fraudsters, BMAs are not directly involved in frauds and are more difficult to detect. This paper illustrates our BMA detection system SGRL (Self-supervised Graph Representation Learning) used in WeChat, a representative MMMA with over a billion users. We tailor Graph Neural Network and Graph Self-supervised Learning in SGRL for BMA detection. The workflow of SGRL contains a pretraining phase that utilizes structural information, node attribute information and available human knowledge, and a lightweight detection phase. In offline experiments, SGRL outperforms state-of-the-art methods by 16.06%-58.17% on offline evaluation measures. We deploy SGRL in the online environment to detect BMAs on the billion-scale WeChat graph, and it exceeds the alternative by 7.27% on the online evaluation measure. In conclusion, SGRL can alleviate label reliance, generalize well to unseen data, and effectively detect BMAs in WeChat.

A resource-adaptive supernet adjusts its subnets for inference to fit the dynamically available resources. In this paper, we propose prioritized subnet sampling to train a resource-adaptive supernet, termed PSS-Net. We maintain multiple subnet pools, each of which stores the information of substantial subnets with similar resource consumption. Considering a resource constraint, subnets conditioned on this resource constraint are sampled from a pre-defined subnet structure space and high-quality ones will be inserted into the corresponding subnet pool. Then, the sampling will gradually be prone to sampling subnets from the subnet pools. Moreover, the one with a better performance metric is assigned with higher priority to train our PSS-Net, if sampling is from a subnet pool. At the end of training, our PSS-Net retains the best subnet in each pool to entitle a fast switch of high-quality subnets for inference when the available resources vary. Experiments on ImageNet using MobileNet-V1/V2 and ResNet-50 show that our PSS-Net can well outperform state-of-the-art resource-adaptive supernets. Our project is publicly available at https://github.com/chenbong/PSS-Net.

Variation of scales or aspect ratios has been one of the main challenges for tracking. To overcome this challenge, most existing methods adopt either multi-scale search or anchor-based schemes, which use a predefined search space in a handcrafted way and therefore limit their performance in complicated scenes. To address this problem, recent anchor-free based trackers have been proposed without using prior scale or anchor information. However, an inconsistency problem between classification and regression degrades the tracking performance. To address the above issues, we propose a simple yet effective tracker (named Siamese Box Adaptive Network, SiamBAN) to learn a target-aware scale handling schema in a data-driven manner. Our basic idea is to predict the target boxes in a per-pixel fashion through a fully convolutional network, which is anchor-free. Specifically, SiamBAN divides the tracking problem into classification and regression tasks, which directly predict objectiveness and regress bounding boxes, respectively. A no-prior box design is proposed to avoid tuning hyper-parameters related to candidate boxes, which makes SiamBAN more flexible. SiamBAN further uses a target-aware branch to address the inconsistency problem. Experiments on benchmarks including VOT2018, VOT2019, OTB100, UAV123, LaSOT and TrackingNet show that SiamBAN achieves promising performance and runs at 35 FPS.

Patient survival prediction based on gigapixel whole-slide histopathological images (WSIs) has become increasingly prevalent in recent years. A key challenge of this task is achieving an informative survival-specific global representation from those WSIs with highly complicated data correlation. This article proposes a multi-hypergraph based learning framework, called "HGSurvNet," to tackle this challenge. HGSurvNet achieves an effective high-order global representation of WSIs via multilateral correlation modeling in multiple spaces and a general hypergraph convolution network. It has the ability to alleviate over-fitting issues caused by the lack of training data by using a new convolution structure called hypergraph max-mask convolution. Extensive validation experiments were conducted on three widely-used carcinoma datasets: Lung Squamous Cell Carcinoma (LUSC), Glioblastoma Multiforme (GBM), and National Lung Screening Trial (NLST). Quantitative analysis demonstrated that the proposed method consistently outperforms state-of-the-art methods, coupled with the Bayesian Concordance Readjust loss. We also demonstrate the individual effectiveness of each module of the proposed framework and its application potential for pathology diagnosis and reporting empowered by its interpretability potential.

Graph Neural Networks have attracted increasing attention in recent years. However, existing GNN frameworks are deployed based upon simple graphs, which limits their applications in dealing with complex data correlation of multi-modal/multi-type data in practice. A few hypergraph-based methods have recently been proposed to address the problem of multi-modal/multi-type data correlation by directly concatenating the hypergraphs constructed from each single individual modality/type, which is difficult to learn an adaptive weight for each modality/type. In this paper, we extend the original conference version HGNN, and introduce a general high-order multi-modal/multi-type data correlation modeling framework called HGNN$^+$to learn an optimal representation in a single hypergraph based framework. It is achieved by bridging multi-modal/multi-type data and hyperedge with hyperedge groups. Specifically, in our method, hyperedge groups are first constructed to represent latent high-order correlations in each specific modality/type with explicit or implicit graph structures. An adaptive hyperedge group fusion strategy is then used to effectively fuse the correlations from different modalities/types in a unified hypergraph. After that a new hypergraph convolution scheme performed in spatial domain is used to learn a general data representation for various tasks. We have evaluated this framework on several popular datasets and compared it with recent state-of-the-art methods. The comprehensive evaluations indicate that the proposed HGNN$^+$framework can consistently outperform existing methods with a significant margin, especially when modeling implicit data correlations. We also release a toolbox called THU-DeepHypergraph for the proposed framework, which can be used for various of applications, such as data classification, retrieval and recommendation.

The mainstream approach for filter pruning is usually either to force a hard-coded importance estimation upon a computation-heavy pretrained model to select "important" filters, or to impose a hyperparameter-sensitive sparse constraint on the loss objective to regularize the network training. In this paper, we present a novel filter pruning method, dubbed dynamic-coded filter fusion (DCFF), to derive compact CNNs in a computation-economical and regularization-free manner for efficient image classification. Each filter in our DCFF is first given an inter-similarity distribution with a temperature parameter as a filter proxy, on top of which, a fresh Kullback-Leibler divergence based dynamic-coded criterion is proposed to evaluate the filter importance. In contrast to simply keeping high-score filters in other methods, we propose the concept of filter fusion, i.e., the weighted averages using the assigned proxies, as our preserved filters. We obtain a one-hot inter-similarity distribution as the temperature parameter approaches infinity. Thus, the relative importance of each filter can vary along with the training of the compact CNN, leading to dynamically changeable fused filters without both the dependency on the pretrained model and the introduction of sparse constraints. Extensive experiments on classification benchmarks demonstrate the superiority of our DCFF over the compared counterparts. For example, our DCFF derives a compact VGGNet-16 with only 72.77M FLOPs and 1.06M parameters while reaching top-1 accuracy of 93.47% on CIFAR-10. A compact ResNet-50 is obtained with 63.8% FLOPs and 58.6% parameter reductions, retaining 75.60% top-1 accuracy on ILSVRC-2012. Our code, narrower models and training logs are available at https://github.com/lmbxmu/DCFF.

Binary neural networks (BNNs) have attracted broad research interest due to their efficient storage and computational ability. Nevertheless, a significant challenge of BNNs lies in handling discrete constraints while ensuring bit entropy maximization, which typically makes their weight optimization very difficult. Existing methods relax the learning using the sign function, which simply encodes positive weights into +1s, and -1s otherwise. Alternatively, we formulate an angle alignment objective to constrain the weight binarization to$\lbrace$0,+1$\rbrace$to solve the challenge. In this paper, we show that our weight binarization provides an analytical solution by encoding high-magnitude weights into +1s, and 0 s otherwise. Therefore, a high-quality discrete solution is established in a computationally efficient manner without the sign function. We prove that the learned weights of binarized networks roughly follow a Laplacian distribution that does not allow entropy maximization, and further demonstrate that it can be effectively solved by simply removing the$\ell _{2}$regularization during network training. Our method, dubbed sign-to-magnitude network binarization (SiMaN), is evaluated on CIFAR-10 and ImageNet, demonstrating its superiority over the sign-based state-of-the-arts. Our source code, experimental settings, training logs and binary models are available athttps://github.com/lmbxmu/SiMaN.

Though network pruning receives popularity in reducing the complexity of convolutional neural networks (CNNs), it remains an open issue to concurrently maintain model accuracy as well as achieve significant speedups on general CPUs. In this paper, we propose a novel 1×N pruning pattern to break this limitation. In particular, consecutive N output kernels with the same input channel index are grouped into one block, which serves as a basic pruning granularity of our pruning pattern. Our 1×N pattern prunes these blocks considered unimportant. We also provide a workflow of filter rearrangement that first rearranges the weight matrix in the output channel dimension to derive more influential blocks for accuracy improvements and then applies similar rearrangement to the next-layer weights in the input channel dimension to ensure correct convolutional operations. Moreover, the output computation after our 1×N pruning can be realized via a parallelized block-wise vectorized operation, leading to significant speedups on general CPUs. The efficacy of our pruning pattern is proved with experiments on ILSVRC-2012. For example, given the pruning rate of 50% and N=4, our pattern obtains about 3.0% improvements over filter pruning in the top-1 accuracy of MobileNet-V2. Meanwhile, it obtains 56.04ms inference savings on Cortex-A7 CPU over weight pruning. Our project is made available at https://github.com/lmbxmu/1xN.

Few-shot class-incremental learning (FSCIL) is challenged by catastrophically forgetting old classes and over-fitting new classes. Revealed by our analyses, the problems are caused by feature distribution crumbling, which leads to class confusion when continuously embedding few samples to a fixed feature space. In this study, we propose a Dynamic Support Network (DSN), which refers to an adaptively updating network with compressive node expansion to "support" the feature space. In each training session, DSN tentatively expands network nodes to enlarge feature representation capacity for incremental classes. It then dynamically compresses the expanded network by node self-activation to pursue compact feature representation, which alleviates over-fitting. Simultaneously, DSN selectively recalls old class distributions during incremental learning to support feature distributions and avoid confusion between classes. DSN with compressive node expansion and class distribution recalling provides a systematic solution for the problems of catastrophic forgetting and overfitting. Experiments on CUB, CIFAR-100, and miniImage datasets show that DSN significantly improves upon the baseline approach, achieving new state-of-the-arts.

Network pruning is an effective approach to reduce network complexity with acceptable performance compromise. Existing studies achieve the sparsity of neural networks via time-consuming weight training or complex searching on networks with expanded width, which greatly limits the applications of network pruning. In this paper, we show that high-performing and sparse sub-networks without the involvement of weight training, termed "lottery jackpots", exist in pre-trained models with unexpanded width. Our presented lottery jackpots are traceable through empirical and theoretical outcomes. For example, we obtain a lottery jackpot that has only 10% parameters and still reaches the performance of the original dense VGGNet-19 without any modifications on the pre-trained weights on CIFAR-10. Furthermore, we improve the efficiency for searching lottery jackpots from two perspectives. First, we observe that the sparse masks derived from many existing pruning criteria have a high overlap with the searched mask of our lottery jackpot, among which, the magnitude-based pruning results in the most similar mask with ours. In compliance with this insight, we initialize our sparse mask using the magnitude-based pruning, resulting in at least 3× cost reduction on the lottery jackpot searching while achieving comparable or even better performance. Second, we conduct an in-depth analysis of the searching process for lottery jackpots. Our theoretical result suggests that the decrease in training loss during weight searching can be disturbed by the dependency between weights in modern networks. To mitigate this, we propose a novel short restriction method to restrict change of masks that may have potential negative impacts on the training loss, which leads to a faster convergence and reduced oscillation for searching lottery jackpots. Consequently, our searched lottery jackpot removes 90% weights in ResNet-50, while it easily obtains more than 70% top-1 accuracy using only 5 searching epochs on ImageNet.

Robust tracking has a variety of practical applications. Despite many years of progress, it is still a difficult problem due to enormous uncertainties in real-world scenes. To address this issue, we propose a robust anchor-free based tracking model with uncertainty estimation. Within the model, a new data-driven uncertainty estimation strategy is proposed to generate uncertainty-aware features with promising discriminative and descriptive power. Then, a simple yet effective pyramid-wise cross correlation operation is constructed to extract multi-scale semantic features that provide rich correlation information for uncertainty-aware estimation and thus enhances the tracking robustness. Finally, a semantic consistency checking branch is designed to further estimate uncertainty of output results from the classification and regression branches by adaptively generating semantically consistent labels. Experiments on six benchmarks (i.e., OTB100, VOT2018, VOT2020, TrackingNet, GOT-10K and LaSOT) show the competing performance of our tracker with 130 FPS.

Despite that both local and context information are crucial for robust tracking, existing CNN-based and transformer-based methods mainly focus on one of these aspects. Consequently, the former fails to exploit rich global context information due to the limited receptive field, while the latter suffers from the deficiencies in constructing the local relationship among neighboring regions. To address this issue, we propose the SiamPIN tracker, based on our Parallel Interaction Network. It consists of two effective modules, namely Global Aggregation Block (GAB) and Local Process Block (LPB). GAB perceives the global context to capture the long-range spatial dependency through a transformer-based architecture. Meanwhile, LPB performs local information extraction using a CNN model to retain the detailed appearance information of the target. These two modules are connected consecutively to compose a Trans-Conv unit block, which transmits the global context information to the local feature extraction procedure, hence enables the interaction of global-local information flow. Several such blocks are cascaded so that our model can learn to aggregate local and context information interactively. The proposed tracker achieves state-of-the-art performance on six benchmark datasets, while maintaining a real time running speed.

Deep neural networks have been widely used in various domains owing to the success of deep learning. However, recent studies have shown that these models are vulnerable to adversarial examples, leading to inaccurate predictions. In this paper, we focus on the issue of adversarial robustness by examining it through the lens of semantic information, which drives us to propose a new perspective,i.e., adversarial attacks destroy the correlation between visual representations and semantic word vectors, while adversarial training restores it. Additionally, we discover that the correlation among robust representations of different categories aligns with the correlation among the corresponding semantic word vectors. Based on these empirical observations, we incorporate the semantic information into the model training process and propose Semantic Constraint Adversarial Robust Learning (SCARL). Firstly, inspired by the information-theoretical perspective, we maximize mutual information to bridge the information gap between the visual representations and the corresponding semantic word vectors in the embedding space. We further provide a differentiable lower bound to optimize such mutual information efficiently. Secondly, we introduce a novel semantic structure constraint that maintains the structure of visual representations consistent with that of semantic word vectors. Finally, we integrate these techniques with adversarial training to learn robust visual representations. We conduct extensive experiments on several datasets (such as CIFAR and TinyImageNet) and evaluate the robustness against various adversarial attacks (such as PGD-attack and AutoAttack), demonstrating the benefits of incorporating semantic information for improving model robustness. Our code is available at https://github.com/SkyKuang/SCARL.

Self-attention (SA) based networks have achieved great success in image captioning, constantly dominating the leaderboards of online benchmarks. However, existing SA networks still suffer from distance insensitivity and low-rank bottleneck. In this paper, we aim to optimize SA in terms of two aspects, thereby addressing the above issues. First, we introduce a Distance-sensitive Self-Attention (DSA), which considers the raw geometric distances between query-key pairs in the 2D images during SA modeling. Second, we present a simple yet effective approach, named Multi-branch Self-Attention (MSA) to compensate for the low-rank bottleneck. MSA treats a multi-head self-attention layer as a branch and duplicates it multiple times to increase the expressive power of SA. To validate the effectiveness of the two designs, we apply them to the standard self-attention network, and conduct extensive experiments on the highly competitive MS-COCO dataset. We achieve new state-of-the-art performance on both the local and online test sets,i.e., 135.1% CIDEr on the Karpathy split and 135.4% CIDEr on the official online split.

Generative Adversarial Networks (GANs) have been widely-used in image translation, but their high computational and storage costs impede the deployment on mobile devices. Prevalent methods for CNN compression cannot be directly applied to GANs due to the specificity of GAN tasks and the unstable adversarial training. To solve these, in this paper, we introduce a novel GAN compression method, termed DMAD, by proposing a Differentiable Mask and a co-Attention Distillation. The former searches for a light-weight generator architecture in a training-adaptive manner. To overcome channel inconsistency when pruning the residual connections, an adaptive cross-block group sparsity is further incorporated. The latter simultaneously distills informative attention maps from both the generator and discriminator of a pre-trained model to the searched generator, effectively stabilizing the adversarial training of our light-weight model. Experiments show that DMAD can reduce the Multiply Accumulate Operations (MACs) of CycleGAN by 13x and that of Pix2Pix by 4x while retaining a comparable performance against the full model. Our code can be available at https://github.com/SJLeo/DMAD.

Attention has become an indispensable component of the models of various multimedia tasks likeImage Captioning(IC) andVisual Question Answering(VQA). However, most existing attention modules are designed for capturing the spatial dependency, and are still insufficient in semantic understanding,e.g., the categories of objects and their attributes, which is also critical for image captioning. To compensate for this defect, we propose a novel attention module termedChannel-wise Attention Block(CAB) to model channel-wise dependency for both visual modality and linguistic modality, thereby improving semantic learning and multi-modal reasoning simultaneously. Specifically, CAB has two novel designs to tackle with the high overhead of channel-wise attention, which are thereduction-reconstruction block structureand thegating-based attention prediction. Based on CAB, we further propose a novelSemantic-enhanced Dual Attention Transformer(termed SDATR), which combines the merits of spatial and channel-wise attentions. To validate SDATR, we conduct extensive experiments on the MS COCO dataset and yield new state-of-the-art performance of 134.5 CIDEr score on COCO Karpathy test split and 136.0 CIDEr score on the official online testing server. To examine the generalization of SDATR, we also apply it to the task of visual question answering, where superior performance gains are also witnessed. The code and models are publicly available athttps://github.com/xmu-xiaoma666/SDATR.

Recent works have validated the benefit of integrating spatial information into deep networks to improve pixel-level prediction tasks such as monocular depth estimation. However, how to efficiently and robustly integrate spatial cues retains as an open problem. In this paper, we introduce the Side Prediction Aggregation (termed SPA) method to enhance the embedding of scene structural information from low-level to high-level layers. To improve the estimation accuracy, the proposed method is further equipped with continuous Spatial Refinement Loss (termed SRL) at multiple resolutions with negligible extra computation. Besides, the proposed sequential network can further perform adversarial learning at multiple resolutions. Such an adversarial refinement strategy greatly improves the accuracy of estimated depth with a little extra computation. Without using any pre-trained models, our network achieves the the-state-of-art accuracy on KITTI, NYUD V2, and Cityscapes datasets, which has achieved real-time depth estimation online.

Existing online knowledge distillation approaches either adopt the student with the best performance or construct an ensemble model for better holistic performance. However, the former strategy ignores other students' information, while the latter increases the computational complexity during deployment. In this article, we propose a novel method for online knowledge distillation, termed feature fusion and self-distillation (FFSD), which comprises two key components: FFSD, toward solving the above problems in a unified framework. Different from previous works, where all students are treated equally, the proposed FFSD splits them into a leader student set and a common student set. Then, the feature fusion module converts the concatenation of feature maps from all common students into a fused feature map. The fused representation is used to assist the learning of the leader student. To enable the leader student to absorb more diverse information, we design an enhancement strategy to increase the diversity among students. Besides, a self-distillation module is adopted to convert the feature map of deeper layers into a shallower one. Then, the shallower layers are encouraged to mimic the transformed feature maps of the deeper layers, which helps the students to generalize better. After training, we simply adopt the leader student, which achieves superior performance, over the common students, without increasing the storage or inference cost. Extensive experiments on CIFAR-100 and ImageNet demonstrate the superiority of our FFSD over existing works. The code is available at https://github.com/SJLeo/FFSD.

This article focuses on filter-level network pruning. A novel pruning method, termed CLR-RNF, is proposed. We first reveal a "long-tail" pruning problem in magnitude-based weight pruning methods and then propose a computation-aware measurement for individual weight importance, followed by a cross-layer ranking (CLR) of weights to identify and remove the bottom-ranked weights. Consequently, the per-layer sparsity makes up the pruned network structure in our filter pruning. Then, we introduce a recommendation-based filter selection scheme where each filter recommends a group of its closest filters. To pick the preserved filters from these recommended groups, we further devise a k -reciprocal nearest filter (RNF) selection scheme where the selected filters fall into the intersection of these recommended groups. Both our pruned network structure and the filter selection are nonlearning processes, which, thus, significantly reduces the pruning complexity and differentiates our method from existing works. We conduct image classification on CIFAR-10 and ImageNet to demonstrate the superiority of our CLR-RNF over the state-of-the-arts. For example, on CIFAR-10, CLR-RNF removes 74.1% FLOPs and 95.0% parameters from VGGNet-16 with even 0.3% accuracy improvements. On ImageNet, it removes 70.2% FLOPs and 64.8% parameters from ResNet-50 with only 1.7% top-five accuracy drops. Our project is available at https://github.com/lmbxmu/CLR-RNF.

Channel pruning has been long studied to compress convolutional neural networks (CNNs), which significantly reduces the overall computation. Prior works implement channel pruning in an unexplainable manner, which tends to reduce the final classification errors while failing to consider the internal influence of each channel. In this article, we conduct channel pruning in a white box. Through deep visualization of feature maps activated by different channels, we observe that different channels have a varying contribution to different categories in image classification. Inspired by this, we choose to preserve channels contributing to most categories. Specifically, to model the contribution of each channel to differentiating categories, we develop a class-wise mask for each channel, implemented in a dynamic training manner with respect to the input image's category. On the basis of the learned class-wise mask, we perform a global voting mechanism to remove channels with less category discrimination. Lastly, a fine-tuning process is conducted to recover the performance of the pruned model. To our best knowledge, it is the first time that CNN interpretability theory is considered to guide channel pruning. Extensive experiments on representative image classification tasks demonstrate the superiority of our White-Box over many state-of-the-arts (SOTAs). For instance, on CIFAR-10, it reduces 65.23% floating point operations per seconds (FLOPs) with even 0.62% accuracy improvement for ResNet-110. On ILSVRC-2012, White-Box achieves a 45.6% FLOP reduction with only a small loss of 0.83% in the top-1 accuracy for ResNet-50. Code is available at https://github.com/zyxxmu/White-Box.

