3D captioning, which aims to describe the content of 3D scenes in natural language, remains highly challenging due to the inherent sparsity of point clouds and weak cross-modal alignment in existing methods. To address these challenges, we propose 3D CoCa, a novel unified framework that seamlessly combines contrastive vision-language learning with 3D caption generation into a single architecture. We design a frozen CLIP vision-language backbone to provide rich semantic priors, a spatially-aware 3D scene encoder to capture geometric context, and a multi-modal decoder to generate descriptive captions. Unlike the prior two-stage methods that rely on explicit object proposals, 3D CoCa jointly optimizes contrastive and captioning objectives in a shared feature space, eliminating the need for external detectors or handcrafted proposals. This joint training paradigm yields stronger spatial reasoning and richer semantic grounding by aligning 3D and textual representations. Extensive experiments on the ScanRefer and Nr3D benchmarks demonstrate that 3D CoCa significantly outperforms current state-of-the-arts by 10.2% and 5.76% in [email protected], respectively. Code will be available at https://github.com/AIGeeksGroup/3DCoCa.

Gaussian fields are a promising representation for robot body modeling due to their differentiability and inherently low sim-to-real gap. However, existing methods like Dr-Robot overlook explicit geometric constraints, leading to artifacts under novel poses or views. Directly enforcing depth and normal supervision on articulated Gaussians is unstable due to entanglement between pose deformation and 3D appearance learning. To address this, we propose a two-stage training strategy: we first learn a canonical Gaussian field in a canonical pose using dense RGB, depth, and normal supervision, establishing a geometryaware reconstruction. We then fine-tune the Gaussian parameters jointly with a deformation network conditioned on joint angles using only RGB losses, ensuring consistent geometry and appearance across poses. To further mitigate rendering artifacts in novel poses and viewpoints, we integrate a diffusion-based refinement module. This module conditions on both the initial Gaussian renderings and the target robot skeletons, and significantly enhances visual fidelity while preserving pose accuracy. Experiments across multiple robotic platforms show that GRADRobot outperforms DrRobot by a large margin in both rendering quality (PSNR) and geometric accuracy (Chamfer Distance). https://github.com/liyunlooong/GRADRobot

Enabling multi-task adaptation in pre-trained Low-Rank Adaptation (LoRA) models is crucial for enhancing their generalization capabilities. Most existing pre-trained LoRA fusion methods decompose weight matrices, sharing similar parameters, while fusion divergent ones. However, this paradigm inevitably induces inter-weight conflicts and leads to catastrophic domain forgetting. While incremental learning enables adaptation to multiple tasks, it struggles to achieve generalization in few-shot scenarios. Consequently, when the weight data follows a long-tailed distribution, it can lead to forgetting in the fused weights. To address this issue, we propose In-Context Meta LoRA Fusion (ICM-Fusion), a novel framework that synergizes meta-learning with in-context adaptation. The key innovation lies in our task vector arithmetic, which dynamically balances conflicting optimization directions across domains through learned manifold projections. ICM-Fusion obtains the optimal task vector orientation for the fused model in the latent space by adjusting the orientation of the task vectors. Subsequently, the fused LoRA is reconstructed by a self-designed Fusion VAE (F-VAE) to realize multi-task LoRA generation. We have conducted extensive experiments on visual and linguistic tasks, and the experimental results demonstrate that ICM-Fusion can be adapted to a wide range of architectural models and applied to various tasks. Compared to the current pre-trained LoRA fusion method, ICM-Fusion fused LoRA can significantly reduce the multi-tasking loss and can even achieve task enhancement in few-shot scenarios.

Diffusion models have been widely adopted in image and video generation. However, their complex network architecture leads to high inference overhead for its generation process. Existing diffusion quantization methods primarily focus on the quantization of the model structure while ignoring the impact of time-steps variation during sampling. At the same time, most current approaches fail to account for significant activations that cannot be eliminated, resulting in substantial performance degradation after quantization. To address these issues, we propose Time-Rotation Diffusion Quantization (TR-DQ), a novel quantization method incorporating time-step and rotation-based optimization. TR-DQ first divides the sampling process based on time-steps and applies a rotation matrix to smooth activations and weights dynamically. For different time-steps, a dedicated hyperparameter is introduced for adaptive timing modeling, which enables dynamic quantization across different time steps. Additionally, we also explore the compression potential of Classifier-Free Guidance (CFG-wise) to establish a foundation for subsequent work. TR-DQ achieves state-of-the-art (SOTA) performance on image generation and video generation tasks and a 1.38-1.89× speedup and 1.97-2.58× memory reduction in inference compared to existing quantization methods.