Referring expression comprehension (REC) is an emerging research topic in computer vision, which refers to the detection of a target region in an image given a test description. Most existing REC methods follow a multistage pipeline, which is computationally expensive and greatly limits the applications of REC. In this article, we propose a one-stage model toward real-time REC, termed real-time global inference network (RealGIN). RealGIN addresses the issues of expression diversity and complexity of REC with two innovative designs: adaptive feature selection (AFS) and Global Attentive ReAsoNing (GARAN). Expression diversity concerns varying expression content, which includes information such as colors, attributes, locations, and fine-grained categories. To address this issue, AFS adaptively fuses features of different semantic levels to tackle the changes in expression content. In contrast, expression complexity concerns the complex relational conditions in expressions that are used to identify the referent. To this end, GARAN uses the textual feature as a pivot to collect expression-aware visual information from all regions and then diffuses this information back to each region, which provides sufficient context for modeling the relational conditions in expressions. On five benchmark datasets, i.e., RefCOCO, RefCOCO+, RefCOCOg, ReferIT, and Flickr30k, the proposed RealGIN outperforms most existing methods and achieves very competitive performances against the most advanced one, i.e., MAttNet. More importantly, under the same hardware, RealGIN can boost the processing speed by 10-20 times over the existing methods.

Transfer adversarial attack is a non-trivial black-box adversarial attack that aims to craft adversarial perturbations on the surrogate model and then apply such perturbations to the victim model. However, the transferability of perturbations from existing methods is still limited, since the adversarial perturbations are easily overfitting with a single surrogate model and specific data pattern. In this paper, we propose a Learning to Learn Transferable Attack (LLTA) method, which makes the adversarial perturbations more generalized via learning from both data and model augmentation. For data augmentation, we adopt simple random resizing and padding. For model augmentation, we randomly alter the back propagation instead of the forward propagation to eliminate the effect on the model prediction. By treating the attack of both specific data and a modified model as a task, we expect the adversarial perturbations to adopt enough tasks for generalization. To this end, the meta-learning algorithm is further introduced during the iteration of perturbation generation. Empirical results on the widely-used dataset demonstrate the effectiveness of our attack method with a 12.85% higher success rate of transfer attack compared with the state-of-the-art methods. We also evaluate our method on the real-world online system, i.e., Google Cloud Vision API, to further show the practical potentials of our method.

With various facial manipulation techniques arising, face forgery detection has drawn growing attention due to security concerns. Previous works always formulate face forgery detection as a classification problem based on cross-entropy loss, which emphasizes category-level differences rather than the essential discrepancies between real and fake faces, limiting model generalization in unseen domains. To address this issue, we propose a novel face forgery detection framework, named Dual Contrastive Learning (DCL), which specially constructs positive and negative paired data and performs designed contrastive learning at different granularities to learn generalized feature representation. Concretely, combined with the hard sample selection strategy, Inter-Instance Contrastive Learning (Inter-ICL) is first proposed to promote task-related discriminative features learning by especially constructing instance pairs. Moreover, to further explore the essential discrepancies, Intra-Instance Contrastive Learning (Intra-ICL) is introduced to focus on the local content inconsistencies prevalent in the forged faces by constructing local region pairs inside instances. Extensive experiments and visualizations on several datasets demonstrate the generalization of our method against the state-of-the-art competitors. Our Code is available at https://github.com/Tencent/TFace.git.

This paper focuses on the challenging crowd counting task. As large-scale variations often exist within crowd images, neither fixed-size convolution kernel of CNN nor fixed-size attention of recent vision transformers can well handle this kind of variations. To address this problem, we propose a Multifaceted Attention Network (MAN) to improve transformer models in local spatial relation encoding. MAN incorporates global attention from vanilla transformer, learnable local attention, and instance attention into a counting model. Firstly, the local Learnable Region Attention (LRA) is proposed to assign attention exclusive for each feature location dynamically. Secondly, we design the Local Attention Regularization to supervise the training of LRA by minimizing the deviation among the attention for different feature locations. Finally, we provide an Instance Attention mechanism to focus on the most important instances dynamically during training. Extensive experiments on four challenging crowd counting datasets namely ShanghaiTech, UCF-QNRF, JHU++, and NWPU have validated the proposed method. Code: https://github.com/LoraLinH/Boosting-Crowd-Counting-via-Multifaceted-Attention.

In this paper, we study teacher-student learning from the perspective of data initialization and propose a novel algorithm called Active Teacher11Source code are available at: https://github.com/HunterJLin/ActiveTeacher for semi-supervised object detection (SSOD). Active Teacher extends the teacher-student framework to an iterative version, where the label set is partially initialized and gradually augmented by evaluating three key factors of unlabeled examples, including difficulty, information and diversity. With this design, Active Teacher can maximize the effect of limited label information while improving the quality of pseudo-labels. To validate our approach, we conduct extensive experiments on the MS-COCO benchmark and compare Active Teacher with a set of recently proposed SSOD methods. The experimental results not only validate the superior performance gain of Active Teacher over the compared methods, but also show that it enables the baseline network, i.e., Faster-RCNN, to achieve 100% supervised performance with much less label expenditure, i.e. 40% labeled examples on MS-COCO. More importantly, we believe that the experimental analyses in this paper can provide useful empirical knowledge for data annotation in practical applications.

Current Image Captioning (IC) methods predict textual words sequentially based on the input visual information from the visual feature extractor and the partially generated sentence information. However, for most cases, the partially generated sentence may dominate the target word prediction due to the insufficiency of visual information, making the generated descriptions irrelevant to the content of the given image. In this paper, we propose a Dual Information Flow Network (DIFNet11Source code is available at: https://github.com/mrwu-mac/DIFNet) to address this issue, which takes segmentation feature as another visual information source to enhance the contribution of visual information for prediction. To maximize the use of two information flows, we also propose an effective feature fusion module termed Iterative Independent Layer Normalization (IILN) which can condense the most relevant inputs while retraining modality-specific information in each flow. Experiments show that our method is able to enhance the dependence of prediction on visual information, making word prediction more focused on the visual content, and thus achieves new state-of-the-art performance on the MSCOCO dataset, e.g., 136.2 CIDEr on COCO Karpathy test split.

Performance evaluation strategy is one of the most important factors that determine the effectiveness and efficiency in Neural Architecture Search (NAS). Existing strategies, such as employing standard training or performance predictor, often suffer from high computational complexity and low generality. To address this issue, we propose to rank architectures by Representation Mutual Information (RMI). Specifically, given an arbitrary architecture that has decent accuracy, architectures that have high RMI with it always yield good accuracies. As an accurate performance indicator to facilitate NAS, RMI not only generalizes well to different search spaces, but is also efficient enough to evaluate architectures using only one batch of data. Building upon RMI, we further propose a new search algorithm termed RMI-NAS, facilitating with a theorem to guarantee the global optimal of the searched architecture. In particular, RMI-NAS first randomly samples architectures from the search space, which are then effectively classified as positive or negative samples by RMI. We then use these samples to train a random forest to explore new regions, while keeping track of the distribution of positive architectures. When the sample size is sufficient, the architecture with the largest probability from the aforementioned distribution is selected, which is theoretically proved to be the optimal solution. The architectures searched by our method achieve remarkable top-1 accuracies with the magnitude times faster search process. Besides, RMI-NAS also generalizes to different datasets and search spaces. Our code has been made available at https://git.openi.org.cn/PCL_AutoML/XNAS.

Learning to synthesize data has emerged as a promising direction in zero-shot quantization (ZSQ), which represents neural networks by low-bit integer without accessing any of the real data. In this paper, we observe an interesting phenomenon of intra-class heterogeneity in real data and show that existing methods fail to retain this property in their synthetic images, which causes a limited performance increase. To address this issue, we propose a novel zero-shot quantization method referred to as IntraQ. First, we propose a local object reinforcement that locates the target objects at different scales and positions of the synthetic images. Second, we introduce a marginal distance constraint to form class-related features distributed in a coarse area. Lastly, we devise a soft inception loss which injects a soft prior label to prevent the synthetic images from being over-fitting to a fixed object. Our IntraQ is demonstrated to well retain the intra-class heterogeneity in the synthetic images and also observed to perform state-of-the-art. For example, compared to the advanced ZSQ, our IntraQ obtains 9.17% increase of the top-1 accuracy on ImageNet when all layers of MobileNetV1 are quantized to 4-bit. Code is at https://github.com/zysxmu/IntraQ

Recently, Vision Transformer (ViT) has achieved remarkable success in several computer vision tasks. The progresses are highly relevant to the architecture design, then it is worthwhile to propose Transformer Architecture Search (TAS) to search for better ViTs automatically. However, current TAS methods are time-consuming and existing zero-cost proxies in CNN do not generalize well to the ViT search space according to our experimental observations. In this paper, for the first time, we investigate how to conduct TAS in a training-free manner and devise an effective training-free TAS (TF-TAS) scheme. Firstly, we observe that the properties of multi-head self-attention (MSA) and multi-layer perceptron (MLP) in ViTs are quite different and that the synaptic diversity of MSA affects the performance notably. Secondly, based on the observation, we devise a modular strategy in TF-TAS that evaluates and ranks ViT architectures from two theoretical perspectives: synaptic diversity and synaptic saliency, termed as DSS-indicator. With DSS-indicator, evaluation results are strongly corre-lated with the test accuracies of ViT models. Experimental results demonstrate that our TF- TAS achieves a competitive performance against the state-of-the-art manually or automatically design ViT architectures, and it promotes the searching efficiency in ViT search space greatly: from about 24 GPU days to less than 0.5 GPU days. Moreover, the proposed DSS-indicator outperforms the existing cutting-edge zero-cost approaches (e.g., TE-score and NASWOT).

Motion information is critical for action recognition. Most existing methods perform motion enhancement only from the channel or spatiotemporal dimension. They may fail to achieve fine-grained motion modeling, leading to suboptimal performance. In this paper, we propose a novel motion distinction network (MDNet) to address this challenging problem. Specifically, we first propose a channel-wise motion enhancement (CME) module, which aims to emphasize motion-related channels by leveraging a channel-wise gating mechanism. Then, we propose a cascaded spatiotemporal enhancement (CSTE) module to enhance motion features along the spatiotemporal dimension. Moreover, we design a multi-attention fusion strategy to further refine the enhanced motion features in a moderate manner, enabling the network to focus on discriminative motion regions. The proposed two modules and fusion strategy are complementary for fine-grained motion enhancement. Extensive experiments are conducted on the challenging Something-Something V1 and Kinetics-400 datasets to show the effectiveness of the proposed MDNet.

Few-shot action recognition aims to learn novel action classes with only a few annotated samples. This is a challenging problem because motion modeling is difficult, especially when a few training samples are available. Visual tempo, which is an essential variation factor of video semantics, characterizes the dynamic motion information in action videos. In this work, we propose a visual tempo contrastive learning framework (VTCL) to tackle the few-shot action recognition problem. Specifically, we propose a visual tempo encoding (VTE) module for visual tempo learning. The VTE module samples the same action instance at different frame sampling rates to obtain different visual tempo encoding vectors, which jointly form the features of each instance. To enhance the discriminability of visual tempo encoding vectors, we propose a visual tempo contrastive encoding (VTCE) loss to promote intra-class compactness and inter-class difference of visual tempo encoding vectors. Extensive experiments demonstrate that the proposed VTCL achieves promising results among the competing state-of-the-art methods on three few-shot action recognition datasets, including Kinetics, UCF101 and HMDB51.

Recently, it has been shown that the traditional Image Signal Processing (ISP) can be replaced by deep neural networks due to their superior performance. However, most of these networks require heavy computation burden and thus are far from sufficient to be deployed on resource-limited platforms, including but not limited to mobile devices and FPGA. To tackle this challenge, we propose an automated search framework that derives ISP models with high image quality while satisfying the low-computation requirement. To reduce the search cost, we adopt the weight-sharing strategy by introducing a supernet and decouple the architecture search into two stages, supernet training and hard-aware evolutionary search. With the proposed framework, we can train the ISP model once and quickly find high-performance but low-computation models on multiple devices. Experiments demonstrate that the searched ISP models have an excellent trade-off between image quality and model complexity, i.e., achieve compelling reconstruction quality with more than 90% reduction in FLOPs as compared to the state-of-the-art networks.

Video-text retrieval has been a crucial and fundamental task in multi-modal research. The development of video-text retrieval has been considerably promoted by large-scale multi-modal contrastive pre-training, which primarily focuses on coarse-grained or fine-grained contrast. However, cross-grained contrast, which is the contrast between coarse-grained representations and fine-grained representations, has rarely been explored in prior research. Compared with fine-grained or coarse-grained contrasts, cross-grained contrast calculate the correlation between coarse-grained features and each fine-grained feature, and is able to filter out the unnecessary fine-grained features guided by the coarse-grained feature during similarity calculation, thus improving the accuracy of retrieval. To this end, this paper presents a novel multi-grained contrastive model, namely X-CLIP, for video-text retrieval. However, another challenge lies in the similarity aggregation problem, which aims to aggregate fine-grained and cross-grained similarity matrices to instance-level similarity. To address this challenge, we propose the Attention Over Similarity Matrix (AOSM) module to make the model focus on the contrast between essential frames and words, thus lowering the impact of unnecessary frames and words on retrieval results. With multi-grained contrast and the proposed AOSM module, X-CLIP achieves outstanding performance on five widely-used video-text retrieval datasets, including MSR-VTT (49.3 [email protected]), MSVD (50.4 [email protected]), LSMDC (26.1 [email protected]), DiDeMo (47.8 [email protected]) and ActivityNet (46.2 [email protected]).

GAN inversion aims to invert an input image into the latent space of a pre-trained GAN. Despite the recent advances in GAN inversion, there remain challenges to mitigate the tradeoff between distortion and editability, i.e. reconstructing the input image accurately and editing the inverted image with a small visual quality drop. The recently proposed pivotal tuning model makes significant progress towards reconstruction and editability, by using a two-step approach that first inverts the input image into a latent code, called pivot code, and then alters the generator so that the input image can be accurately mapped into the pivot code. Here, we show that both reconstruction and editability can be improved by a proper design of the pivot code. We present a simple yet effective method, named cycle encoding, for a high-quality pivot code. The key idea of our method is to progressively train an encoder in varying spaces according to a cycle scheme: W->W+->W. This training methodology preserves the properties of both W and W+ spaces, i.e. high editability of W and low distortion of W+. To further decrease the distortion, we also propose to refine the pivot code with an optimization-based method, where a regularization term is introduced to reduce the degradation in editability. Qualitative and quantitative comparisons to several state-of-the-art methods demonstrate the superiority of our approach.

Text-to-face (T2F) generation is an emerging research hot spot in multimedia, which aims to synthesize vivid portraits based on the given descriptions. Its main challenge lies in the accurate alignments from texts to image pixels, which pose a high demand in generation fidelity. We define T2F as a pixel synthesis problem conditioned on the texts and propose a novel dynamic pixel synthesis network, PixelFace, for end-to-end T2F generation in this paper. To fully exploit the prior knowledge for T2F synthesis, we propose a novel dynamic parameter generation module, which transforms text features into dynamic knowledge embeddings for end-to-end pixel regression. These knowledge embeddings are example-dependent and spatially related to image pixels, based on which PixelFace can exploit the text priors for high-quality text-guided face generation. To validate the proposed PixelFace, we conduct extensive experiments on the MMCelebA, and compare PixelFace with a set of state-of-the-art methods in T2F and T2I generations, e.g., StyleCLIP and TediGAN. The experimental results not only show the greater performance of PixelFace than the compared methods but also validates its merits over existing T2F methods in both text-image matching and inference speed. Codes will be released at: \textcolormagenta \urlhttps://github.com/pengjunn/PixelFace .

Text-to-face (T2F) generation is an emerging research hot spot in multimedia, and its main challenge lies in the high fidelity requirement of generated portraits. Many existing works resort to exploring the latent space in a pre-trained generator, e.g., StyleGAN, which has obvious shortcomings in efficiency and generalization ability. In this paper, we propose a generative network for open-ended text-to-face generation, which is termed OpenFaceGAN. Differing from existing StyleGAN-based methods, OpenFaceGAN constructs an effective multi-modal latent space that directly converts the natural language description into a face. This mapping paradigm can fit the real data distribution well and make the model capable of open-ended and even zero-shot T2F generation. Our method improves the inference speed by an order of magnitude, e.g., 294 times than TediGAN. Based on OpenFaceGAN, we further explore text-guided face manipulation (editing). In particular, we propose a parameterized module, OpenEditor, to automatically disentangle the target latent code and update the original style information. OpenEditor also makes OpenFaceGAN directly applicable for most manipulation instructions without example-dependent searches or optimizations, greatly improving the efficiency of face manipulation. We conduct extensive experiments on two benchmark datasets namely Multi-Modal CelebA-HQ and Face2Text-v1.0. The experimental results not only show the superior performance of OpenFaceGAN to the existing T2F methods in both image quality and image-text matching degree but also greatly confirm its outstanding ability in the zero-shot generation. Codes will be released at: \textcolormagenta \urlhttps://github.com/pengjunn/OpenFace

Although person re-identification has achieved an impressive improvement in recent years, the common occlusion case caused by different obstacles is still an unsettled issue in real application scenarios. Existing methods mainly address this issue by employing body clues provided by an extra network to distinguish the visible part. Nevertheless, the inevitable domain gap between the assistant model and the ReID datasets has highly increased the difficulty to obtain an effective and efficient model. To escape from the extra pre-trained networks and achieve an automatic alignment in an end-to-end trainable network, we propose a novel Dynamic Prototype Mask (DPM) based on two self-evident prior knowledge. Specifically, we first devise a Hierarchical Mask Generator which utilizes the hierarchical semantic to select the visible pattern space between the high-quality holistic prototype and the feature representation of the occluded input image. Under this condition, the occluded representation could be well aligned in a selected subspace spontaneously. Then, to enrich the feature representation of the high-quality holistic prototype and provide a more complete feature space, we introduce a Head Enrich Module to encourage different heads to aggregate different patterns representation in the whole image. Extensive experimental evaluations conducted on occluded and holistic person re-identification benchmarks demonstrate the superior performance of the DPM over the state-of-the-art methods.

By forcing N out of M consecutive weights to be non-zero, the recent N:M fine-grained network sparsity has received increasing attention with its two attractive advantages over traditional irregular network sparsity methods: 1) Promising performance at a high sparsity. 2) Significant speedups when performed on NVIDIA A100 GPUs. Current implementation on N:M sparsity requires a tedious pre-training phase or computationally heavy from-scratch training. To circumvent these problems, this paper presents an efficient solution for achieving N:M fine-grained sparsity from scratch. Specifically, we first make a re-formulation to convert the N:M fine-grained sparsity into a combinatorial problem, in which, the object falls into choosing the best weight combination among $C_M^N$ candidates. Then, we equip each combination with a learnable importance score, which can be jointly optimized along with its associated weights. Through rigorous proof, we demonstrate that the magnitude of the optimized score well reflects the importance of its corresponding weights combination to the training loss. Therefore, by gradually removing combinations with smaller scores till the best one is left, N:M fine-grained sparsity can be efficiently optimized during the normal training phase without any extra expenditure. Comprehensive experimental results have demonstrated that our proposed method for learning best combination, dubbed as LBC, consistently increases the efficacy of the off-the-shelf N:M methods across varying networks and datasets. Our project is released at https://github.com/zyxxmu/LBC.

Large-scale vision-language pre-training has achieved promising results on downstream tasks. Existing methods highly rely on the assumption that the image-text pairs crawled from the Internet are in perfect one-to-one correspondence. However, in real scenarios, this assumption can be difficult to hold: the text description, obtained by crawling the affiliated metadata of the image, often suffers from the semantic mismatch and the mutual compatibility. To address these issues, we introduce PyramidCLIP, which constructs an input pyramid with different semantic levels for each modality, and aligns visual elements and linguistic elements in the form of hierarchy via peer-level semantics alignment and cross-level relation alignment. Furthermore, we soften the loss of negative samples (unpaired samples) so as to weaken the strict constraint during the pre-training stage, thus mitigating the risk of forcing the model to distinguish compatible negative pairs. Experiments on five downstream tasks demonstrate the effectiveness of the proposed PyramidCLIP. In particular, with the same amount of 15 million pre-training image-text pairs, PyramidCLIP exceeds CLIP on ImageNet zero-shot classification top-1 accuracy by 10.6%/13.2%/10.0% with ResNet50/ViT-B32/ViT-B16 based image encoder respectively. When scaling to larger datasets, PyramidCLIP achieves the state-of-the-art results on several downstream tasks. In particular, the results of PyramidCLIP-ResNet50 trained on 143M image-text pairs surpass that of CLIP using 400M data on ImageNet zero-shot classification task, significantly improving the data efficiency of CLIP.

Deep neural networks often suffer from poor generalization caused by complex and non-convex loss landscapes. One of the popular solutions is Sharpness-Aware Minimization (SAM), which smooths the loss landscape via minimizing the maximized change of training loss when adding a perturbation to the weight. However, we find the indiscriminate perturbation of SAM on all parameters is suboptimal, which also results in excessive computation,~\emph{i.e.}, double the overhead of common optimizers like Stochastic Gradient Descent~(SGD). In this paper, we propose an efficient and effective training scheme coined as Sparse SAM (SSAM), which achieves sparse perturbation by a binary mask. To obtain the sparse mask, we provide two solutions which are based on Fisher information and dynamic sparse training, respectively. In addition, we theoretically prove that SSAM can converge at the same rate as SAM,~\emph{i.e.}, $O(\log T/\sqrt{T})$. Sparse SAM not only has the potential for training acceleration but also smooths the loss landscape effectively. Extensive experimental results on CIFAR10, CIFAR100, and ImageNet-1K confirm the superior efficiency of our method to SAM, and the performance is preserved or even better with a perturbation of merely 50\% sparsity. Code is available at \url{https://github.com/Mi-Peng/Sparse-Sharpness-Aware-Minimization}.

Online image hashing has received increasing research attention recently, which processes large-scale data in a streaming fashion to update the hash functions on-the-fly. To this end, most existing works exploit this problem under a supervised setting, i.e., using class labels to boost the hashing performance, which suffers from the defects in both adaptivity and efficiency: First, large amounts of training batches are required to learn up-to-date hash functions, which leads to poor online adaptivity. Second, the training is time-consuming, which contradicts with the core need of online learning. In this paper, a novel supervised online hashing scheme, termed Fast Class-wise Updating for Online Hashing (FCOH), is proposed to address the above two challenges by introducing a novel and efficient inner product operation. To achieve fast online adaptivity, a class-wise updating method is developed to decompose the binary code learning and alternatively renew the hash functions in a class-wise fashion, which well addresses the burden on large amounts of training batches. Quantitatively, such a decomposition further leads to at least 75 percent storage saving. To further achieve online efficiency, we propose a semi-relaxation optimization, which accelerates the online training by treating different binary constraints independently. Without additional constraints and variables, the time complexity is significantly reduced. Such a scheme is also quantitatively shown to well preserve past information during updating hashing functions. We have quantitatively demonstrated that the collective effort of class-wise updating and semi-relaxation optimization provides a superior performance comparing to various state-of-the-art methods, which is verified through extensive experiments on three widely-used datasets.

Visual Question Answering (VQA) has attracted extensive research focus recently. Along with the ever-increasing data scale and model complexity, the enormous training cost has become an emerging challenge for VQA. In this article, we show such a massive training cost is indeed plague. In contrast, a fine-grained design of the learning paradigm can be extremely beneficial in terms of both training efficiency and model accuracy. In particular, we argue that there exist two essential and unexplored issues in the existing VQA training paradigm that randomly samples data in each epoch, namely, the "difficulty diversity" and the "label redundancy". Concretely, "difficulty diversity" refers to the varying difficulty levels of different question types, while "label redundancy" refers to the redundant and noisy labels contained in individual question type. To tackle these two issues, in this article we propose a fine-grained VQA learning paradigm with an actor-critic based learning agent, termed FG-A1C. Instead of using all training data from scratch, FG-A1C includes a learning agent that adaptively and intelligently schedules the most difficult question types in each training epoch. Subsequently, two curriculum learning based schemes are further designed to identify the most useful data to be learned within each inidividual question type. We conduct extensive experiments on the VQA2.0 and VQA-CP v2 datasets, which demonstrate the significant benefits of our approach. For instance, on VQA-CP v2, with less than 75 percent of the training data, our learning paradigms can help the model achieves better performance than using the whole dataset. Meanwhile, we also shows the effectivenesss of our method in guiding data labeling. Finally, the proposed paradigm can be seamlessly integrated with any cutting-edge VQA models, without modifying their structures.

Unsupervised domain adaptation (UDA) aims to address the domain-shift problem between a labeled source domain and an unlabeled target domain. Many efforts have been made to eliminate the mismatch between the distributions of training and testing data by learning domain-invariant representations. However, the learned representations are usually not task-oriented, i.e., being class-discriminative and domain-transferable simultaneously. This drawback limits the flexibility of UDA in complicated open-set tasks where no labels are shared between domains. In this paper, we break the concept of task-orientation into task-relevance and task-irrelevance, and propose a dynamic task-oriented disentangling network (DTDN) to learn disentangled representations in an end-to-end fashion for UDA. The dynamic disentangling network effectively disentangles data representations into two components: the task-relevant ones embedding critical information associated with the task across domains, and the task-irrelevant ones with the remaining non-transferable or disturbing information. These two components are regularized by a group of task-specific objective functions across domains. Such regularization explicitly encourages disentangling and avoids the use of generative models or decoders. Experiments in complicated, open-set scenarios (retrieval tasks) and empirical benchmarks (classification tasks) demonstrate that the proposed method captures rich disentangled information and achieves superior performance.

Despite considerable progress, image captioning still suffers from the huge difference in quality between easy and hard examples, which is left unexploited in existing methods. To address this issue, we explore the hard example mining in image captioning, and propose a simple yet effective mechanism to instruct the model to pay more attention to hard examples, thereby improving the performance in both general and complex scenarios. We first propose a novel learning strategy, termed Metric-oriented Focal Mechanism (MFM), for hard example mining in image captioning. Differing from the existing strategies for classification tasks, MFM can adopt the generative metrics of image captioning to measure the difficulties of examples, and then up-weight the rewards of hard examples during training. To make MFM applicable to different datasets without tedious parameter tuning, we further introduce an adaptive reward metric called Effective CIDEr (ECIDEr), which considers the data distribution of easy and hard examples during reward estimation. Extensive experiments are conducted on the MS COCO benchmark, and the results show that while maintaining the performance on simple examples, MFM can significantly improve the quality of captions for hard examples. The ECIDEr-based MFM is equipped on the current SOTA method, e.g., DLCT (Luo et al., 2021), which outperforms all existing methods and achieves new state-of-the-art performance on both the off-line and on- line testing, i.e., 134.3 CIDEr for the off-line testing and 136.1 for the on- line testing of MSCOCO. To validate the generalization ability of ECIDEr-based MFM, we also apply it to another dataset, namely Flickr30k, and superior performance gains can also be obtained.

Despite the exciting performance, Transformer is criticized for its excessive parameters and computation cost. However, compressing Transformer remains as an open problem due to its internal complexity of the layer designs, i.e., Multi-Head Attention (MHA) and Feed-Forward Network (FFN). To address this issue, we introduce Group-wise Transformation towards a universal yet lightweight Transformer for vision-and-language tasks, termed as LW-Transformer. LW-Transformer applies Group-wise Transformation to reduce both the parameters and computations of Transformer, while also preserving its two main properties, i.e., the efficient attention modeling on diverse subspaces of MHA, and the expanding-scaling feature transformation of FFN. We apply LW-Transformer to a set of Transformer-based networks, and quantitatively measure them on three vision-and-language tasks and six benchmark datasets. Experimental results show that while saving a large number of parameters and computations, LW-Transformer achieves very competitive performance against the original Transformer networks for vision-and-language tasks. To examine the generalization ability, we apply LW-Transformer to the task of image classification, and build its network based on a recently proposed image Transformer called Swin-Transformer, where the effectiveness can be also confirmed.

Text-to-Image (T2I) synthesis is a challenging task that aims to convert natural language descriptions to real images. It remains an open problem mainly due to the diversity of text descriptions, which poses a huge obstacle in generating vivid and relevant images. Moreover, the existing evaluation metrics in T2I synthesis are mainly used to evaluate the visual quality of the generated images, while the semantic consistency between the two modalities is often ignored. To address these issues, we present a novelKnowledge-Driven Generative Adversarial Network, termed KD-GAN, and a new evaluation system, namedPseudo Turing Test(PTT for short). Concretely, KD-GAN takes a further step in imitating the behavior of human painting,i.e., drawing an image according to reference knowledge. The introduction of reference knowledge in KD-GAN not only improves the quality of the generated images but also enhances the semantic consistency between them and the input texts. In addition, KD-GAN can also greatly avoid some flaws against common sense during image generation,e.g., skiing in the blue sky. The proposed PTT is an important supplement to the existing evaluation system of T2I synthesis. It includes a set of pseudo-experts of different multimedia tasks to evaluate the semantic consistency between the given texts and the generated images. To validate the proposed KD-GAN, we conducted extensive experiments on two benchmark datasets,i.e., Caltech-UCSD Birds (CUB), and MS-COCO (COCO). The experimental results demonstrate that KD-GAN outperforms state-of-the-art methods on IS, FID, and the proposed PTT metrics.11The codes of KD-GAN are at [Online]. Available:https://github.com/pengjunn/KD-GANand the codes and models of PTT are at [Online]. Available:https://github.com/pengjunn/PTT.

We propose a novel network pruning approach by information preserving of pretrained network weights (filters). Network pruning with the information preserving is formulated as a matrix sketch problem, which is efficiently solved by the off-the-shelf frequent direction method. Our approach, referred to as FilterSketch, encodes the second-order information of pretrained weights, which enables the representation capacity of pruned networks to be recovered with a simple fine-tuning procedure. FilterSketch requires neither training from scratch nor data-driven iterative optimization, leading to a several-orders-of-magnitude reduction of time cost in the optimization of pruning. Experiments on CIFAR-10 show that FilterSketch reduces 63.3% of floating-point operations (FLOPs) and prunes 59.9% of network parameters with negligible accuracy cost for ResNet-110. On ILSVRC-2012, it reduces 45.5% of FLOPs and removes 43.0% of parameters with only 0.69% accuracy drop for ResNet-50. Our code and pruned models can be found at https://github.com/lmbxmu/FilterSketch.

Popular network pruning algorithms reduce redundant information by optimizing hand-crafted models, and may cause suboptimal performance and long time in selecting filters. We innovatively introduce adaptive exemplar filters to simplify the algorithm design, resulting in an automatic and efficient pruning approach called EPruner. Inspired by the face recognition community, we use a message-passing algorithm Affinity Propagation on the weight matrices to obtain an adaptive number of exemplars, which then act as the preserved filters. EPruner breaks the dependence on the training data in determining the "important" filters and allows the CPU implementation in seconds, an order of magnitude faster than GPU-based SOTAs. Moreover, we show that the weights of exemplars provide a better initialization for the fine-tuning. On VGGNet-16, EPruner achieves a 76.34%-FLOPs reduction by removing 88.80% parameters, with 0.06% accuracy improvement on CIFAR-10. In ResNet-152, EPruner achieves a 65.12%-FLOPs reduction by removing 64.18% parameters, with only 0.71% top-5 accuracy loss on ILSVRC-2012. Our code is available at https://github.com/lmbxmu/EPruner.

Recent years have witnessed significant progress of person reidentification (reID) driven by expert-designed deep neural network architectures. Despite the remarkable success, such architectures often suffer from high model complexity and time-consuming pretraining process, as well as the mismatches between the image classification-driven backbones and the reID task. To address these issues, we introduce neural architecture search (NAS) into automatically designing person reID backbones, i.e., reID-NAS, which is achieved via automatically searching attention-based network architectures from scratch. Different from traditional NAS approaches that originated for image classification, we design a reID-based search space as well as a search objective to fit NAS for the reID tasks. In terms of the search space, reID-NAS includes a lightweight attention module to precisely locate arbitrary pedestrian bounding boxes, which is automatically added as attention to the reID architectures. In terms of the search objective, reID-NAS introduces a new retrieval objective to search and train reID architectures from scratch. Finally, we propose a hybrid optimization strategy to improve the search stability in reID-NAS. In our experiments, we validate the effectiveness of different parts in reID-NAS, and show that the architecture searched by reID-NAS achieves a new state of the art, with one order of magnitude fewer parameters on three-person reID datasets. As a concomitant benefit, the reliance on the pretraining process is vastly reduced by reID-NAS, which facilitates one to directly search and train a lightweight reID model from scratch.

Domain generalization (DG) offers a preferable real-world setting for Person Re-Identification (Re-ID), which trains a model using multiple source domain datasets and expects it to perform well in an unseen target domain without any model updating. Unfortunately, most DG approaches are designed explicitly for classification tasks, which fundamentally differs from the retrieval task Re-ID. Moreover, existing applications of DG in Re-ID cannot correctly handle the massive variation among Re-ID datasets. In this paper, we identify two fundamental challenges in DG for Person Re-ID: domain-wise variations and identity-wise similarities. To this end, we propose an end-to-end Dual Distribution Alignment Network (DDAN) to learn domain-invariant features with dual-level constraints: the domain-wise adversarial feature learning and the identity-wise similarity enhancement. These constraints effectively reduce the domain-shift among multiple source domains further while agreeing to real-world scenarios. We evaluate our method in a large-scale DG Re-ID benchmark and compare it with various cutting-edge DG approaches. Quantitative results show that DDAN achieves state-of-the-art performance.

With the rapid development of facial manipulation techniques, face forgery has received considerable attention in digital media forensics due to security concerns. Most existing methods formulate face forgery detection as a classification problem and utilize binary labels or manipulated region masks as supervision. However, without considering the correlation between local regions, these global supervisions are insufficient to learn a generalized feature and prone to overfitting. To address this issue, we propose a novel perspective of face forgery detection via local relation learning. Specifically, we propose a Multi-scale Patch Similarity Module (MPSM), which measures the similarity between features of local regions and forms a robust and generalized similarity pattern. Moreover, we propose an RGB-Frequency Attention Module (RFAM) to fuse information in both RGB and frequency domains for more comprehensive local feature representation, which further improves the reliability of the similarity pattern. Extensive experiments show that the proposed method consistently outperforms the state-of-the-arts on widely-used benchmarks. Furthermore, detailed visualization shows the robustness and interpretability of our method.

Transformer-based architectures have shown great success in image captioning, where object regions are encoded and then attended into the vectorial representations to guide the caption decoding. However, such vectorial representations only contain region-level information without considering the global information reflecting the entire image, which fails to expand the capability of complex multi-modal reasoning in image captioning. In this paper, we introduce a Global Enhanced Transformer (termed GET) to enable the extraction of a more comprehensive global representation, and then adaptively guide the decoder to generate high-quality captions. In GET, a Global Enhanced Encoder is designed for the embedding of the global feature, and a Global Adaptive Decoder are designed for the guidance of the caption generation. The former models intra- and inter-layer global representation by taking advantage of the proposed Global Enhanced Attention and a layer-wise fusion module. The latter contains a Global Adaptive Controller that can adaptively fuse the global information into the decoder to guide the caption generation. Extensive experiments on MS COCO dataset demonstrate the superiority of our GET over many state-of-the-arts.

Descriptive region features extracted by object detection networks have played an important role in the recent advancements of image captioning. However, they are still criticized for the lack of contextual information and fine-grained details, which in contrast are the merits of traditional grid features. In this paper, we introduce a novel Dual-Level Collaborative Transformer (DLCT) network to realize the complementary advantages of the two features. Concretely, in DLCT, these two features are first processed by a novel Dual-way Self Attenion (DWSA) to mine their intrinsic properties, where a Comprehensive Relation Attention component is also introduced to embed the geometric information. In addition, we propose a Locality-Constrained Cross Attention module to address the semantic noises caused by the direct fusion of these two features, where a geometric alignment graph is constructed to accurately align and reinforce region and grid features. To validate our model, we conduct extensive experiments on the highly competitive MS-COCO dataset, and achieve new state-of-the-art performance on both local and online test sets, i.e., 133.8% CIDEr on Karpathy split and 135.4% CIDEr on the official split.

In this paper, we propose a domain-general model, termed learning-to-weight (LTW), that guarantees face detection performance across multiple domains, particularly the target domains that are never seen before. However, various face forgery methods cause complex and biased data distributions, making it challenging to detect fake faces in unseen domains. We argue that different faces contribute differently to a detection model trained on multiple domains, making the model likely to fit domain-specific biases. As such, we propose the LTW approach based on the meta-weight learning algorithm, which configures different weights for face images from different domains. The LTW network can balance the model's generalizability across multiple domains. Then, the meta-optimization calibrates the source domain's gradient enabling more discriminative features to be learned. The detection ability of the network is further improved by introducing an intra-class compact loss. Extensive experiments on several commonly used deepfake datasets to demonstrate the effectiveness of our method in detecting synthetic faces. Code and supplemental material are available at https://github.com/skJack/LTW.

One significant factor we expect the video representation learning to capture, especially in contrast with the image representation learning, is the object motion. However, we found that in the current mainstream video datasets, some action categories are highly related with the scene where the action happens, making the model tend to degrade to a solution where only the scene information is encoded. For example, a trained model may predict a video as playing football simply because it sees the field, neglecting that the subject is dancing as a cheerleader on the field. This is against our original intention towards the video representation learning and may bring scene bias on a different dataset that can not be ignored. In order to tackle this problem, we propose to decouple the scene and the motion (DSM) with two simple operations, so that the model attention towards the motion information is better paid. Specifically, we construct a positive clip and a negative clip for each video. Compared to the original video, the positive/negative is motion-untouched/broken but scene-broken/untouched by Spatial Local Disturbance and Temporal Local Disturbance. Our objective is to pull the positive closer while pushing the negative farther to the original clip in the latent space. In this way, the impact of the scene is weakened while the temporal sensitivity of the network is further enhanced. We conduct experiments on two tasks with various backbones and different pre-training datasets, and find that our method surpass the SOTA methods with a remarkable 8.1% and 8.8% improvement towards action recognition task on the UCF101 and HMDB51 datasets respectively using the same backbone.

Channel pruning and tensor decomposition have received extensive attention in convolutional neural network compression. However, these two techniques are traditionally deployed in an isolated manner, leading to significant accuracy drop when pursuing high compression rates. In this paper, we propose a Collaborative Compression (CC) scheme, which joints channel pruning and tensor decomposition to compress CNN models by simultaneously learning the model sparsity and low-rankness. Specifically, we first investigate the compression sensitivity of each layer in the network, and then propose a Global Compression Rate Optimization that transforms the decision problem of compression rate into an optimization problem. After that, we propose multi-step heuristic compression to remove redundant compression units step-by-step, which fully considers the effect of the remaining compression space (i.e., unremoved compression units). Our method demonstrates superior performance gains over previous ones on various datasets and backbone architectures. For example, we achieve 52.9% FLOPs reduction by removing 48.4% parameters on ResNet-50 with only a Top-1 accuracy drop of 0.56% on ImageNet 2012.

Few-shot object detection has made substantial progress by representing novel class objects using the feature representation learned upon a set of base class objects. However, an implicit contradiction between novel class classification and representation is unfortunately ignored. On the one hand, to achieve accurate novel class classification, the distributions of either two base classes must be far away from each other (max-margin). On the other hand, to precisely represent novel classes, the distributions of base classes should be close to each other to reduce the intra-class distance of novel classes (min-margin). In this paper, we propose a class margin equilibrium (CME) approach, with the aim to optimize both feature space partition and novel class reconstruction in a systematic way. CME first converts the few-shot detection problem to the few-shot classification problem by using a fully connected layer to decouple localization features. CME then reserves adequate margin space for novel classes by introducing simple-yet-effective class margin loss during feature learning. Finally, CME pursues margin equilibrium by disturbing the features of novel class instances in an adversarial min-max fashion. Experiments on Pascal VOC and MS-COCO datasets show that CME significantly improves upon two baseline detectors (up to 3 ~ 5% in average), achieving state-of-the-art performance. Code is available at https://github.com/BohaoLee/CME.

Recently, image-to-image translation has made significant progress in achieving both multi-label (i.e., translation conditioned on different labels) and multi-style (i.e., generation with diverse styles) tasks. However, due to the unexplored independence and exclusiveness in the labels, existing endeavors are defeated by involving uncontrolled manipulations to the translation results. In this paper, we propose Hierarchical Style Disentanglement (HiSD) to address this issue. Specifically, we organize the labels into a hierarchical tree structure, in which independent tags, exclusive attributes, and disentangled styles are allocated from top to bottom. Correspondingly, a new translation process is designed to adapt the above structure, in which the styles are identified for controllable translations. Both qualitative and quantitative results on the CelebA-HQ dataset verify the ability of the proposed HiSD. The code has been released at https://github.com/imlixinyang/HiSD.

Panoptic segmentation aims to partition an image to object instances and semantic content for thing and stuff categories, respectively. To date, learning weakly supervised panoptic segmentation (WSPS) with only image-level labels remains unexplored. In this paper, we propose an efficient jointly thing-and-stuff mining (JTSM) framework for WSPS. To this end, we design a novel mask of interest pooling (MoIPool) to extract fixed-size pixel-accurate feature maps of arbitrary-shape segmentations. MoIPool enables a panoptic mining branch to leverage multiple instance learning (MIL) to recognize things and stuff segmentation in a unified manner. We further refine segmentation masks with parallel instance and semantic segmentation branches via self-training, which collaborates the mined masks from panoptic mining with bottom-up object evidence as pseudo-ground-truth labels to improve spatial coherence and contour localization. Experimental results demonstrate the effectiveness of JTSM on PASCAL VOC and MS COCO. As a by-product, we achieve competitive results for weakly supervised object detection and instance segmentation. This work is a first step towards tackling challenge panoptic segmentation task with only image-level labels.

Self-supervised learning has shown great potentials in improving the video representation ability of deep neural networks by getting supervision from the data itself. However, some of the current methods tend to cheat from the background, i.e., the prediction is highly dependent on the video background instead of the motion, making the model vulnerable to background changes. To mitigate the model reliance towards the background, we propose to remove the background impact by adding the background. That is, given a video, we randomly select a static frame and add it to every other frames to construct a distracting video sample. Then we force the model to pull the feature of the distracting video and the feature of the original video closer, so that the model is explicitly restricted to resist the background influence, focusing more on the motion changes. We term our method as Background Erasing (BE). It is worth noting that the implementation of our method is so simple and neat and can be added to most of the SOTA methods without much efforts. Specifically, BE brings 16.4% and 19.1% improvements with MoCo on the severely biased datasets UCF101 and HMDB51, and 14.5% improvement on the less biased dataset Diving48.