Severe weather restoration models often face the simultaneous interaction of multiple degradations in real-world scenarios. Existing approaches typically handle single or composite degradations based on scene descriptors derived from text or image embeddings. However, due to the varying proportions of different degradations within an image, these scene descriptors may not accurately differentiate between degradations, leading to suboptimal restoration in practical applications. To address this issue, we propose a novel Transformer-based restoration framework, AllRestorer, for dealing with four physical severe weather impairments: low-light, haze, rain, and snow. In AllRestorer, we enable the model to adaptively consider all weather impairments, thereby avoiding errors from scene descriptor misdirection. Specifically, we introduce the All-in-One Transformer Block (AiOTB), the core innovation of which is the ability to adaptively handle multiple degradations in a single image, beyond the limitation of existing Transformers that can only handle one type of degradation at a time. To accurately address different variations potentially present within the same type of degradation and minimize ambiguity, AiOTB utilizes a Composite Scene Embedding consisting of both image and text embeddings to define the degradation. Moreover, AiOTB includes an adaptive weight for each degradation, allowing for precise control of the restoration intensity. By leveraging AiOTB, AllRestorer avoids misdirection caused by inaccurate scene descriptors, achieving a 5.00 dB increase in PSNR compared to the baseline on the CDD-11 dataset.

With the rapid rise of short video social platforms, the spread of fake news videos has become a global challenge. Short videos, which integrate multiple modalities such as text, images, and audio, have a powerful visual and auditory impact, making fake news more prone to widespread dissemination and causing serious societal consequences. However, the complex fusion of multimodal information in fake news videos, coupled with editing artifacts that often blur the distinction between real and fake content, presents considerable challenges to traditional detection methods. To address these challenges, this paper proposes a fake news video detection method based on the Knowledge-Enhanced Dynamic Scene Graph Attention Network (KDSGAT). This method captures temporal correlations and local semantic differences in visual scenes by leveraging dynamic scene graph networks, while enhancing semantic understanding through knowledge distillation from external knowledge graphs. Specifically, we first use pre-trained models such as BERT, HuBERT, and Swin Transformer to extract text semantic features, audio emotion features, and visual features, respectively. Next, we apply an unbiased scene graph generation approach to convert keyframes from the video into scene graphs, which are then processed by the dynamic scene graph attention network to capture temporal correlations and local semantic variations within the scene graph sequences. Finally, co-attention is used to interactively fuse multimodal features, enabling precise detection of fake news in videos. We conduct extensive experiments on two real-world datasets from short video social platforms, FakeSV and FakeTT. The results show that our method outperforms state-of-the-art baselines, improving accuracy by 1.86% and 2.68% on the two datasets, respectively. The source code and data are available athttps://github.com/xuejianhuang/KDSGAT-FNVD.

To integrate digital humans into everyday life, there is a strong demand for generating high-quality, fine-grained disentangled 3D avatars that support expressive animation and simulation capabilities, ideally from low-cost textual inputs. Although text-driven 3D avatar generation has made significant progress by leveraging 2D generative priors, existing methods still struggle to fulfill all these requirements simultaneously. To address this challenge, we propose DreamBarbie, a novel text-driven framework for generating animatable 3D avatars with separable shoes, accessories, and simulation-ready garments, truly capturing the iconic "Barbie doll" aesthetic. The core of our framework lies in an expressive 3D representation combined with appropriate modeling constraints. Unlike prior methods, we use G-Shell to uniformly model watertight components (e.g., bodies, shoes) and non-watertight garments. By reformulating boundaries as euclidean field intersections instead of manifold geodesics, we propose an SDF-based initialization and a hole regularization loss that together achieve a $100\times$100× speedup and stable open topology without image input. These disentangled 3D representations are then optimized by specialized expert diffusion models tailored to each domain, ensuring high-fidelity outputs. To mitigate geometric artifacts and texture conflicts when combining different expert models, we further propose several effective geometric losses and strategies. Extensive experiments demonstrate that DreamBarbie outperforms existing methods in both dressed human and outfit generation. Our framework further enables diverse applications, including apparel combination, editing, expressive animation, and physical simulation.