Visible-infrared person re-identification (Re-ID) aims to match the pedestrian images of the same identity from different modalities. Existing works mainly focus on alleviating the modality discrepancy by aligning the distributions of features from different modalities. However, nuanced but discriminative information, such as glasses, shoes, and the length of clothes, has not been fully explored, especially in the infrared modality. Without discovering nuances, it is challenging to match pedestrians across modalities using modality alignment solely, which inevitably reduces feature distinctiveness. In this paper, we propose a joint Modality and Pattern Alignment Network (MPANet) to discover cross-modality nuances in different patterns for visible-infrared person Re-ID, which introduces a modality alleviation module and a pattern alignment module to jointly extract discriminative features. Specifically, we first propose a modality alleviation module to dislodge the modality information from the extracted feature maps. Then, We devise a pattern alignment module, which generates multiple pattern maps for the diverse patterns of a person, to discover nuances. Finally, we introduce a mutual mean learning fashion to alleviate the modality discrepancy and propose a center cluster loss to guide both identity learning and nuances discovering. Extensive experiments on the public SYSU-MM01 and RegDB datasets demonstrate the superiority of MPANet over state-of-the-arts.

Recent progress on visual question answering has explored the merits of grid features for vision language tasks. Meanwhile, transformer-based models have shown remarkable performance in various sequence prediction problems. However, the spatial information loss of grid features caused by flattening operation, as well as the defect of the transformer model in distinguishing visual words and non visual words, are still left unexplored. In this paper, we first propose Grid-Augmented (GA) module, in which relative geometry features between grids are incorporated to enhance visual representations. Then, we build a BERT-based language model to extract language context and propose Adaptive-Attention (AA) module on top of a transformer decoder to adaptively measure the contribution of visual and language cues before making decisions for word prediction. To prove the generality of our proposals, we apply the two modules to the vanilla transformer model to build our Relationship-Sensitive Transformer (RSTNet) for image captioning task. The proposed model is tested on the MSCOCO benchmark, where it achieves new state-of-art results on both the Karpathy test split and the online test server. Source code is available at GitHub1.

Understanding the inner workings of deep neural networks (DNNs) is essential to provide trustworthy artificial intelligence techniques for practical applications. Existing studies typically involve linking semantic concepts to units or layers of DNNs, but fail to explain the inference process. In this paper, we introduce neural architecture disentanglement (NAD) to fill the gap. Specifically, NAD learns to disentangle a pre-trained DNN into sub-architectures according to independent tasks, forming information flows that describe the inference processes. We investigate whether, where, and how the disentanglement occurs through experiments conducted with handcrafted and automatically-searched network architectures, on both object-based and scene-based datasets. Based on the experimental results, we present three new findings that provide fresh insights into the inner logic of DNNs. First, DNNs can be divided into sub-architectures for independent tasks. Second, deeper layers do not always correspond to higher semantics. Third, the connection type in a DNN affects how the information flows across layers, leading to different disentanglement behaviors. With NAD, we further explain why DNNs sometimes give wrong predictions. Experimental results show that misclassified images have a high probability of being assigned to task sub-architectures similar to the correct ones. Our code is available at https://github.com/hujiecpp/NAD.

Person Re-Identification (ReID) has achieved remarkable performance along with the deep learning era. However, most approaches carry out ReID only based upon holistic pedestrian regions. In contrast, real-world scenarios involve occluded pedestrians, which provide partial visual appearances and destroy the ReID accuracy. A common strategy is to locate visible body parts by auxiliary model, which however suffers from significant domain gaps and data bias issues. To avoid such problematic models in occluded person ReID, we propose the Occlusion-Aware Mask Network (OAMN). In particular, we incorporate an attention-guided mask module, which requires guidance from labeled occlusion data. To this end, we propose a novel occlusion augmentation scheme that produces diverse and precisely labeled occlusion for any holistic dataset. The proposed scheme suits real-world scenarios better than existing schemes, which consider only limited types of occlusions. We also offer a novel occlusion unification scheme to tackle ambiguity information at the test phase. The above three components enable existing attention mechanisms to precisely capture body parts regardless of the occlusion. Comprehensive experiments on a variety of person ReID benchmarks demonstrate the superiority of OAMN over state-of-the-arts.

Annotation burden has become one of the biggest barriers to semantic segmentation. Approaches based on click-level annotations have therefore attracted increasing attention due to their superior trade-off between supervision and annotation cost. In this paper, we propose seminar learning, a new learning paradigm for semantic segmentation with click-level supervision. The fundamental rationale of seminar learning is to leverage the knowledge from different networks to compensate for insufficient information provided in click-level annotations. Mimicking a seminar, our seminar learning involves a teacher-student and a student-student module, where a student can learn from both skillful teachers and other students. The teacher-student module uses a teacher network based on the exponential moving average to guide the training of the student network. In the student-student module, heterogeneous pseudo-labels are proposed to bridge the transfer of knowledge among students to enhance each other’s performance. Experimental results demonstrate the effectiveness of seminar learning, which achieves the new state-of-the-art performance of 72.51% (mIOU), surpassing previous methods by a large margin of up to 16.88% on the Pascal VOC 2012 dataset.

The decision-based black-box attack means to craft adversarial examples with only the top-1 label of the victim model available. A common practice is to start from a large perturbation and then iteratively reduce it with a deterministic direction and a random one while keeping it adversarial. The limited information obtained from each query and inefficient direction sampling impede attack efficiency, making it hard to obtain a small enough perturbation within a limited number of queries. To tackle this problem, we propose a novel attack method termed Adaptive History-driven Attack (AHA) which gathers information from all historical queries as the prior for current sampling. Moreover, to balance between the deterministic direction and the random one, we dynamically adjust the coefficient according to the ratio of the actual magnitude reduction to the expected one. Such a strategy improves the success rate of queries during optimization, letting adversarial examples move swiftly along the decision boundary. Our method can also integrate with subspace optimization like dimension reduction to further improve efficiency. Extensive experiments on both ImageNet and CelebA datasets demonstrate that our method achieves at least 24.3% lower magnitude of perturbation on average with the same number of queries. Finally, we prove the practical potential of our method by evaluating it on popular defense methods and a real-world system provided by MEGVII Face++.

Weakly supervised instance segmentation (WSIS) with only image-level labels has recently drawn much attention. To date, bottom-up WSIS methods refine discriminative cues from classifiers with sophisticated multi-stage training procedures, which also suffer from inconsistent object boundaries. And top-down WSIS methods are formulated as cascade detection-to-segmentation pipeline, in which the quality of segmentation learning heavily depends on pseudo masks generated from detectors. In this paper, we propose a unified parallel detection-and-segmentation learning (PDSL) framework to learn instance segmentation with only image-level labels, which draws inspiration from both top-down and bottom-up instance segmentation approaches. The detection module is the same as the typical design of any weakly supervised object detection, while the segmentation module leverages self-supervised learning to model class-agnostic foreground extraction, following by self-training to refine class-specific segmentation. We further design instance-activation correlation module to improve the coherence between detection and segmentation branches. Extensive experiments verify that the proposed method outperforms baselines and achieves the state-of-the-art results on PASCAL VOC and MS COCO.

Due to the superior ability of global dependency modeling, Transformer and its variants have become the primary choice of many vision-and-language tasks. However, in tasks like Visual Question Answering (VQA) and Referring Expression Comprehension (REC), the multimodal prediction often requires visual information from macro- to micro-views. Therefore, how to dynamically schedule the global and local dependency modeling in Transformer has become an emerging issue. In this paper, we propose an example-dependent routing scheme called TRAnsformer Routing (TRAR) to address this issue1. Specifically, in TRAR, each visual Transformer layer is equipped with a routing module with different attention spans. The model can dynamically select the corresponding attentions based on the output of the previous inference step, so as to formulate the optimal routing path for each example. Notably, with careful designs, TRAR can reduce the additional computation and memory overhead to almost negligible. To validate TRAR, we conduct extensive experiments on five benchmark datasets of VQA and REC, and achieve superior performance gains than the standard Transformers and a bunch of state-of-the-art methods.

Based on the relaxed search space, differential architecture search (DARTS) is efficient in searching for a high-performance architecture. However, the unbalanced competition among operations that have different trainable parameters causes the model collapse. Besides, the inconsistent structures in the search and retraining stages causes cross-stage evaluation to be unstable. In this paper, we call these issues as an operation gap and a structure gap in DARTS. To shrink these gaps, we propose to induce equalized and consistent optimization in differentiable architecture search (EC-DARTS). EC-DARTS decouples different operations based on their categories to optimize the operation weights so that the operation gap between them is shrinked. Besides, we introduce an induced structural transition to bridge the structure gap between the model structures in the search and retraining stages. Extensive experiments on CIFAR10 and ImageNet demonstrate the effectiveness of our method. Specifically, on CIFAR10, we achieve a test error of 2.39%, while only 0.3 GPU days on NVIDIA TITAN V. On ImageNet, our method achieves a top-1 error of 23.6% under the mobile setting.

In this work, we propose a high fidelity face swapping method, called HifiFace, which can well preserve the face shape of the source face and generate photo-realistic results. Unlike other existing face swapping works that only use face recognition model to keep the identity similarity, we propose 3D shape-aware identity to control the face shape with the geometric supervision from 3DMM and 3D face reconstruction method. Meanwhile, we introduce the Semantic Facial Fusion module to optimize the combination of encoder and decoder features and make adaptive blending, which makes the results more photo-realistic. Extensive experiments on faces in the wild demonstrate that our method can preserve better identity, especially on the face shape, and can generate more photo-realistic results than previous state-of-the-art methods. Code is available at: https://johann.wang/HifiFace

Weakly supervised object localization (WSOL) has gained recent popularity, which seeks to train localizers with only image-level labels. However, due to relying heavily on classification objective for training, prevailing WSOL methods only localize discriminative parts of object, ignoring other useful information, such as the wings of a bird, and suffer from severe rotation variations. Moreover, learning object localization imposes CNNs to attend non-salient regions under weak supervision, which may negatively influence image classification results. To address these challenges, this paper proposes a novel end-to-end Excitation-Expansion network, coined as E$^2$Net, to localize entire objects with only image-level labels, which served as the base of most multimedia tasks. The proposed E$^2$Net consists of two key components: Maxout-Attention Excitation (MAE) and Orientation-Sensitive Expansion (OSE). Firstly, MAE module aims to activate non-discriminative localization features while simultaneously recovering discriminative classification cues. To this end, we couple erasing strategy with maxout learning efficiently to facilitate entire-object localization without hurting classification accuracy. Secondly, to address rotation variations, the proposed OSE module expands less salient object parts along with all possible orientations. Particularly, OSE module dynamically combines selective attention banks from various orientated expansions of receptive-field, which introduces additional multi-parallel localization heads. Extensive experiments on ILSVRC 2012 and CUB-200-2011 demonstrate that the proposed E$^2$Net outperforms the previous state-of-the-art WSOL methods and also significantly improves classification performance.

Text recognition is the key pillar for many real-world multimedia applications. Existing text recognition approaches focus on recognizing isolated instances, whose text fields are visually separated and have no interference with each other. Moreover, these approaches cannot handle overlapped instances that often appear in sheets like invoices, receipts and math exercises, where printed templates are generated beforehand and extra contents are added afterward on existing texts. In this paper, we aim to tackle this problem by proposing RecycleNet, which automatically extracts and reconstructs overlapped instances by fully recycling the intersecting pixels that used to be obstacles for recognition. RecycleNet parallels to existing recognition systems, and serves as a plug-and-play module to boost recognition performance with zero-effort. We also released an OverlapText-500 dataset, which helps to boost the design of better overlapped text recovery and recognition solutions.

We investigate the challenging problem of table structure recognition in this work. Many recent methods adopt graph-based context aggregator with strong inductive bias to reason sparse contextual relationships of table elements. However, the strong constraints may be too restrictive to represent the complicated table relationships. In order to learn more appropriate inductive bias from data, we try to introduce Transformer as context aggregator in this work. Nevertheless, Transformer taking dense context as input requires larger scale data and may suffer from unstable training procedure due to the weakening of inductive bias. To overcome the above limitations, we in this paper design a FLAG (FLexible context AGgregator), which marries Transformer with graph-based context aggregator in an adaptive way. Based on FLAG, an end-to-end framework requiring no extra meta-data or OCR information, termed FLAG-Net, is proposed to flexibly modulate the aggregation of dense context and sparse one for the relational reasoning of table elements. We investigate the modulation pattern in FLAG and show what contextual information is focused, which is vital for recognizing table structure. Extensive experimental results on benchmarks demonstrate the performance of our proposed FLAG-Net surpasses other compared methods by a large margin.

Natural image matting estimates the alpha values of unknown regions in the trimap. Recently, deep learning based methods propagate the alpha values from the known regions to unknown regions according to the similarity between them. However, we find that more than 50% pixels in the unknown regions cannot be correlated to pixels in known regions due to the limitation of small effective reception fields of common convolutional neural networks, which leads to inaccurate estimation when the pixels in the unknown regions cannot be inferred only with pixels in the reception fields. To solve this problem, we propose Long-Range Feature Propagating Network (LFPNet), which learns the long-range context features outside the reception fields for alpha matte estimation. Specifically, we first design the propagating module which extracts the context features from the downsampled image. Then, we present Center-Surround Pyramid Pooling (CSPP) that explicitly propagates the context features from the surrounding context image patch to the inner center image patch. Finally, we use the matting module which takes the image, trimap and context features to estimate the alpha matte. Experimental results demonstrate that the proposed method performs favorably against the state-of-the-art methods on the AlphaMatting and Adobe Image Matting datasets.

Neural network pruning has shown promising performance in reducing computational complexity and facilitate the deployment of deep neural networks on resource-limited edge devices. Most existing pruning methods focus on the indicators of the filter's weight, gradient, or feature map and regard the weak or similar filters as network redundancy. In contrast, the representation of discriminative power is also a fundamental attribute that analog neural networks to have extraordinary performance in various tasks. However, such representation is neglected in existing works. Alternatively, we propose a novel filter pruning strategy via class-wise discriminative power (CDP). Unlike the previous methods, CDP treats the filters that always yield large or small activation values as redundant and reserves the filters that show different magnitudes in activations as they yield high discriminative power. We further propose to obtain such discriminative power by employing the widely-used Term Frequency-Inverse Document Frequency (TF-IDF) on feature representations across classes. Specifically, the output of a filter is considered as a word, and the whole feature map is considered as a document. Then, TF-IDF is used to generate the relevant score between words and all documents. If a filter has low TF-IDF scores is less discriminate and can be pruned. Thus, the filters with high TF-IDF scores are reserved. To our best knowledge, this is the first work that prunes neural networks through class-wise discriminative power and measures such power by introducing TF-IDF in feature representation among different classes. Without any iterative process, CDP achieves better compression trade-offs comparing to the state-of-the-art compression algorithms. For instance, in VGG-16, we achieve a 68.05%-FLOPs reduction, with a 94.86% Top-1 accuracy on CIFAR-10. Specifically, we compress a 90.12%-FLOPs reduction VGG-16, even retains 93.30% Top-1 accuracy on CIFAR-10. The code is available at https://github.com/Tianshuo-Xu/CDP-Towards-Optimal-Filter-Pruning-via-Class-wise-Discriminative-Power.git

Recently, a series of algorithms have been explored for GAN compression, which aims to reduce tremendous computational overhead and memory usages when deploying GANs on resource-constrained edge devices. However, most of the existing GAN compression work only focuses on how to compress the generator, while fails to take the discriminator into account. In this work, we revisit the role of discriminator in GAN compression and design a novel generator-discriminator cooperative compression scheme for GAN compression, termed GCC. Within GCC, a selective activation discriminator automatically selects and activates convolutional channels according to a local capacity constraint and a global coordination constraint, which help maintain the Nash equilibrium with the lightweight generator during the adversarial training and avoid mode collapse. The original generator and discriminator are also optimized from scratch, to play as a teacher model to progressively refine the pruned generator and the selective activation discriminator. A novel online collaborative distillation scheme is designed to take full advantage of the intermediate feature of the teacher generator and discriminator to further boost the performance of the lightweight generator. Extensive experiments on various GAN-based generation tasks demonstrate the effectiveness and generalization of GCC. Among them, GCC contributes to reducing 80% computational costs while maintains comparable performance in image translation tasks.

This paper reports the results and post-challenge analyses of ChaLearn's AutoDL challenge series, which helped sorting out a profusion of AutoML solutions for Deep Learning (DL) that had been introduced in a variety of settings, but lacked fair comparisons. All input data modalities (time series, images, videos, text, tabular) were formatted as tensors and all tasks were multi-label classification problems. Code submissions were executed on hidden tasks, with limited time and computational resources, pushing solutions that get results quickly. In this setting, DL methods dominated, though popular Neural Architecture Search (NAS) was impractical. Solutions relied on fine-tuned pre-trained networks, with architectures matching data modality. Post-challenge tests did not reveal improvements beyond the imposed time limit. While no component is particularly original or novel, a high level modular organization emerged featuring a "meta-learner", "data ingestor", "model selector", "model/learner", and "evaluator". This modularity enabled ablation studies, which revealed the importance of (off-platform) meta-learning, ensembling, and efficient data management. Experiments on heterogeneous module combinations further confirm the (local) optimality of the winning solutions. Our challenge legacy includes an ever-lasting benchmark (http://autodl.chalearn.org), the open-sourced code of the winners, and a free "AutoDL self-service."

Neural architecture search (NAS) has achieved unprecedented performance in various computer vision tasks. However, most existing NAS methods are defected in search efficiency and model generalizability. In this paper, we propose a novel NAS framework, termed MIGO-NAS, with the aim to guarantee the efficiency and generalizability in arbitrary search spaces. On the one hand, we formulate the search space as a multivariate probabilistic distribution, which is then optimized by a novel multivariate information-geometric optimization (MIGO). By approximating the distribution with a sampling, training, and testing pipeline, MIGO guarantees the memory efficiency, training efficiency, and search flexibility. Besides, MIGO is the first time to decrease the estimation error of natural gradient in multivariate distribution. On the other hand, for a set of specific constraints, the neural architectures are generated by a novel dynamic programming network generation (DPNG), which significantly reduces the training cost under various hardware environments. Experiments validate the advantages of our approach over existing methods by establishing a superior accuracy and efficiency i.e., 2.39 test error on CIFAR-10 benchmark and 21.7 on ImageNet benchmark, with only 1.5 GPU hours and 96 GPU hours for searching, respectively. Besides, the searched architectures can be well generalize to computer vision tasks including object detection and semantic segmentation, i.e., 25× FLOPs compression, with 6.4 mAP gain over Pascal VOC dataset, and 29.9× FLOPs compression, with only 1.41 percent performance drop over Cityscapes dataset. The code is publicly available.

Automated machine learning (AutoML) has achieved remarkable progress on various tasks, which is attributed to its minimal involvement of manual feature and model designs. However, most of existing AutoML pipelines only touch parts of the full machine learning pipeline, e.g., neural architecture search or optimizer selection. This leaves potentially important components such as data cleaning and model ensemble out of the optimization, and still results in considerable human involvement and suboptimal performance. The main challenges lie in the huge search space assembling all possibilities over all components, as well as the generalization ability over different tasks like image, text, and tabular etc. In this paper, we present a first-of-its-kind fully AutoML pipeline, to comprehensively automate data preprocessing, feature engineering, model generation/selection/training and ensemble for an arbitrary dataset and evaluation metric. Our innovation lies in the comprehensive scope of a learning pipeline, with a novel "life-long" knowledge anchor design to fundamentally accelerate the search over the full search space. Such knowledge anchors record detailed information of pipelines and integrates them with an evolutionary algorithm for joint optimization across components. Experiments demonstrate that the result pipeline achieves state-of-the-art performance on multiple datasets and modalities. Specifically, the proposed framework was extensively evaluated in the NeurIPS 2019 AutoDL challenge, and won the only champion with a significant gap against other approaches, on all the image, video, speech, text and tabular tracks.

Computational prediction of facial aesthetics has attracted ever-increasing research focus, which has wide range of prospects in multimedia applications. The key challenge lies in extracting discriminative and perception-aware features to characterize the facial beautifulness. To this end, the existing schemes simply adopt a direct feature mapping, which relies on handcraft-designed low-level features that cannot reflect human-level aesthetic perception. In this paper, we present a systematic framework towards designing biology-inspired, discriminative representation for facial aesthetic prediction. First, we design a group of biological experiments that adopt eye tracker to identify spatial regions of interest during the facial aesthetic judgments of subjects, which forms a Bio-inspired Facial Aesthetic Ontology (Bio-FAO) and is made public available. Second, we adopt the cutting-edge convolutional neural network to train a set of Bio-inspired Attribute features, termed Bio-AttriBank, which forms a mid-level interpretable representation corresponding to the aforementioned Bio-FAO. For a given image, the facial aesthetic prediction is then formulated as a classification problem over the Bio-AttriBank descriptor responses, which well bridges the affective gap, and provides explainable evidences on why/how a face is beautiful or not. We have carried out extensive experiments on both JAFFE and FaceWarehouse datasets, with comparisons to a set of state-of-the-art and alternative approaches. Superior performance gains in the experiments have demonstrated the merits of the proposed scheme.

Deep learning-based person re-identification (Re-ID) has made great progress and achieved high performance recently. In this paper, we make the first attempt to examine the vulnerability of current person Re-ID models against a dangerous attack method, i.e., the universal adversarial perturbation (UAP) attack, which has been shown to fool classification models with a little overhead. We propose a more universal adversarial perturbation (MUAP) method for both image-agnostic and model-insensitive person Re-ID attack. Firstly, we adopt a list-wise attack objective function to disrupt the similarity ranking list directly. Secondly, we propose a model-insensitive mechanism for cross-model attack. Extensive experiments show that the proposed attack approach achieves high attack performance and outperforms other state of the arts by large margin in cross-model scenario. The results also demonstrate the vulnerability of current Re-ID models to MUAP and further suggest the need of designing more robust Re-ID models.

It is known that media outlets, such as CNN and FOX, have intrinsic political bias that is reflected in their news reports. The computational prediction of such bias has broad application prospects. However, the prediction is difficult via directly analyzing the news content without high-level context. In contrast, social signals (e.g., the network structure of media followers) provide inspiring cues to uncover such bias. In this article, we realize the first attempt of predicting the latent bias of media outlets by analyzing their social network structures. In particular, we address two key challenges:network sparsityandlabel sparsity. The network sparsity refers to the partial sampling of the entire follower network in practical analysis and computing, whereas the label sparsity refers to the difficulty of annotating sufficient labels to train the prediction model. To cope with the network sparsity, we propose a hybrid sampling strategy to construct a training corpus that contains network information from micro to macro views. Based on this training corpus, a semi-supervised network embedding approach is proposed to learn low-dimensional yet effective network representations. To deal with the label sparsity, we adopt a graph-based label propagation scheme to supplement the missing links and augment label information for model training. The preceding two steps are iteratively optimized to reinforce each other. We further collect a large-scale dataset containing social networks of 10 media outlets together with about 300,000 followers and more than 5 million connections. Over this dataset, we compare our model to a range of state of the art. Superior performance gains demonstrate the merits of the proposed approach. More importantly, the experimental results and analyses confirm the validity of our approach for the computerized prediction of media bias.

Vehicle Re-Identification targets at searching for vehicle instances of the same identity with a given query. It has gained an increasing attention recently with a wide application prospects in video surveillance and intelligent transportation. The main challenge lies in how to distinguish the subtle differences between different vehicles, while capturing the slight similarity between the instance of the same vehicle in different viewpoints or illuminations. To this end, most existing methods focus on learning discriminative global representations, which leave the unique local details such as stickers, inspection labels and driver wearing being unexploited. In this paper, we present a novel coarse-to-fine scheme that aggregates global and local visual representations to boost the retrieval accuracy. Specifically, a multi-task learning framework combining an Attribute Learning branch and a Deep Ranking branch (termed ALDR) is first adopted to learn robust global features, which produces an initial ranking list. Then the local similarities between image patches in the initial ranking list and in the query is computed via a Multi-Channel and Multi-Scale Siamese network (termed MCMS-Siam). Finally, the retrieval result is returned after re-ranking the initial list according to such a combination of global and local similarities. Experimental results on the widely-used VehicleID dataset and VECH-WILD dataset demonstrate the merits of the proposed method over the state-of-the-art methods.

Neural architecture search (NAS) can have a significant impact in computer vision by automatically designing optimal neural network architectures for various tasks. A variant, binarized neural architecture search (BNAS), with a search space of binarized convolutions, can produce extremely compressed models. Unfortunately, this area remains largely unexplored. BNAS is more challenging than NAS due to the learning inefficiency caused by optimization requirements and the huge architecture space. To address these issues, we introduce channel sampling and operation space reduction into a differentiable NAS to significantly reduce the cost of searching. This is accomplished through a performance-based strategy used to abandon less potential operations. Two optimization methods for binarized neural networks are used to validate the effectiveness of our BNAS. Extensive experiments demonstrate that the proposed BNAS achieves a performance comparable to NAS on both CIFAR and ImageNet databases. An accuracy of 96.53% vs. 97.22% is achieved on the CIFAR-10 dataset, but with a significantly compressed model, and a 40% faster search than the state-of-the-art PC-DARTS.

Generating temporal action proposals remains a very challenging problem, where the main issue lies in predicting precise temporal proposal boundaries and reliable action confidence in long and untrimmed real-world videos. In this paper, we propose an efficient and unified framework to generate temporal action proposals named Dense Boundary Generator (DBG), which draws inspiration from boundary-sensitive methods and implements boundary classification and action completeness regression for densely distributed proposals. In particular, the DBG consists of two modules: Temporal boundary classification (TBC) and Action-aware completeness regression (ACR). The TBC aims to provide two temporal boundary confidence maps by low-level two-stream features, while the ACR is designed to generate an action completeness score map by high-level action-aware features. Moreover, we introduce a dual stream BaseNet (DSB) to encode RGB and optical flow information, which helps to capture discriminative boundary and actionness features. Extensive experiments on popular benchmarks ActivityNet-1.3 and THUMOS14 demonstrate the superiority of DBG over the state-of-the-art proposal generator (e.g., MGG and BMN).

Visual aesthetic assessment has been an active research field for decades. Although latest methods have achieved promising performance on benchmark datasets, they typically rely on a large number of manual annotations including both aesthetic labels and related image attributes. In this paper, we revisit the problem of image aesthetic assessment from the self-supervised feature learning perspective. Our motivation is that a suitable feature representation for image aesthetic assessment should be able to distinguish different expert-designed image manipulations, which have close relationships with negative aesthetic effects. To this end, we design two novel pretext tasks to identify the types and parameters of editing operations applied to synthetic instances. The features from our pretext tasks are then adapted for a one-layer linear classifier to evaluate the performance in terms of binary aesthetic classification. We conduct extensive quantitative experiments on three benchmark datasets and demonstrate that our approach can faithfully extract aesthetics-aware features and outperform alternative pretext schemes. Moreover, we achieve comparable results to state-of-the-art supervised methods that use 10 million labels from ImageNet.

Person re-identification (re-ID), is a challenging task due to the high variance within identity samples and imaging conditions. Although recent advances in deep learning have achieved remarkable accuracy in settled scenes, i.e., source domain, few works can generalize well on the unseen target domain. One popular solution is assigning unlabeled target images with pseudo labels by clustering, and then retraining the model. However, clustering methods tend to introduce noisy labels and discard low confidence samples as outliers, which may hinder the retraining process and thus limit the generalization ability. In this study, we argue that by explicitly adding a sample filtering procedure after the clustering, the mined examples can be much more efficiently used. To this end, we design an asymmetric co-teaching framework, which resists noisy labels by cooperating two models to select data with possibly clean labels for each other. Meanwhile, one of the models receives samples as pure as possible, while the other takes in samples as diverse as possible. This procedure encourages that the selected training samples can be both clean and miscellaneous, and that the two models can promote each other iteratively. Extensive experiments show that the proposed framework can consistently benefit most clustering based methods, and boost the state-of-the-art adaptation accuracy. Our code is available at https://github.com/FlyingRoastDuck/ACT_AAAI20.

Employing attention mechanisms to model both global and local features as a final pedestrian representation has become a trend for person re-identification (Re-ID) algorithms. A potential limitation of these methods is that they focus on the most salient features, but the re-identification of a person may rely on diverse clues masked by the most salient features in different situations, e.g., body, clothes or even shoes. To handle this limitation, we propose a novel Salience-guided Cascaded Suppression Network (SCSN) which enables the model to mine diverse salient features and integrate these features into the final representation by a cascaded manner. Our work makes the following contributions: (i) We observe that the previously learned salient features may hinder the network from learning other important information. To tackle this limitation, we introduce a cascaded suppression strategy, which enables the network to mine diverse potential useful features that be masked by the other salient features stage-by-stage and each stage integrates different feature embedding for the last discriminative pedestrian representation. (ii) We propose a Salient Feature Extraction (SFE) unit, which can suppress the salient features learned in the previous cascaded stage and then adaptively extracts other potential salient feature to obtain different clues of pedestrians. (iii) We develop an efficient feature aggregation strategy that fully increases the network’s capacity for all potential salience features. Finally, experimental results demonstrate that our proposed method outperforms the state-of-the-art methods on four large-scale datasets. Especially, our approach exceeds the current best method by over 7% on the CUHK03 dataset.

Almost all adversarial attacks in computer vision are aimed at pre-known object categories, which could be offline trained for generating perturbations. But as for visual object tracking, the tracked target categories are normally unknown in advance. However, the tracking algorithms also have potential risks of being attacked, which could be maliciously used to fool the surveillance systems. Meanwhile, it is still a challenging task that adversarial attacks on tracking since it has the free-model tracked target. Therefore, to help draw more attention to the potential risks, we study adversarial attacks on tracking algorithms. In this paper, we propose a novel one-shot adversarial attack method to generate adversarial examples for free-model single object tracking, where merely adding slight perturbations on the target patch in the initial frame causes state-of-the-art trackers to lose the target in subsequent frames. Specifically, the optimization objective of the proposed attack consists of two components and leverages the dual attention mechanisms. The first component adopts a targeted attack strategy by optimizing the batch confidence loss with confidence attention while the second one applies a general perturbation strategy by optimizing the feature loss with channel attention. Experimental results show that our approach can significantly lower the accuracy of the most advanced Siamese network-based trackers on three benchmarks.

Most of the existing trackers usually rely on either a multi-scale searching scheme or pre-defined anchor boxes to accurately estimate the scale and aspect ratio of a target. Unfortunately, they typically call for tedious and heuristic configurations. To address this issue, we propose a simple yet effective visual tracking framework (named Siamese Box Adaptive Network, SiamBAN) by exploiting the expressive power of the fully convolutional network (FCN). SiamBAN views the visual tracking problem as a parallel classification and regression problem, and thus directly classifies objects and regresses their bounding boxes in a unified FCN. The no-prior box design avoids hyper-parameters associated with the candidate boxes, making SiamBAN more flexible and general. Extensive experiments on visual tracking benchmarks including VOT2018, VOT2019, OTB100, NFS, UAV123, and LaSOT demonstrate that SiamBAN achieves state-of-the-art performance and runs at 40 FPS, confirming its effectiveness and efficiency. The code will be available at https://github.com/hqucv/siamban.

Generating adversarial examples in a black-box setting retains a significant challenge with vast practical application prospects. In particular, existing black-box attacks suffer from the need for excessive queries, as it is non-trivial to find an appropriate direction to optimize in the high-dimensional space. In this paper, we propose Projection & Probability-driven Black-box Attack (PPBA) to tackle this problem by reducing the solution space and providing better optimization. For reducing the solution space, we first model the adversarial perturbation optimization problem as a process of recovering frequency-sparse perturbations with compressed sensing, under the setting that random noise in the low-frequency space is more likely to be adversarial. We then propose a simple method to construct a low-frequency constrained sensing matrix, which works as a plug-and-play projection matrix to reduce the dimensionality. Such a sensing matrix is shown to be flexible enough to be integrated into existing methods like NES and BanditsTD. For better optimization, we perform a random walk with a probability-driven strategy, which utilizes all queries over the whole progress to make full use of the sensing matrix for a less query budget. Extensive experiments show that our method requires at most 24% fewer queries with a higher attack success rate compared with state-of-the-art approaches. Finally, the attack method is evaluated on the real-world online service, i.e., Google Cloud Vision API, which further demonstrates our practical potentials.

Neural network pruning offers a promising prospect to facilitate deploying deep neural networks on resource-limited devices. However, existing methods are still challenged by the training inefficiency and labor cost in pruning designs, due to missing theoretical guidance of non-salient network components. In this paper, we propose a novel filter pruning method by exploring the High Rank of feature maps (HRank). Our HRank is inspired by the discovery that the average rank of multiple feature maps generated by a single filter is always the same, regardless of the number of image batches CNNs receive. Based on HRank, we develop a method that is mathematically formulated to prune filters with low-rank feature maps. The principle behind our pruning is that low-rank feature maps contain less information, and thus pruned results can be easily reproduced. Besides, we experimentally show that weights with high-rank feature maps contain more important information, such that even when a portion is not updated, very little damage would be done to the model performance. Without introducing any additional constraints, HRank leads to significant improvements over the state-of-the-arts in terms of FLOPs and parameters reduction, with similar accuracies. For example, with ResNet-110, we achieve a 58.2%-FLOPs reduction by removing 59.2% of the parameters, with only a small loss of 0.14% in top-1 accuracy on CIFAR-10. With Res-50, we achieve a 43.8%-FLOPs reduction by removing 36.7% of the parameters, with only a loss of 1.17% in the top-1 accuracy on ImageNet. The codes can be available at https://github.com/lmbxmu/HRank.

Referring expression comprehension (REC) and segmentation (RES) are two highly-related tasks, which both aim at identifying the referent according to a natural language expression. In this paper, we propose a novel Multi-task Collaborative Network (MCN) to achieve a joint learning of REC and RES for the first time. In MCN, RES can help REC to achieve better language-vision alignment, while REC can help RES to better locate the referent. In addition, we address a key challenge in this multi-task setup, i.e., the prediction conflict, with two innovative designs namely, Consistency Energy Maximization (CEM) and Adaptive Soft Non-Located Suppression (ASNLS). Specifically, CEM enables REC and RES to focus on similar visual regions by maximizing the consistency energy between two tasks. ASNLS supresses the response of unrelated regions in RES based on the prediction of REC. To validate our model, we conduct extensive experiments on three benchmark datasets of REC and RES, i.e., RefCOCO, RefCOCO+ and RefCOCOg. The experimental results report the significant performance gains of MCN over all existing methods, i.e., up to +7.13% for REC and +11.50% for RES over SOTA, which well confirm the validity of our model for joint REC and RES learning.

This paper proposes a new learning paradigm called filter grafting, which aims to improve the representation capability of Deep Neural Networks (DNNs). The motivation is that DNNs have unimportant (invalid) filters (e.g., l1norm close to 0). These filters limit the potential of DNNs since they are identified as having little effect on the network. While filter pruning removes these invalid filters for efficiency consideration, filter grafting re-activates them from an accuracy boosting perspective. The activation is processed by grafting external information (weights) into invalid filters. To better perform the grafting process, we develop an entropy-based criterion to measure the information of filters and an adaptive weighting strategy for balancing the grafted information among networks. After the grafting operation, the network has very few invalid filters compared with its untouched state, empowering the model with more representation capacity. We also perform extensive experiments on the classification and recognition tasks to show the superiority of our method. For example, the grafted MobileNetV2 outperforms the non-grafted MobileNetV2 by about 7 percent on CIFAR-100 dataset.

We investigate the emerging task of learning object detectors with sole image-level labels on the web without requiring any other supervision like precise annotations or additional images from well-annotated benchmark datasets. Such a task, termed as fully webly supervised object detection, is extremely challenging, since image-level labels on the web are always noisy, leading to poor performance of the learned detectors. In this work, we propose an end-to-end framework to jointly learn webly supervised detectors and reduce the negative impact of noisy labels. Such noise is heterogeneous, which is further categorized into two types, namely background noise and foreground noise. Regarding the background noise, we propose a residual learning structure incorporated with weakly supervised detection, which decomposes background noise and models clean data. To explicitly learn the residual feature between clean data and noisy labels, we further propose a spatially-sensitive entropy criterion, which exploits the conditional distribution of detection results to estimate the confidence of background categories being noise. Regarding the foreground noise, a bagging-mixup learning is introduced, which suppresses foreground noisy signals from incorrectly labelled images, whilst maintaining the diversity of training data. We evaluate the proposed approach on popular benchmark datasets by training detectors on web images, which are retrieved by the corresponding category tags from photo-sharing sites. Extensive experiments show that our method achieves significant improvements over the state-of-the-art methods.

Domain adaptive person re-identification (re-ID) is a challenging task, especially when person identities in target domains are unknown. Existing methods attempt to address this challenge by transferring image styles or aligning feature distributions across domains, whereas the rich unlabeled samples in target domains are not sufficiently exploited. This paper presents a novel augmented discriminative clustering (AD-Cluster) technique that estimates and augments person clusters in target domains and enforces the discrimination ability of re-ID models with the augmented clusters. AD-Cluster is trained by iterative density-based clustering, adaptive sample augmentation, and discriminative feature learning. It learns an image generator and a feature encoder which aim to maximize the intra-cluster diversity in the sample space and minimize the intra-cluster distance in the feature space in an adversarial min-max manner. Finally, AD-Cluster increases the diversity of sample clusters and improves the discrimination capability of re-ID models greatly. Extensive experiments over Market-1501 and DukeMTMC-reID show that AD-Cluster outperforms the state-of-the-art with large margins.

Neural architecture search (NAS) remains a challenging problem, which is attributed to the indispensable and time-consuming component of performance estimation (PE). In this paper, we provide a novel yet systematic rethinking of PE in a resource constrained regime, termed budgeted PE (BPE), which precisely and effectively estimates the performance of an architecture sampled from an architecture space. Since searching an optimal BPE is extremely time-consuming as it requires to train a large number of networks for evaluation, we propose a Minimum Importance Pruning (MIP) approach. Given a dataset and a BPE search space, MIP estimates the importance of hyper-parameters using random forest and subsequently prunes the minimum one from the next iteration. In this way, MIP effectively prunes less important hyper-parameters to allocate more computational resource on more important ones, thus achieving an effective exploration. By combining BPE with various search algorithms including reinforcement learning, evolution algorithm, random search, and differentiable architecture search, we achieve 1, 000× of NAS speed up with a negligible performance drop comparing to the SOTA.

Conventional learning methods simplify the bilinear model by regarding two intrinsically coupled factors independently, which degrades the optimization procedure. One reason lies in the insufficient training due to the asynchronous gradient descent, which results in vanishing gradients for the coupled variables. In this paper, we introduce a Cogradient Descent algorithm (CoGD) to address the bilinear problem, based on a theoretical framework to coordinate the gradient of hidden variables via a projection function. We solve one variable by considering its coupling relationship with the other, leading to a synchronous gradient descent to facilitate the optimization procedure. Our algorithm is applied to solve problems with one variable under the sparsity constraint, which is widely used in the learning paradigm. We validate our CoGD considering an extensive set of applications including image reconstruction, inpainting, and network pruning. Experiments show that it improves the state-of-the-art by a significant margin.

In this paper, we focus on Universal Adversarial Perturbations (UAP) attack on state-of-the-art person re-identification (Re-ID) methods. Existing UAP methods usually compute a perturbation image and add it to the images of interest. Such a simple constant form greatly limits the attack power. To address this problem, we extend the formulation of UAP to a polynomial form and propose the Polynomial Universal Adversarial Perturbation (PUAP). Unlike traditional UAP methods which only rely on the additive perturbation signal, the proposed PUAP consists of both an additive perturbation and a multiplicative modulation factor. The additive perturbation produces the fundamental component of the signal, while the multiplicative factor modulates the perturbation signal in line with the unit impulse pattern of the input image. Moreover, we introduce a Pearson correlation coefficient loss to generate universal perturbations, for disrupting the outputs of person Re-ID models. Extensive experiments on DukeMTMC-reID, Market-1501, and MARS show that the proposed method can efficiently improve the attack performance, especially when the magnitude of UAP is constrained to a relatively small value.

Channel pruning is among the predominant approaches to compress deep neural networks. To this end, most existing pruning methods focus on selecting channels (filters) by importance/optimization or regularization based on rule-of-thumb designs, which defects in sub-optimal pruning. In this paper, we propose a new channel pruning method based on artificial bee colony algorithm (ABC), dubbed as ABCPruner, which aims to efficiently find optimal pruned structure, i.e., channel number in each layer, rather than selecting "important" channels as previous works did. To solve the intractably huge combinations of pruned structure for deep networks, we first propose to shrink the combinations where the preserved channels are limited to a specific space, thus the combinations of pruned structure can be significantly reduced. And then, we formulate the search of optimal pruned structure as an optimization problem and integrate the ABC algorithm to solve it in an automatic manner to lessen human interference. ABCPruner has been demonstrated to be more effective, which also enables the fine-tuning to be conducted efficiently in an end-to-end manner. The source codes can be available at https: //github.com/lmbxmu/ABCPruner.

Collaborative filtering (CF) is one of the most popular and important recommendation methodologies in the heart of numerous recommender systems today. Although widely adopted, existing CF-based methods, ranging from matrix factorization to the emerging graph-based methods, suffer inferior performance especially when the data for training are very limited. In this paper, we first pinpoint the root causes of such deficiency and observe two main disadvantages that stem from the inherent designs of existing CF-based methods, i.e., 1) inflexible modeling of users and items and 2) insufficient modeling of high-order correlations among the subjects. Under such circumstances, we propose a dual channel hypergraph collaborative filtering (DHCF) framework to tackle the above issues. First, a dual channel learning strategy, which holistically leverages the divide-and-conquer strategy, is introduced to learn the representation of users and items so that these two types of data can be elegantly interconnected while still maintaining their specific properties. Second, the hypergraph structure is employed for modeling users and items with explicit hybrid high-order correlations. The jump hypergraph convolution (JHConv) method is proposed to support the explicit and efficient embedding propagation of high-order correlations. Comprehensive experiments on two public benchmarks and two new real-world datasets demonstrate that DHCF can achieve significant and consistent improvements against other state-of-the-art methods.

In this paper, we investigate the fragility of deep image captioning models against adversarial attacks. Different from existing works that generate common words and concepts, we focus on the adversarial attacks towards controllable image captioning, i.e., removing target words from captions by imposing adversarial noises to images while maintaining the captioning accuracy for the remaining visual content. We name this new task as Masked Image Captioning (MIC), which is expected to be training and labeling free for end-to-end captioning models. Meanwhile, we propose a novel adversarial learning approach for this new task, termed Show, Mask, and Tell (SMT), which crafts adversarial examples to mask the target concepts via minimizing an objective loss while training the noise generator. Concretely, three novel designs are introduced in this loss, i.e., word removal regularization, captioning accuracy regularization, and noise filtering regularization. For quantitative validation, we propose a benchmark dataset for MIC based on the MS COCO dataset, together with a new evaluation metric called Attack Quality. Experimental results show that the proposed approach achieves successful attacks by removing 93.8% and 91.9% target words while maintaining 97.3% and 97.4% accuracies on two cutting-edge captioning models, respectively.

Referring expression segmentation (RES) aims to segment the target instance in a given image according to a natural language expression. Its main challenge lies in how to quickly and accurately align the text expression to the referred visual instances. In this paper, we focus on addressing this issue by proposing a Cascade Grouped Attention Network (CGAN) with two innovative designs: Cascade Grouped Attention (CGA) and Instance-level Attention (ILA) loss. Specifically, CGA is used to perform step-wise reasoning over the entire image to perceive the differences between instances accurately yet efficiently, so as to identify the referent. ILA loss is further embedded into each step of CGA to directly supervise the attention modeling, which improves the alignments between the text expression and the visual instances. Through these two novel designs, CGAN can achieve the high efficiency of one-stage RES while possessing a strong reasoning ability comparable to the two-stage methods. To validate our model, we conduct extensive experiments on three RES benchmark datasets and achieve significant performance gains over existing one-stage and multi-stage models

Modeling human visual attention mechanism is a fundamental problem for the understanding of human vision, which has also been demonstrated as an important module for various multimedia applications such as image captioning and visual question answering. In this paper, we propose a new probabilistic framework for attention, and introduce the concept ofmode to model the flexibility and adaptability of attention modulation in complex environments. We characterize the correlations between the visual input, the activated mode, the saliency and the spatial allocation of attention via a graphical model representation, based on which we explore the lingual guidance from captioning data for the implementation of a mode-sensitive attention (MSA) model. The proposed framework explicitly justifies the usage of center bias for fixation prediction and can convert an arbitrary learning-based backbone attention model to a more robust multi-mode version. Experimental results on the York120, MIT1003 and PASCAL datasets demonstrate the effectiveness of the proposed method.

In this paper, we propose a cross-modal network architecture search (NAS) algorithm for VQA, termed as k-Armed Bandit based NAS (KAB-NAS). KAB-NAS regards the design of each layer as a k-armed bandit problem and updates the preference of each candidate via numerous samplings in a single-shot search framework. To establish an effective search space, we further propose a new architecture termed Automatic Graph Attention Network (AGAN), and extend the popular self-attention layer with three graph structures, denoted as dense-graph, co-graph and separate-graph.These graph layers are used to form the direction of information propagation in the graph network, and their optimal combinations are searched by KAB-NAS. To evaluate KAB-NAS and AGAN, we conduct extensive experiments on two VQA benchmark datasets, i.e., VQA2.0 and GQA, and also test AGAN with the popular BERT-style pre-training. The experimental results show that with the help of KAB-NAS, AGAN can achieve the state-of-the-art performance on both benchmark datasets with much fewer parameters and computations.

Binary Neural Network (BNN) shows its predominance in reducing the complexity of deep neural networks. However, it suffers severe performance degradation. One of the major impediments is the large quantization error between the full-precision weight vector and its binary vector. Previous works focus on compensating for the norm gap while leaving the angular bias hardly touched. In this paper, for the first time, we explore the influence of angular bias on the quantization error and then introduce a Rotated Binary Neural Network (RBNN), which considers the angle alignment between the full-precision weight vector and its binarized version. At the beginning of each training epoch, we propose to rotate the full-precision weight vector to its binary vector to reduce the angular bias. To avoid the high complexity of learning a large rotation matrix, we further introduce a bi-rotation formulation that learns two smaller rotation matrices. In the training stage, we devise an adjustable rotated weight vector for binarization to escape the potential local optimum. Our rotation leads to around 50% weight flips which maximize the information gain. Finally, we propose a training-aware approximation of the sign function for the gradient backward. Experiments on CIFAR-10 and ImageNet demonstrate the superiorities of RBNN over many state-of-the-arts. Our source code, experimental settings, training logs and binary models are available at https://github.com/lmbxmu/RBNN.

Weakly supervised object detection (WSOD) has attracted extensive research attention due to its great flexibility of exploiting large-scale dataset with only image-level annotations for detector training. Despite its great advance in recent years, WSOD still suffers limited performance, which is far below that of fully supervised object detection (FSOD). As most WSOD methods depend on object proposal algorithms to generate candidate regions and are also confronted with challenges like low-quality predicted bounding boxes and large scale variation. In this paper, we propose a unified WSOD framework, termed UWSOD, to develop a high-capacity general detection model with only image-level labels, which is self-contained and does not require external modules or additional supervision. To this end, we exploit three important components, i.e., object proposal generation, bounding-box fine-tuning and scale-invariant features. First, we propose an anchor-based self-supervised proposal generator to hypothesize object locations, which is trained end-to-end with supervision created by UWSOD for both objectness classification and regression. Second, we develop a step-wise bounding-box fine-tuning to refine both detection scores and coordinates by progressively select high-confidence object proposals as positive samples, which bootstraps the quality of predicted bounding boxes. Third, we construct a multi-rate resampling pyramid to aggregate multi-scale contextual information, which is the first in-network feature hierarchy to handle scale variation in WSOD. Extensive experiments on PASCAL VOC and MS COCO show that the proposed UWSOD achieves competitive results with the state-of-the-art WSOD methods while not requiring external modules or additional supervision. Moreover, the upper-bound performance of UWSOD with class-agnostic ground-truth bounding boxes approaches Faster R-CNN, which demonstrates UWSOD has fully-supervised-level capacity.

In this paper, we address the problem of monocular depth estimation when only a limited number of training image-depth pairs are available. To achieve a high regression accuracy, the state-of-the-art estimation methods rely on CNNs trained with a large number of image-depth pairs, which are prohibitively costly or even infeasible to acquire. Aiming to break the curse of such expensive data collections, we propose a semi-supervised adversarial learning framework that only utilizes a small number of image-depth pairs in conjunction with a large number of easily-available monocular images to achieve high performance. In particular, we use one generator to regress the depth and two discriminators to evaluate the predicted depth, i.e., one inspects the image-depth pair while the other inspects the depth channel alone. These two discriminators provide their feedbacks to the generator as the loss to generate more realistic and accurate depth predictions. Experiments show that the proposed approach can (1) improve most state-of-the-art models on the NYUD v2 dataset by effectively leveraging additional unlabeled data sources; (2) reach state-of-the-art accuracy when the training set is small, e.g., on the Make3D dataset; (3) adapt well to an unseen new dataset (Make3D in our case) after training on an annotated dataset (KITTI in our case).

Online image hashing aims to update hash functions on-the-fly along with newly arriving data streams, which has found broad applications in computer vision and beyond. To this end, most existing methods update hash functions simply using discrete labels or pairwise similarity to explore intra-class relationships, which, however, often deteriorates search performance when facing a domain gap or semantic shift. One reason is that they ignore the particular semantic relationships among different classes, which should be taken into account in updating hash functions. Besides, the common characteristics between the label vectors (can be regarded as a sort of binary codes) and to-be-learned binary hash codes have left unexploited. In this paper, we present a novel online hashing method, termed Similarity Preserving Linkage Hashing (SPLH), which not only utilizes pairwise similarity to learn the intra-class relationships, but also fully exploits a latent linkage space to capture the inter-class relationships and the common characteristics between label vectors and to-be-learned hash codes. Specifically, SPLH first maps the independent discrete label vectors and binary hash codes into a linkage space, through which the relative semantic distance between data points can be assessed precisely. As a result, the pairwise similarities within the newly arriving data stream are exploited to learn the latent semantic space to benefit binary code learning. To learn the model parameters effectively, we further propose an alternating optimization algorithm. Extensive experiments conducted on three widely-used datasets demonstrate the superior performance of SPLH over several state-of-the-art online hashing methods.

Weakly supervised object detection has attracted increasing research attention recently. To this end, most existing schemes rely on scoring category-independent region proposals, which is formulated as a multiple instance learning problem. During this process, the proposal scores are aggregated and supervised by only image-level labels, which often fails to locate object boundaries precisely. In this paper, we break through such a restriction by taking a deeper look into the score aggregation stage and propose a Category-aware Spatial Constraint (CSC) scheme for proposals, which is integrated into weakly supervised object detection in an end-to-end learning manner. In particular, we incorporate the global shape information of objects as an unsupervised constraint, which is inferred from build-in foreground-and-background cues, termed Category-specific Pixel Gradient (CPG) maps. Specifically, each region proposal is weighted according to how well it covers the estimated shape of objects. For each category, a multi-center regularization is further introduced to penalize the violations between centers cluster and high-score proposals in a given image. Extensive experiments are done on the most widely-used benchmark Pascal VOC and COCO, which shows that our approach significantly improves weakly supervised object detection without adding new learnable parameters to the existing models nor changing the structures of CNNs.

Recent advances of person re-identification have well advocated the usage of human body cues to boost performance. However, most existing methods still retain on exploiting a relatively coarse-grained local information. Such information may include redundant backgrounds that are sensitive to the apparently similar persons when facing challenging scenarios like complex poses, inaccurate detection, occlusion and misalignment. In this paper we propose a novel Fine-Grained Spatial Alignment Model (FGSAM) to mine fine-grained local information to handle the aforementioned challenge effectively. In particular, we first design a pose resolve net with channel parse blocks (CPB) to extract pose information in pixel-level. This network allows the proposed model to be robust to complex pose variations while suppressing the redundant backgrounds caused by inaccurate detection and occlusion. Given the extracted pose information, a locally reinforced alignment mode is further proposed to address the misalignment problem between different local parts by considering different local parts along with attribute information in a fine-grained way. Finally, a focal triplet loss is designed to effectively train the entire model, which imposes a constraint on the intra-class and an adaptively weight adjustment mechanism to handle the hard sample problem. Extensive evaluations and analysis on Market1501, DukeMTMC-reid and PETA datasets demonstrate the effectiveness of FGSAM in coping with the problems of misalignment, occlusion and complex poses.

The success of convolutional neural networks (CNNs) in computer vision applications has been accompanied by a significant increase of computation and memory costs, which prohibits their usage on resource-limited environments, such as mobile systems or embedded devices. To this end, the research of CNN compression has recently become emerging. In this paper, we propose a novel filter pruning scheme, termed structured sparsity regularization (SSR), to simultaneously speed up the computation and reduce the memory overhead of CNNs, which can be well supported by various off-the-shelf deep learning libraries. Concretely, the proposed scheme incorporates two different regularizers of structured sparsity into the original objective function of filter pruning, which fully coordinates the global output and local pruning operations to adaptively prune filters. We further propose an alternative updating with Lagrange multipliers (AULM) scheme to efficiently solve its optimization. AULM follows the principle of alternating direction method of multipliers (ADMM) and alternates between promoting the structured sparsity of CNNs and optimizing the recognition loss, which leads to a very efficient solver ( 2.5× to the most recent work that directly solves the group sparsity-based regularization). Moreover, by imposing the structured sparsity, the online inference is extremely memory-light since the number of filters and the output feature maps are simultaneously reduced. The proposed scheme has been deployed to a variety of state-of-the-art CNN structures, including LeNet, AlexNet, VGGNet, ResNet, and GoogLeNet, over different data sets. Quantitative results demonstrate that the proposed scheme achieves superior performance over the state-of-the-art methods. We further demonstrate the proposed compression scheme for the task of transfer learning, including domain adaptation and object detection, which also show exciting performance gains over the state-of-the-art filter pruning methods.

Aortic stent has been widely used in restoring vascular stenosis and assisting patients with cardiovascular disease. The effective visualization of aortic stent is considered to be critical to ensure the effectiveness and functions of the aortic stent in clinical practice. Volume rendering with ray casting has been used as an effective approach to enable the effective visualization of aortic stent. The volume rendering relies on the transfer function that converts the medical images into optical attributes including color and transparency. This article proposes a new transfer function, namely, the multi-dimensional transfer function, to provide additional transparency value of a voxel. The proposed approach using the additional transparency value effectively assists the distinguishing of tissues that have the same CT value. The transparency values are simultaneously determined by gray threshold and gray change threshold, which can recognize the unnecessary structures such as bones transparent. A series of experimental results demonstrate that the situation of aorta stent of a patient can be directly observed, and the angle of view can be switched arbitrarily. The proposed method provides a new way for the operation of a virtual endoscopy to reach the place of blood vessels that a traditional endoscopy fails to reach.

In this paper, we present a hypergraph neural networks (HGNN) framework for data representation learning, which can encode high-order data correlation in a hypergraph structure. Confronting the challenges of learning representation for complex data in real practice, we propose to incorporate such data structure in a hypergraph, which is more flexible on data modeling, especially when dealing with complex data. In this method, a hyperedge convolution operation is designed to handle the data correlation during representation learning. In this way, traditional hypergraph learning procedure can be conducted using hyperedge convolution operations efficiently. HGNN is able to learn the hidden layer representation considering the high-order data structure, which is a general framework considering the complex data correlations. We have conducted experiments on citation network classification and visual object recognition tasks and compared HGNN with graph convolutional networks and other traditional methods. Experimental results demonstrate that the proposed HGNN method outperforms recent state-of-theart methods. We can also reveal from the results that the proposed HGNN is superior when dealing with multi-modal data compared with existing methods.

When facing large-scale image datasets, online hashing serves as a promising solution for online retrieval and prediction tasks. It encodes the online streaming data into compact binary codes, and simultaneously updates the hash functions to renew codes of the existing dataset. To this end, the existing methods update hash functions solely based on the new data batch, without investigating the correlation between such new data and the existing dataset. In addition, existing works update the hash functions using a relaxation process in its corresponding approximated continuous space. And it remains as an open problem to directly apply discrete optimizations in online hashing. In this paper, we propose a novel supervised online hashing method, termed Balanced Similarity for Online Discrete Hashing (BSODH), to solve the above problems in a unified framework. BSODH employs a well-designed hashing algorithm to preserve the similarity between the streaming data and the existing dataset via an asymmetric graph regularization. We further identify the “data-imbalance” problem brought by the constructed asymmetric graph, which restricts the application of discrete optimization in our problem. Therefore, a novel balanced similarity is further proposed, which uses two equilibrium factors to balance the similar and dissimilar weights and eventually enables the usage of discrete optimizations. Extensive experiments conducted on three widely-used benchmarks demonstrate the advantages of the proposed method over the stateof-the-art methods.

Symmetric positive defined (SPD) matrix has attracted increasing research focus in image/video analysis, which merits in capturing the Riemannian geometry in its structured 2D feature representation. However, computation in the vector space on SPD matrices cannot capture the geometric properties, which corrupts the classification performance. To this end, Riemannian based deep network has become a promising solution for SPD matrix classification, because of its excellence in performing non-linear learning over SPD matrix. Besides, Riemannian metric learning typically adopts a kNN classifier that cannot be extended to large-scale datasets, which limits its application in many time-efficient scenarios. In this paper, we propose a Bag-of-Matrix-Summarization (BoMS) method to be combined with Riemannian network, which handles the above issues towards highly efficient and scalable SPD feature representation. Our key innovation lies in the idea of summarizing data in a Riemannian geometric space instead of the vector space. First, the whole training set is compressed with a small number of matrix features to ensure high scalability. Second, given such a compressed set, a constant-length vector representation is extracted by efficiently measuring the distribution variations between the summarized data and the latent feature of the Riemannian network. Finally, the proposed BoMS descriptor is integrated into the Riemannian network, upon which the whole framework is end-to-end trained via matrix back-propagation. Experiments on four different classification tasks demonstrate the superior performance of the proposed method over the state-of-the-art methods.

Three-dimensional (3D) shape recognition has drawn much research attention in the field of computer vision. The advances of deep learning encourage various deep models for 3D feature representation. For point cloud and multi-view data, two popular 3D data modalities, different models are proposed with remarkable performance. However the relation between point cloud and views has been rarely investigated. In this paper, we introduce Point-View Relation Network (PVRNet), an effective network designed to well fuse the view features and the point cloud feature with a proposed relation score module. More specifically, based on the relation score module, the point-single-view fusion feature is first extracted by fusing the point cloud feature and each single view feature with point-singe-view relation, then the pointmulti- view fusion feature is extracted by fusing the point cloud feature and the features of different number of views with point-multi-view relation. Finally, the point-single-view fusion feature and point-multi-view fusion feature are further combined together to achieve a unified representation for a 3D shape. Our proposed PVRNet has been evaluated on ModelNet40 dataset for 3D shape classification and retrieval. Experimental results indicate our model can achieve significant performance improvement compared with the state-of-the-art models.

Recent advances on fine-grained image retrieval prefer learning convolutional neural network (CNN) with specific fullyconnect layer designed loss function for discriminative feature representation. Essentially, such loss should establish a robust metric to efficiently distinguish high-dimensional features within and outside fine-grained categories. To this end, the existing loss functions are defected in two aspects: (a) The feature relationship is encoded inside the training batch. Such a local scope leads to low accuracy. (b) The error is established by the mean square, which needs pairwise distance computation in training set and results in low efficiency. In this paper, we propose a novel metric learning scheme, termed Normalize-Scale Layer and Decorrelated Global Centralized Ranking Loss, which achieves extremely efficient and discriminative learning, i.e., 5× speedup over triplet loss and 12% recall boost on CARS196. Our method originates from the classic softmax loss, which has a global structure but does not directly optimize the distance metric as well as the inter/intra class distance. We tackle this issue through a hypersphere layer and a global centralized ranking loss with a pairwise decorrelated learning. In particular, we first propose a Normalize-Scale Layer to eliminate the gap between metric distance (for measuring distance in retrieval) and dot product (for dimension reduction in classification). Second, the relationship between features is encoded under a global centralized ranking loss, which targets at optimizing metric distance globally and accelerating learning procedure. Finally, the centers are further decorrelated by Gram-Schmidt process, leading to extreme efficiency (with 20 epochs in training procedure) and discriminability in feature learning. We have conducted quantitative evaluations on two fine-grained retrieval benchmark. The superior performance demonstrates the merits of the proposed approach over the state-of-the-arts.

In this paper, we uncover the issue of knowledge inertia in visual question answering (VQA), which commonly exists in most VQA models and forces the models to mainly rely on the question content to “guess” answer, without regard to the visual information. Such an issue not only impairs the performance of VQA models, but also greatly reduces the credibility of the answer prediction. To this end, simply highlighting the visual features in the model is undoable, since the prediction is built upon the joint modeling of two modalities and largely influenced by the data distribution. In this paper, we propose a Pairwise Inconformity Learning (PIL) to tackle the issue of knowledge inertia. In particular, PIL takes full advantage of the similar image pairs with diverse answers to an identical question provided in VQA2.0 dataset. It builds a multi-modal embedding space to project pos./neg. feature pairs, upon which word vectors of answers are modeled as anchors. By doing so, PIL strengthens the importance of visual features in prediction with a novel dynamic-margin based triplet loss that efficiently increases the semantic discrepancies between pos./neg. image pairs. To verify the proposed PIL, we plug it on a baseline VQA model as well as a set of recent VQA models, and conduct extensive experiments on two benchmark datasets, i.e., VQA1.0 and VQA2.0. Experimental results show that PIL can boost the accuracy of the existing VQA models (1.56%-2.93% gain) with a negligible increase in parameters (0.85%-5.4% parameters). Qualitative results also reveal the elimination of knowledge inertia in the existing VQA models after implementing our PIL.

In visual question answering (VQA), recent advances have well advocated the use of attention mechanism to precisely link the question to the potential answer areas. As the difficulty of the question increases, more VQA models adopt multiple attention layers to capture the deeper visual-linguistic correlation. But a negative consequence is the explosion of parameters, which makes the model vulnerable to over-fitting, especially when limited training examples are given. In this paper, we propose an extremely compact alternative to this static multi-layer architecture towards accurate yet efficient attention modeling, termed as Dynamic Capsule Attention (CapsAtt). Inspired by the recent work of Capsule Network, CapsAtt treats visual features as capsules and obtains the attention output via dynamic routing, which updates the attention weights by calculating coupling coefficients between the underlying and output capsules. Meanwhile, CapsAtt also discards redundant projection matrices to make the model much more compact. We quantify CapsAtt on three benchmark VQA datasets, i.e., COCO-QA, VQA1.0 and VQA2.0. Compared to the traditional multi-layer attention model, CapsAtt achieves significant improvements of up to 4.1%, 5.2% and 2.2% on three datasets, respectively. Moreover, with much fewer parameters, our approach also yields competitive results compared to the latest VQA models. To further verify the generalization ability of CapsAtt, we also deploy it on another challenging multi-modal task of image captioning, where state-of-the-art performance is achieved with a simple network structure.

Most existing visual search systems are deployed based upon fixed kinds of visual features, which prohibits the feature reusing across different systems or when upgrading systems with a new type of feature. Such a setting is obviously inflexible and time/memory consuming, which is indeed mendable if visual features can be ``translated" across systems. In this paper, we make the first attempt towards visual feature translation to break through the barrier of using features across different visual search systems. To this end, we propose a Hybrid Auto-Encoder (HAE) to translate visual features, which learns a mapping by minimizing the translation and reconstruction errors. Based upon HAE, an Undirected Affinity Measurement (UAM) is further designed to quantify the affinity among different types of visual features. Extensive experiments have been conducted on several public datasets with sixteen different types of widely-used features in visual search systems. Quantitative results show the encouraging possibilities of feature translation. For the first time, the affinity among widely-used features like SIFT and DELF is reported.

Compressing convolutional neural networks (CNNs) has received ever-increasing research focus. However, most existing CNN compression methods do not interpret their inherent structures to distinguish the implicit redundancy. In this paper, we investigate the problem of CNN compression from a novel interpretable perspective. The relationship between the input feature maps and 2D kernels is revealed in a theoretical framework, based on which a kernel sparsity and entropy (KSE) indicator is proposed to quantitate the feature map importance in a feature-agnostic manner to guide model compression. Kernel clustering is further conducted based on the KSE indicator to accomplish high-precision CNN compression. KSE is capable of simultaneously compressing each layer in an efficient way, which is significantly faster compared to previous data-driven feature map pruning methods. We comprehensively evaluate the compression and speedup of the proposed method on CIFAR-10, SVHN and ImageNet 2012. Our method demonstrates superior performance gains over previous ones. In particular, it achieves 4.7× FLOPs reduction and 2.9× compression on ResNet-50 with only a top-5 accuracy drop of 0.35% on ImageNet 2012, which significantly outperforms state-of-the-art methods.

Structured pruning of filters or neurons has received increased focus for compressing convolutional neural networks. Most existing methods rely on multi-stage optimizations in a layer-wise manner for iteratively pruning and retraining which may not be optimal and may be computation intensive. Besides, these methods are designed for pruning a specific structure, such as filter or block structures without jointly pruning heterogeneous structures. In this paper, we propose an effective structured pruning approach that jointly prunes filters as well as other structures in an end-to-end manner. To accomplish this, we first introduce a soft mask to scale the output of these structures by defining a new objective function with sparsity regularization to align the output of baseline and network with this mask. We then effectively solve the optimization problem by generative adversarial learning (GAL), which learns a sparse soft mask in a label-free and an end-to-end manner. By forcing more scale factors in the soft mask to zero, the fast iterative shrinkage-thresholding algorithm (FISTA) can be leveraged to fast and reliably remove the corresponding structures. Extensive experiments demonstrate the effectiveness of GAL on different datasets, including MNIST, CIFAR-10 and ImageNet ILSVRC 2012. For example, on ImageNet ILSVRC 2012, the pruned ResNet-50 achieves 10.88% Top-5 error and results in a factor of 3.7x speedup. This significantly outperforms state-of-the-art methods.

The rapidly decreasing computation and memory cost has recently driven the success of many applications in the field of deep learning. Practical applications of deep learning in resource-limited hardware, such as embedded devices and smart phones, however, remain challenging. For binary convolutional networks, the reason lies in the degraded representation caused by binarizing full-precision filters. To address this problem, we propose new circulant filters (CiFs) and a circulant binary convolution (CBConv) to enhance the capacity of binarized convolutional features via our circulant back propagation (CBP). The CiFs can be easily incorporated into existing deep convolutional neural networks (DCNNs), which leads to new Circulant Binary Convolutional Networks (CBCNs). Extensive experiments confirm that the performance gap between the 1-bit and full-precision DCNNs is minimized by increasing the filter diversity, which further increases the representational ability in our networks. Our experiments on ImageNet show that CBCNs achieve 61.4% top-1 accuracy with ResNet18. Compared to the state-of-the-art such as XNOR, CBCNs can achieve up to 10% higher top-1 accuracy with more powerful representational ability.

Weakly supervised learning has attracted growing research attention due to the significant saving in annotation cost for tasks that require intra-image annotations, such as object detection and semantic segmentation. To this end, existing weakly supervised object detection and semantic segmentation approaches follow an iterative label mining and model training pipeline. However, such a self-enforcement pipeline makes both tasks easy to be trapped in local minimums. In this paper, we join weakly supervised object detection and segmentation tasks with a multi-task learning scheme for the first time, which uses their respective failure patterns to complement each other's learning. Such cross-task enforcement helps both tasks to leap out of their respective local minimums. In particular, we present an efficient and effective framework termed Weakly Supervised Joint Detection and Segmentation (WS-JDS). WS-JDS has two branches for the above two tasks, which share the same backbone network. In the learning stage, it uses the same cyclic training paradigm but with a specific loss function such that the two branches benefit each other. Extensive experiments have been conducted on the widely-used Pascal VOC and COCO benchmarks, which demonstrate that our model has achieved competitive performance with the state-of-the-art algorithms.

Most existing Re-IDentification (Re-ID) methods are highly dependent on precise bounding boxes that enable images to be aligned with each other. However, due to the challenging practical scenarios, current detection models often produce inaccurate bounding boxes, which inevitably degenerate the performance of existing Re-ID algorithms. In this paper, we propose a novel coarse-to-fine pyramid model to relax the need of bounding boxes, which not only incorporates local and global information, but also integrates the gradual cues between them. The pyramid model is able to match at different scales and then search for the correct image of the same identity, even when the image pairs are not aligned. In addition, in order to learn discriminative identity representation, we explore a dynamic training scheme to seamlessly unify two losses and extract appropriate shared information between them. Experimental results clearly demonstrate that the proposed method achieves the state-of-the-art results on three datasets. Especially, our approach exceeds the current best method by 9.5% on the most challenging CUHK03 dataset.

In this paper, we address the problem of real-time image semantic segmentation for indoor scene. As for scene parsing, both accuracy and speed are equally important. However, most of existing methods mainly focus on improving accuracy rather than speed. How to find a balance between accuracy and speed is crucial for real-time semantic segmentation tasks. To tackle this problem, we propose a lightweight framework with depthwise dilation residual module and multi-scale information integration module. A single DSNet yields the performance of mIoU accuracy 32.12% on SUN RGB-D dataset and accuracy 26.32% on ADE20K dataset, which is the most challenge scene parsing dataset. Besides, our system yields real-time inference on a single NVIDIA GPU.

In this paper, we study the problem of zero-shot fine-grained recognition. It aims to distinguish unseen subordinate categories through some other seen categories within an entry-level category. We demonstrate the necessity to learn multiple latent dictionaries through joint training with specific set of instances, human-defined attributes and the class labels. A novel approach that is capable of 1) automatically assigning suitable dictionaries for each instance and 2) learning similarity-specific semantic representations for zero-shot fine-grained recognition is proposed. Experimental results on three benchmark datasets demonstrate that the proposed method achieves superior or comparable performance.

The cross-modality topic detection in social media retains as an open problem mainly due to the difficulty of dealing with modality independence and modality missing. In this paper, we present a novel Deep Topical Correlation Analysis (DTCA) approach, which achieves robust and accurate topic detecting for micro-blogs and handles aforementioned challenges simultaneously. In particular, bidirectional recurrent neural networks and convolutional neural networks are used to learn deep textual and visual features, respectively. Then a Canonical Correlation Analysis based fusion scheme is proposed, which has two innovations to deal with both two problems mentioned above. We further release a large-scale cross-modal twitter dataset for topic detection. Extensive and quantitative evaluations are conducted with comparisons to several state-of-the-arts and alternative approaches on this dataset. Significant performance gains are reported to demonstrate the merits of proposed approach.

Deep learning models have shown their vulnerabilities to universal adversarial perturbations (UAP), which are quasi-imperceptible. Compared to the conventional supervised UAPs that suffer from the knowledge of training data, the data-independent unsupervised UAPs are more applicable. Existing unsupervised methods fail to take advantage of the model uncertainty to produce robust perturbations. In this paper, we propose a new unsupervised universal adversarial perturbation method, termed as Prior Driven Uncertainty Approximation (PD-UA), to generate a robust UAP by fully exploiting the model uncertainty at each network layer. Specifically, a Monte Carlo sampling method is deployed to activate more neurons to increase the model uncertainty for a better adversarial perturbation. Thereafter, a textural bias prior to revealing a statistical uncertainty is proposed, which helps to improve the attacking performance. The UAP is crafted by the stochastic gradient descent algorithm with a boosted momentum optimizer, and a Laplacian pyramid frequency model is finally used to maintain the statistical uncertainty. Extensive experiments demonstrate that our method achieves well attacking performances on the ImageNet validation set, and significantly improves the fooling rate compared with the state-of-the-art methods.

While it is trivial for humans to quickly assess the perceptual similarity between two images, the underlying mechanism are thought to be quite complex. Despite this, the most widely adopted perceptual metrics today, such as SSIM and FSIM, are simple, shallow functions, and fail to consider many factors of human perception. Recently, the facial modeling community has observed that the inclusion of both structure and texture has a significant positive benefit for face sketch synthesis (FSS). But how perceptual are these so-called “perceptual features”? Which elements are critical for their success? In this paper, we design a perceptual metric, called Structure Co-Occurrence Texture (Scoot), which simultaneously considers the block-level spatial structure and co-occurrence texture statistics. To test the quality of metrics, we propose three novel meta-measures based on various reliable properties. Extensive experiments verify that our Scoot metric exceeds the performance of prior work. Besides, we built the first largest scale (152k judgments) human-perception-based sketch database that can evaluate how well a metric consistent with human perception. Our results suggest that “spatial structure” and “co-occurrence texture” are two generally applicable perceptual features in face sketch synthesis.

Deep convolutional neural networks (DCNNs) have dominated the recent developments in computer vision through making various record-breaking models. However, it is still a great challenge to achieve powerful DCNNs in resource-limited environments, such as on embedded devices and smart phones. Researchers have realized that 1-bit CNNs can be one feasible solution to resolve the issue; however, they are baffled by the inferior performance compared to the full-precision DCNNs. In this paper, we propose a novel approach, called Bayesian optimized 1-bit CNNs (denoted as BONNs), taking the advantage of Bayesian learning, a well-established strategy for hard problems, to significantly improve the performance of extreme 1-bit CNNs. We incorporate the prior distributions of full-precision kernels and features into the Bayesian framework to construct 1-bit CNNs in an end-to-end manner, which have not been considered in any previous related methods. The Bayesian losses are achieved with a theoretical support to optimize the network simultaneously in both continuous and discrete spaces, aggregating different losses jointly to improve the model capacity. Extensive experiments on the ImageNet and CIFAR datasets show that BONNs achieve the best classification performance compared to state-of-the-art 1-bit CNNs.

Universal adversarial perturbations (UAPs), a.k.a. input-agnostic perturbations, has been proved to exist and be able to fool cutting-edge deep learning models on most of the data samples. Existing UAP methods mainly focus on attacking image classification models. Nevertheless, little attention has been paid to attacking image retrieval systems. In this paper, we make the first attempt in attacking image retrieval systems. Concretely, image retrieval attack is to make the retrieval system return irrelevant images to the query at the top ranking list. It plays an important role to corrupt the neighbourhood relationships among features in image retrieval attack. To this end, we propose a novel method to generate retrieval-against UAP to break the neighbourhood relationships of image features via degrading the corresponding ranking metric. To expand the attack method to scenarios with varying input sizes or untouchable network parameters, a multi-scale random resizing scheme and a ranking distillation strategy are proposed. We evaluate the proposed method on four widely-used image retrieval datasets, and report a significant performance drop in terms of different metrics, such as mAP and mP@10. Finally, we test our attack methods on the real-world visual search engine, i.e., Google Images, which demonstrates the practical potentials of our methods.

Recent saliency models extensively explore to incorporate multi-scale contextual information from Convolutional Neural Networks (CNNs). Besides direct fusion strategies, many approaches introduce message-passing to enhance CNN features or predictions. However, the messages are mainly transmitted in two ways, by feature-to-feature passing, and by prediction-to-prediction passing. In this paper, we add message-passing between features and predictions and propose a deep unified CRF saliency model . We design a novel cascade CRFs architecture with CNN to jointly refine deep features and predictions at each scale and progressively compute a final refined saliency map. We formulate the CRF graphical model that involves message-passing of feature-feature, feature-prediction, and prediction-prediction, from the coarse scale to the finer scale, to update the features and the corresponding predictions. Also, we formulate the mean-field updates for joint end-to-end model training with CNN through back propagation. The proposed deep unified CRF saliency model is evaluated over six datasets and shows highly competitive performance among the state of the arts.

Architectures obtained by Neural Architecture Search (NAS) have achieved highly competitive performance in various computer vision tasks. However, the prohibitive computation demand of forward-backward propagation in deep neural networks and searching algorithms makes it difficult to apply NAS in practice. In this paper, we propose a Multinomial Distribution Learning for extremely effective NAS, which considers the search space as a joint multinomial distribution, i.e., the operation between two nodes is sampled from this distribution, and the optimal network structure is obtained by the operations with the most likely probability in this distribution. Therefore, NAS can be transformed to a multinomial distribution learning problem, i.e., the distribution is optimized to have a high expectation of the performance. Besides, a hypothesis that the performance ranking is consistent in every training epoch is proposed and demonstrated to further accelerate the learning process. Experiments on CIFAR-10 and ImageNet demonstrate the effectiveness of our method. On CIFAR-10, the structure searched by our method achieves 2.55% test error, while being 6.0× (only 4 GPU hours on GTX1080Ti) faster compared with state-of-the-art NAS algorithms. On ImageNet, our model achieves 74% top1 accuracy under MobileNet settings (MobileNet V1/V2), while being 1.2× faster with measured GPU latency. Test code with pre-trained models are available at https: //github.com/tanglang96/MDENAS.

Sentiment analysis has attracted increasing attention recently due to its potential wide applications in opinion analysis, recommendation system, etc. Visual-textual sentiment analysis aims to improve the performance of sentiment analysis by leveraging both visual and textual signals. In this paper, we address the visual-textual sentiment analysis in product reviews. Our main contributions are two-fold. First, instead of crawling data from Flickr or Twitter with positive and negative labels in existing works, we introduce a new dataset for visual-textual sentiment analysis, termed as Product Reviews150K (PR-150K), which is collected from the product reviews of online shopping websites. Second, we propose a deep Tucker fusion method for visual-textual sentiment analysis, which efficiently combines visual and textual deep representations based on the Tucker decomposition and a bilinear pooling operation. Extensive experiments on our PR-150K, MVSO, and VSO datasets show that our method outperforms several state-of-the-art methods.

Most existing fine-grained image retrieval schemes are built based upon deep feature learning paradigms, which typically leverage the feature maps of the last convolutional layer as features. However, such representation focuses only on the global information of the object, leaving the local details unexploited, which is however crucial to identifying subtle differences for fine-grained retrieval. In this paper, we have discovered that the mid-level feature map roles as local salient regions, which well complements the existing global feature representations. To this end, a multi-layer framework is proposed to integrate both local and global representations with generalized mean (GeM) pooling and attention mechanism, trained with the proposed N-pair Center loss to learn more discriminative features. By doing so, state-of-the-art performance can be achieved without using the hard or negative example minings. In the experiments, our approach outperforms favourably compared to the current state-of-the-art methods on the CUB-200-2011, CARS196 and In-shop Clothes Retrieval datasets.

Image captioning has recently attracted ever-increasing research attention in multimedia and computer vision. However, most of image captioning models focus on generating the plain description for images, neglecting its colloquialism under a potential topic, e.g., the topic Movie for a poster. In this paper, we consider colloquial image captioning a new task, which, however, is trapped in the visual-textual semantic gap, as well as the semantic and the sentiment diversities. To this end, we propose a novel topic-guided multi-discriminator captioning model, termed TGMD-Cap. In particular, a topic-guided attention module is firstly designed to compose the effective visual semantic features. Secondly, a multi-discriminator generative adversarial module is designed to generate the semantically-diverse and affectively-diverse expression. To facilitate quantitative evaluation, we further construct and release a colloquial image captioning dataset, CIC-dataset, crawled from real-world social network, serving as the first of its kind. Quantitative results show that the proposed TGMD-Cap model outperforms the state-of-the-art approaches under various standard evaluation metrics.

Vehicle detection and recognition in remote sensing images are challenging, especially when only limited training data are available to accommodate various target categories. In this paper, we introduce a novel coarse-to-fine framework, which decomposes vehicle detection into segmentation-based vehicle localization and generalized zero-shot vehicle classification. Particularly, the proposed framework can well handle the problem of generalized zero-shot vehicle detection, which is challenging due to the requirement of recognizing vehicles that are even unseen during training. Specifically, a hierarchical DeepLab v3 model is proposed in the framework, which fully exploits fine-grained features to locate the target on a pixel-wise level, then recognizes vehicles in a coarse-grained manner. Additionally, the hierarchical DeepLab v3 model is beneficially compatible to combine the generalized zero-shot recognition. To the best of our knowledge, there is no publically available dataset to test comparative methods, we therefore construct a new dataset to fill this gap of evaluation. The experimental results show that the proposed framework yields promising results on the imperative yet difficult task of zero-shot vehicle detection and recognition.

Deep convolutional neural networks (DCNN) with manifold embedding have achieved considerable attention in computer vision. However, prior arts are usually based on the neighborhood-based graph modeling only the pairwise relationship between two samples, which fail to fully capture intra-class variations and thus suffer from severe performance loss for noisy data. While such intra-class variations can be well captured via sophisticated hypergraph structure, we are motivated and lead a hypergraph induced Convolutional Manifold Network (H-CMN) to significantly improve the representation capacity of DCNN for the complex data. Specifically, two innovative designs are provides: 1) our manifold preserving method is implemented based on a mini-batch, which can be efficiently plugged into the existing DCNN training pipelines and be scalable for large datasets; 2) a robust hypergraph is built for each mini-batch, which not only offers a strong robustness against typical noise, but also captures the variances from multiple features. Extensive experiments on the image classification task on large benchmarking datasets demonstrate that our model achieves much better performance than the state-of-the-art

Recently, person re-identification (re-ID) has attracted increasing research attention, which has broad application prospects in video surveillance and beyond. To this end, most existing methods highly relied on well-aligned pedestrian images and hand-engineered part-based model on the coarsest feature map. In this paper, to lighten the restriction of such fixed and coarse input alignment, an end-to-end part power set model with multi-scale features is proposed, which captures the discriminative parts of pedestrians from global to local, and from coarse to fine, enabling part-based scale-free person re-ID. In particular, we first factorize the visual appearance by enumerating $k$-combinations for all $k$ of $n$ body parts to exploit rich global and partial information to learn discriminative feature maps. Then, a combination ranking module is introduced to guide the model training with all combinations of body parts, which alternates between ranking combinations and estimating an appearance model. To enable scale-free input, we further exploit the pyramid architecture of deep networks to construct multi-scale feature maps with a feasible amount of extra cost in term of memory and time. Extensive experiments on the mainstream evaluation datasets, including Market-1501, DukeMTMC-reID and CUHK03, validate that our method achieves the state-of-the-art performance.

Face anti-spoofing detection is critical to guarantee the security of biometric face recognition systems. Despite extensive advances in facial anti-spoofing based on single-model image, little work has been devoted to multi-modal anti-spoofing, which is however widely encountered in real-world scenarios. Following the recent progress, this paper mainly focuses on multi-modal face anti-spoofing and aims to solve the following two challenges: (1) how to effectively fuse multi-modal information; and (2) how to effectively learn distinguishable features despite single cross-entropy loss. We propose a novel Multi-modal Multi-layer Fusion Convolutional Neural Network (mmfCNN), which targets at finding a discriminative model for recognizing the subtle differences between live and spoof faces. The mmfCNN can fully use different information provided by diverse modalities, which is based on a weight-adaptation aggregation approach. Specifically, we utilize a multi-layer fusion model to further aggregate the features from different layers, which fuses the low-, mid- and high-level information from different modalities in a unified framework. Moreover, a novel Average Binary Center (ABC) loss is proposed to maximize the dissimilarity between the features of live and spoof faces, which helps to stabilize the training to generate a robust and discriminative model. Extensive experiments conducted on the CISIA-SURF and 3DMAD datasets verify the significance and generalization capability of the proposed method for the face anti-spoofing task. Code is available at: https://github.com/SkyKuang/Face-anti-spoofing.

Deep networks have recently been applied to medical assistant diagnosis. The brain is the largest and the most complex structure in the central nervous system, which is also complicated in medical images such as computed tomography (CT) scan. While reading the CT image, radiologists generally search across the image to find lesions, characterize and measure them, and then describe them in the radiological report. To automate this process, we quantitatively analyze the cerebral hemorrhage dataset and propose a Multi-scale Feature with Collaborative Learning (MFCL) strategy in terms of Weakly Supervised Lesion Detection (WSLD), which not only adapts to the characteristics of detecting small lesions but also introduces the global constraint classification objective in training. Specifically, a multi-scale feature branch network and a collaborative learning are designed to locate the lesion area. Experimental results demonstrate that the proposed method is valid on the cerebral hemorrhage dataset, and a new baseline of WSLD is established on cerebral hemorrhage dataset.

No abstract available.

Despite the exciting progress in image captioning, generating diverse captions for a given image remains as an open problem. Existing methods typically apply generative models such as Variational Auto-Encoder to diversify the captions, which however neglect two key factors of diverse expression, i.e., the lexical diversity and the syntactic diversity. To model these two inherent diversities in image captioning, we propose a Variational Structured Semantic Inferring model (termed VSSI-cap) executed in a novel structured encoder-inferer-decoder schema. VSSI-cap mainly innovates in a novel structure, i.e., Variational Multi-modal Inferring tree (termed VarMI-tree). In particular, conditioned on the visual-textual features from the encoder, the VarMI-tree models the lexical and syntactic diversities by inferring their latent variables (with variations) in an approximate posterior inference guided by a visual semantic prior. Then, a reconstruction loss and the posterior-prior KL-divergence are jointly estimated to optimize the VSSI-cap model. Finally, diverse captions are generated upon the visual features and the latent variables from this structured encoder-inferer-decoder model. Experiments on the benchmark dataset show that the proposed VSSI-cap achieves significant improvements over the state-of-the-arts.

Learning representations with diversified information remains as an open problem. Towards learning diversified representations, a new approach, termed Information Competing Process (ICP), is proposed in this paper. Aiming to enrich the information carried by feature representations, ICP separates a representation into two parts with different mutual information constraints. The separated parts are forced to accomplish the downstream task independently in a competitive environment which prevents the two parts from learning what each other learned for the downstream task. Such competing parts are then combined synergistically to complete the task. By fusing representation parts learned competitively under different conditions, ICP facilitates obtaining diversified representations which contain rich information. Experiments on image classification and image reconstruction tasks demonstrate the great potential of ICP to learn discriminative and disentangled representations in both supervised and self-supervised learning settings.

Modern CNN-based object detectors assign anchors for ground-truth objects under the restriction of object-anchor Intersection-over-Unit (IoU). In this study, we propose a learning-to-match approach to break IoU restriction, allowing objects to match anchors in a flexible manner. Our approach, referred to as FreeAnchor, updates hand-crafted anchor assignment to "free" anchor matching by formulating detector training as a maximum likelihood estimation (MLE) procedure. FreeAnchor targets at learning features which best explain a class of objects in terms of both classification and localization. FreeAnchor is implemented by optimizing detection customized likelihood and can be fused with CNN-based detectors in a plug-and-play manner. Experiments on MS-COCO demonstrate that FreeAnchor consistently outperforms the counterparts with significant margins.

Convolutional neural networks (CNNs) have achieved remarkable success in various computer vision tasks, which are extremely powerful to deal with massive training data by using tens of millions of parameters. However, CNNs often cost significant memory and computation consumption, which prohibits their usage in resource-limited environments such as mobile or embedded devices. To address the above issues, the existing approaches typically focus on either accelerating the convolutional layers or compressing the fully-connected layers separatedly, without pursuing a joint optimum. In this paper, we overcome such a limitation by introducing a holistic CNN compression framework, termed LRDKT, which works throughout both convolutional and fully-connected layers. First, a low-rank decomposition (LRD) scheme is proposed to remove redundancies across both convolutional kernels and fully-connected matrices, which has a novel closed-form solver to significantly improve the efficiency of the existing iterative optimization solvers. Second, a novel knowledge transfer (KT) based training scheme is introduced. To recover the accumulated accuracy loss and overcome the vanishing gradient, KT explicitly aligns outputs and intermediate responses from a teacher (original) network to its student (compressed) network. We have comprehensively analyzed and evaluated the compression and speedup ratios of the proposed model on MNIST and ILSVRC 2012 benchmarks. In both benchmarks, the proposed scheme has demonstrated superior performance gains over the state-of-the-art methods. We also demonstrate the proposed compression scheme for the task of transfer learning, including domain adaptation and object detection, which show exciting performance gains over the state-of-the-arts. Our source code and compressed models are available at https://github.com/ShaohuiLin/LRDKT.

Binary code learning,a.k.a.hashing, has been successfully applied to the approximate nearest neighbor search in large-scale image collections. The key challenge lies in reducing the quantization error from the original real-valued feature space to a discrete Hamming space. Recent advances in unsupervised hashing advocate the preservation of ranking information, which is achieved by constraining the binary code learning to be correlated with pairwise similarity. However, few unsupervised methods consider the preservation ofordinal relationsin the learning process, which serves as a more basic cue to learn optimal binary codes. In this paper, we propose a novel hashing scheme, termed Ordinal Constraint Hashing (OCH), which embeds the ordinal relation among data points to preserve ranking into binary codes. The core idea is to construct an ordinal graph via tensor product, and then train the hash function over this graph to preserve the permutation relations among data points in the Hamming space. Subsequently, an in-depth acceleration scheme, termed Ordinal Constraint Projection (OCP), is introduced, which approximates the$n$-pair ordinal graph by$L$-pair anchor-based ordinal graph, and reduce the corresponding complexity from$O(n^4)$to$O(L^3)$($L\ll n$). Finally, to make the optimization tractable, we further relax the discrete constrains and design a customized stochastic gradient decent algorithm on the Stiefel manifold. Experimental results on serval large-scale benchmarks demonstrate that the proposed OCH method can achieve superior performance over the state-of-the-art approaches.

Hypergraph learning has been widely exploited in various image processing applications, due to its advantages in modeling the high-order information. Its efficacy highly depends on building an informative hypergraph structure to accurately and robustly formulate the underlying data correlation. However, the existing hypergraph learning methods are sensitive to non- Gaussian noise, which hurts the corresponding performance. In this paper, we present a noise-resistant hypergraph learning model, which provides superior robustness against various non- Gaussian noises. In particular, our model adopts low-rank representation to construct a hypergraph, which captures the globally linear data structure as well as preserving the grouping effect of highly-correlated data. We further introduce a correntropyinduced local metric to measure the reconstruction errors, which is particularly robust to non-Gaussian noises. Finally, the Frobenious-norm based regularization is proposed to combine with the low-rank regularizer, which enables our model to regularize the singular values of the coefficient matrix. By such, the non-zero coefficients are selected to generate a hyperedge set as well as the hyperedge weights. We have evaluated the proposed hypergraph model in the tasks of image clustering and semi-supervised image classification. Quantitatively, our scheme significantly enhances the performance of the state-of-the-art hypergraph models on several benchmark datasets.

While intrinsic data structure in subspace provides useful information for visual recognition, it has not yet been well studied in deep feature learning for action recognition. In this paper, we introduce a new spatio-temporal manifold network (STMN) that leverages data manifold structures to regularize deep action feature learning, aiming at simultaneously minimizing the intra-class variations of learned deep features and alleviating the over-fitting problem. To this end, the manifold prior is imposed from the top layer of a convolutional neural network (CNN), and is propagated across convolutional layers during forward-backward propagation. The observed correspondence of manifold structures in the data space and feature space validates that the manifold priori can be transferred across CNN layers. STMN theoretically recasts the problem of transferring the data structure prior into the deep learning architectures as a projection over the manifold via an embedding method, which can be easily solved by an Alternating Direction Method of Multipliers and Backward Propagation (ADMM-BP) algorithm. STMN is generic in the sense that it can be plugged into various backbone architectures to learn more discriminative representation for action recognition. Extensive experimental results show that our method achieves comparable or even better performance as compared with the state-of-the-art approaches on four benchmark datasets.

Microblog sentiment prediction has attracted extensive research focus with wide application prospects. With the increasing proportion of multimodal tweets consisting of images, texts, and emoticons, new challenges have been raised to the existing sentiment prediction schemes. More crucially, it remains an open problem to model the dependency among multiple modalities, where one or more modalities may be missing. In this paper, we present a novel Bi-layer Multimodal Hypergraph learning (Bi-MHG) toward robust sentiment prediction of multimodal tweets to tackle the above challenges. In particular, we design a two-layer structure for the proposed Bi-MHG model: The first layer, that is, a tweet-level hypergraph, learns the tweet-feature correlation and the tweet relevance to predict the sentiments of unlabeled tweets. The second layer, that is, a feature-level hypergraph learns the relevance among different feature modalities (including the midlevel visual features in Sentibank [1]) by leveraging prior multimodal sentiment dictionaries. These two layers are connected by sharing the relevance of multimodal features in a unified bilayer learning scheme. In such a way, Bi-MHG explicitly models the modality relevance rather than implicitly weighting multimodal features adopted in the existing Multimodal Hypergraph learning [2]. Finally, a nested alternating optimization is further proposed for Bi-MHG parameter learning. We have carried out extensive evaluations on a real-world microblog dataset crawled from Sina Weibo. For the task of multimodal sentiment prediction, superior performance is reported over several state-of-the-art and alternative approaches, which demonstrates the merits of the proposed scheme.

Given a training set of face photo-sketch pairs, face sketch synthesis targets at learning a mapping from the photo domain to the sketch domain. Despite the exciting progresses made in the literature, it retains as an open problem to synthesize high-quality sketches against blurs and deformations. Recent advances in generative adversarial training provide a new insight into face sketch synthesis, from which perspective the existing synthesis pipelines can be fundamentally revisited. In this paper, we present a novel face sketch synthesis method by multidomain adversarial learning (termed MDAL), which overcomes the defects of blurs and deformations toward high-quality synthesis. The principle of our scheme relies on the concept of "interpretation through synthesis." In particular, we first interpret face photographs in the photodomain and face sketches in the sketch domain by reconstructing themselves respectively via adversarial learning. We define the intermediate products in the reconstruction process as latent variables, which form a latent domain. Second, via adversarial learning, we make the distributions of latent variables being indistinguishable between the reconstruction process of the face photograph and that of the face sketch. Finally, given an input face photograph, the latent variable obtained by reconstructing this face photograph is applied for synthesizing the corresponding sketch. Quantitative comparisons to the state-of-the-art methods demonstrate the superiority of the proposed MDAL method.

The dominant neural machine translation (NMT) models apply unified attentional encoder-decoder neural networks for translation. Traditionally, the NMT decoders adopt recurrent neural networks (RNNs) to perform translation in a left-to-right manner, leaving the target-side contexts generated from right to left unexploited during translation. In this paper, we equip the conventional attentional encoder-decoder NMT framework with a backward decoder, in order to explore bidirectional decoding for NMT. Attending to the hidden state sequence produced by the encoder, our backward decoder first learns to generate the target-side hidden state sequence from right to left. Then, the forward decoder performs translation in the forward direction, while in each translation prediction timestep, it simultaneously applies two attention models to consider the source-side and reverse target-side hidden states, respectively. With this new architecture, our model is able to fully exploit source- and target-side contexts to improve translation quality altogether. Experimental results on NIST Chinese-English and WMT English-German translation tasks demonstrate that our model achieves substantial improvements over the conventional NMT by 3.14 and 1.38 BLEU points, respectively. The source code of this work can be obtained from https://github.com/DeepLearnXMU/ABDNMT.

Most image captioning models focus on one-line (single image) captioning, where the correlations like relevance and diversity among group images (e.g., within the same album or event) are simply neglected, resulting in less accurate and diverse captions. Recent works mainly consider imposing the diversity during the online inference only, which neglect the correlation among visual structures in offline training. In this paper, we propose a novel group-based image captioning scheme (termed GroupCap), which jointly models the structured relevance and diversity among group images towards an optimal collaborative captioning. In particular, we first propose a visual tree parser (VP-Tree) to construct the structured semantic correlations within individual images. Then, the relevance and diversity among images are well modeled by exploiting the correlations among their tree structures. Finally, such correlations are modeled as constraints and sent into the LSTM-based captioning generator. We adopt an end-to-end formulation to train the visual tree parser, the structured relevance and diversity constraints, as well as the LSTM based captioning model jointly. To facilitate quantitative evaluation, we further release two group captioning datasets derived from the MS-COCO benchmark, serving as the first of their kind. Quantitative results show that the proposed GroupCap model outperforms the state-of-the-art and alternative approaches.

3D shape recognition has attracted much attention recently. Its recent advances advocate the usage of deep features and achieve the state-of-the-art performance. However, existing deep features for 3D shape recognition are restricted to a view-to-shape setting, which learns the shape descriptor from the view-level feature directly. Despite the exciting progress on view-based 3D shape description, the intrinsic hierarchical correlation and discriminability among views have not been well exploited, which is important for 3D shape representation. To tackle this issue, in this paper, we propose a group-view convolutional neural network (GVCNN) framework for hierarchical correlation modeling towards discriminative 3D shape description. The proposed GVCNN framework is composed of a hierarchical view-group-shape architecture, i.e., from the view level, the group level and the shape level, which are organized using a grouping strategy. Concretely, we first use an expanded CNN to extract a view level descriptor. Then, a grouping module is introduced to estimate the content discrimination of each view, based on which all views can be splitted into different groups according to their discriminative level. A group level description can be further generated by pooling from view descriptors. Finally, all group level descriptors are combined into the shape level descriptor according to their discriminative weights. Experimental results and comparison with state-of-the-art methods show that our proposed GVCNN method can achieve a significant performance gain on both the 3D shape classification and retrieval tasks.

Weakly supervised object detection has attracted extensive research efforts in recent years. Without the need of annotating bounding boxes, the existing methods usually follow a two/multi-stage pipeline with an online compulsive stage to extract object proposals, which is an order of magnitude slower than fast fully supervised object detectors such as SSD [31] and YOLO [34]. In this paper, we speedup online weakly supervised object detectors by orders of magnitude by proposing a novel generative adversarial learning paradigm. In the proposed paradigm, the generator is a one-stage object detector to generate bounding boxes from images. To guide the learning of object-level generator, a surrogator is introduced to mine high-quality bounding boxes for training. We further adapt a structural similarity loss in combination with an adversarial loss into the training objective, which solves the challenge that the bounding boxes produced by the surrogator may not well capture their ground truth. Our one-stage detector outperforms all existing schemes in terms of detection accuracy, running at 118 frames per second, which is up to 438× faster than the state-of-the-art weakly supervised detectors [8, 30, 15, 27, 45]. The code will be available publicly soon.

Despite great effectiveness of very deep and wide Convolutional Neural Networks (CNNs) in various computer vision tasks, the significant cost in terms of storage requirement of such networks impedes the deployment on computationally limited devices. In this paper, we propose new modulated convolutional networks (MCNs) to improve the portability of CNNs via binarized filters. In MCNs, we propose a new loss function which considers the filter loss, center loss and softmax loss in an end-to-end framework. We first introduce modulation filters (M-Filters) to recover the unbinarized filters, which leads to a new architecture to calculate the network model. The convolution operation is further approximated by considering intra-class compactness in the loss function. As a result, our MCNs can reduce the size of required storage space of convolutional filters by a factor of 32, in contrast to the full-precision model, while achieving much better performances than state-of-the-art binarized models. Most importantly, MCNs achieve a comparable performance to the full-precision Resnets and WideResnets. The code will be available publicly soon.

In this paper, we introduce a probabilistic outlier model which is seamlessly integrated into machine learning frameworks (e.g., boosting and deep neural network) to accurately identify outliers in training samples. With two Gamma mixtures, the proposed model can estimate the distribution of inlier and outlier samples respectively and generates their posterior probabilities. The iterative removal method gradually reduces the outliers and boosts the performance of classifier by correcting the bias and variance caused by outliers. Experimental results on INRIA pedestrian dataset and MNIST handwritten-digit dataset demonstrate that the proposed outlier removal method achieves superior detection performance.

Fisher Vector (FV) has been widely used to aggregate the local descriptors of an image into a global representation in large-scale image retrieval. However, FV has limited learning capability and its parameters are mostly fixed after constructing the codebook, which is inflexible and cannot be trained jointly with deep networks. Moreover, the high dimension of FV makes it difficult to be applied in scenarios compact descriptors are needed. In this paper, we propose a novel compact image description scheme based on Fisher network with binary embedding to solve the large-scale image retrieval problem, which consists of two components: a Fisher encoder component and a binary embedding component. Concretely, the Fisher encoder is a trainable neural network functions as the traditional FV, which aggregates the local descriptors into a global representation. And the binary encoder embeds the high-dimensional FV to a binary vector, which outputs the compact global binary descriptor. To learn such a descriptor, we further introduce a novel and effective loss function, in which maximum margin criterion is exploited to minimize the distances of positive pairs, as well as maximizing the distances of negative pairs. Extensive experiments performed on MPEG-7 CDVS benchmarks and ILSVR2010 demonstrate that the proposed framework can achieve very superior performance over the state-of-the-art methods.

This paper focuses on indoor semantic segmentation using RGB-D data. It has been shown that incorporating depth information into RGB information is helpful to improve segmentation accuracy. However, previous studies have revealed two problems. One is about the model size. Recent state-of-the-art methods generally build a network branch for depth images, inherently increasing the model size. The other is about boundary segmentation. The complex and various object configurations with severe occlusions influence the segmentation precision of object boundaries. To address these two problems, we propose a depth-assisted RefineNet (D-RefineNet) for refining the boundary segmentation. The proposed network only uses RGB images to predict segmentation results. Depth images are just used in the proposed loss function without increasing the model size. When the depth values of adjacent pixels change drastically but the adjacent pixels have the same predicted semantic labels, the proposed loss function penalizes the predicted result. Experimental evaluations demonstrate that the proposed method is effective on two challenging RGB-D indoor datasets, NYUDv2 and SUN RGB-D.

Person re-identification (Re-ID) is an important task in video surveillance which automatically searches and identifies people across different cameras. Despite the extensive Re-ID progress in RGB cameras, few works have studied the Re-ID between infrared and RGB images, which is essentially a cross-modality problem and widely encountered in real-world scenarios. The key challenge lies in two folds, i.e., the lack of discriminative information to re-identify the same person between RGB and infrared modalities, and the difficulty to learn a robust metric towards such a large-scale cross-modality retrieval. In this paper, we tackle the above two challenges by proposing a novel cross-modality generative adversarial network (termed cmGAN). To handle the issue of insufficient discriminative information, we leverage the cutting-edge generative adversarial training to design our own discriminator to learn discriminative feature representation from different modalities. To handle the issue of large-scale cross-modality metric learning, we integrates both identification loss and cross-modality triplet loss, which minimize inter-class ambiguity while maximizing cross-modality similarity among instances. The entire cmGAN can be trained in an end-to-end manner by using standard deep neural network framework. We have quantized the performance of our work in the newly-released SYSU RGB-IR Re-ID benchmark, and have reported superior performance, i.e., Cumulative Match Characteristic curve (CMC) and Mean Average Precision (MAP), over the state-of-the-art works [Wu et al., 2017], respectively.

Accelerating convolutional neural networks has recently received ever-increasing research focus. Among various approaches proposed in the literature, filter pruning has been regarded as a promising solution, which is due to its advantage in significant speedup and memory reduction of both network model and intermediate feature maps. To this end, most approaches tend to prune filters in a layer-wise fixed manner, which is incapable to dynamically recover the previously removed filter, as well as jointly optimize the pruned network across layers. In this paper, we propose a novel global & dynamic pruning (GDP) scheme to prune redundant filters for CNN acceleration. In particular, GDP first globally prunes the unsalient filters across all layers by proposing a global discriminative function based on prior knowledge of filters. Second, it dynamically updates the filter saliency all over the pruned sparse network, and then recover the mistakenly pruned filter, followed by a retraining phase to improve the model accuracy. Specially, we effectively solve the corresponding non-convex optimization problem of the proposed GDP via stochastic gradient descent with greedy alternative updating. Extensive experiments show that, comparing to the state-of-the-art filter pruning methods, the proposed approach achieves superior performance to accelerate several cutting-edge CNNs on the ILSVRC 2012 benchmark.

Despite the extensive progress in face sketch synthesis, existing methods are mostly workable under constrained conditions, such as fixed illumination, pose, background and ethnic origin that are hardly to control in real-world scenarios. The key issue lies in the difficulty to use data under fixed conditions to train a model against imaging variations. In this paper, we propose a novel generative adversarial network termed pGAN, which can generate face sketches efficiently using training data under fixed conditions and handle the aforementioned uncontrolled conditions. In pGAN, we embed key photo priors into the process of synthesis and design a parametric sigmoid activation function for compensating illumination variations. Compared to the existing methods, we quantitatively demonstrate that the proposed method can work well on face photos in the wild.

Fine-grained object retrieval has attracted extensive research focus recently. Its state-of-the-art schemesare typically based upon convolutional neural network (CNN) features. Despite the extensive progress, two issues remain open. On one hand, the deep features are coarsely extracted at image level rather than precisely at object level, which are interrupted by background clutters. On the other hand, training CNN features with a standard triplet loss is time consuming and incapable to learn discriminative features. In this paper, we present a novel fine-grained object retrieval scheme that conquers these issues in a unified framework. Firstly, we introduce a novel centralized ranking loss (CRL), which achieves a very efficient (1,000times training speedup comparing to the triplet loss) and discriminative feature learning by a ?centralized? global pooling. Secondly, a weakly supervised attractive feature extraction is proposed, which segments object contours with top-down saliency. Consequently, the contours are integrated into the CNN response map to precisely extract features ?within? the target object. Interestingly, we have discovered that the combination of CRL and weakly supervised learning can reinforce each other. We evaluate the performance ofthe proposed scheme on widely-used benchmarks including CUB200-2011 and CARS196. We havereported significant gains over the state-of-the-art schemes, e.g., 5.4% over SCDA [Wei et al., 2017]on CARS196, and 3.7% on CUB200-2011.

In recent years, binary code learning, a.k.a. hashing, has received extensive attention in large-scale multimedia retrieval. It aims to encode high-dimensional data points into binary codes, hence the original high-dimensional metric space can be efficiently approximated via Hamming space. However, most existing hashing methods adopted offline batch learning, which is not suitable to handle incremental datasets with streaming data or new instances. In contrast, the robustness of the existing online hashing remains as an open problem, while the embedding of supervised/semantic information hardly boosts the performance of the online hashing, mainly due to the defect of unknown category numbers in supervised learning. In this paper, we propose an online hashing scheme, termed Hadamard Codebook based Online Hashing (HCOH), which aims to solving the above problems towards robust and supervised online hashing. In particular, we first assign an appropriate high-dimensional binary codes to each class label, which is generated randomly by Hadamard codes. Subsequently, LSH is adopted to reduce the length of such Hadamard codes in accordance with the hash bits, which can adapt the predefined binary codes online, and theoretically guarantee the semantic similarity. Finally, we consider the setting of stochastic data acquisition, which facilitates our method to efficiently learn the corresponding hashing functions via stochastic gradient descend (SGD) online. Notably, the proposed HCOH can be embedded with supervised labels and is not limited to a predefined category number. Extensive experiments on three widely-used benchmarks demonstrate the merits of the proposed scheme over the state-of-the-art methods.

Cross-modality retrieval has been widely studied, which aims to search images as response to text queries or vice versa. When faced with large-scale dataset, cross-modality hashing serves as an efficient and effective solution, which learns binary codes to approximate the cross-modality similarity in the Hamming space. Most recent cross-modality hashing schemes focus on learning the hash functions from data instances with fully modalities. However, how to learn robust binary codes when facing incomplete modality (i.e., with one modality missed or partially observed), is left unexploited, which however widely occurs in real-world applications. In this paper, we propose a novel cross-modality hashing, termed Dense Auto-encoder Hashing (DAH), which can explicitly impute the missed modality and produce robust binary codes by leveraging the relatedness among different modalities. To that effect, we propose a novel Dense Auto-encoder Network (DAN) to impute the missing modalities, which densely connects each layer to every other layer in a feed-forward fashion. For each layer, a noisy auto-encoder block is designed to calculate the residue between the current prediction and original data. Finally, a hash-layer is added to the end of DAN, which serves as a special binary encoder model to deal with the incomplete modality input. Quantitative experiments on three cross-modality visual search benchmarks, i.e., the Wiki, NUS-WIDE, and FLICKR-25K, have shown that the proposed DAH has superior performance over the state-of-the-art approaches.