Fine-Grained Sketch-Based Image Retrieval (FG-SBIR) aims to minimize the distance between sketches and corresponding images in the embedding space. However, scalability is hindered by the growing complexity of solutions, mainly due to the abstract nature of fine-grained sketches. In this paper, we propose an effective approach to narrow the gap between the two domains. It mainly facilitates unified mutual information sharing both intra- and inter-samples, rather than treating them as a single feature alignment problem between modalities. Specifically, our approach includes: (i) Employing dual weight-sharing networks to optimize alignment within the sketch and image domain, which also effectively mitigates model learning saturation issues. (ii) Introducing an objective optimization function based on contrastive loss to enhance the model's ability to align features in both intra- and inter-samples. (iii) Presenting a self-supervised Multi-Scale Token Recycling (MSTR) Module featured by recycling discarded patch tokens in multi-scale features, further enhancing representation capability and retrieval performance. Our framework achieves excellent results on CNN- and ViT-based backbones. Extensive experiments demonstrate its superiority over existing methods. We also introduce Cloths-V1, the first professional fashion sketch-image dataset, utilized to validate our method and will be beneficial for other applications.

Structured pruning for large language models (LLMs) has garnered significant academic interest due to its ability to efficiently compress and accelerate LLMs by eliminating redundant weight groups at a coarse-grained granularity. Current structured pruning methods for LLMs typically depend on a singular granularity for assessing weight importance, resulting in notable performance degradation in downstream tasks. Intriguingly, our empirical investigations reveal that utilizing unstructured pruning, which achieves better performance retention by pruning weights at a finer granularity, \emph{i.e.}, individual weights, yields significantly varied sparse LLM structures when juxtaposed to structured pruning. This suggests that evaluating both holistic and individual assessments for weight importance are essential for LLM pruning. Building on this insight, we introduce the Hybrid-grained Weight Importance Assessment (HyWIA), a novel method that merges fine-grained and coarse-grained evaluations of weight importance for the pruning of LLMs. Leveraging an attention mechanism, HyWIA adaptively determines the optimal blend of granularity in weight importance assessments in an end-to-end pruning manner. Extensive experiments on LLaMA-V1/V2, Vicuna, Baichuan, and Bloom across various benchmarks demonstrate the effectiveness of HyWIA in pruning LLMs. For example, HyWIA surpasses the cutting-edge LLM-Pruner by an average margin of 2.82% in accuracy across seven downstream tasks when pruning LLaMA-7B by 50%.

Current hair transfer methods struggle to handle diverse and intricate hairstyles, limiting their applicability in real-world scenarios. In this paper, we propose a novel diffusion-based hair transfer framework, named Stable-Hair, which robustly transfers a wide range of real-world hairstyles to user-provided faces for virtual hair try-on. To achieve this goal, our Stable-Hair framework is designed as a two-stage pipeline. In the first stage, we train a Bald Converter alongside stable diffusion to remove hair from the user-provided face images, resulting in bald images. In the second stage, we specifically designed a Hair Extractor and a Latent IdentityNet to transfer the target hairstyle with highly detailed and high-fidelity to the bald image. The Hair Extractor is trained to encode reference images with the desired hairstyles, while the Latent IdentityNet ensures consistency in identity and background. To minimize color deviations between source images and transfer results, we introduce a novel Latent ControlNet architecture, which functions as both the Bald Converter and Latent IdentityNet. After training on our curated triplet dataset, our method accurately transfers highly detailed and high-fidelity hairstyles to the source images. Extensive experiments demonstrate that our approach achieves state-of-the-art performance compared to existing hair transfer methods.

As interest grows in world models that predict future states from current observations and actions, accurately modeling part-level dynamics has become increasingly relevant for various applications. Existing approaches, such as Puppet-Master, rely on fine-tuning large-scale pre-trained video diffusion models, which are impractical for real-world use due to the limitations of 2D video representation and slow processing times. To overcome these challenges, we present PartRM, a novel 4D reconstruction framework that simultaneously models appearance, geometry, and part-level motion from multi-view images of a static object. PartRM builds upon large 3D Gaussian reconstruction models, leveraging their extensive knowledge of appearance and geometry in static objects. To address data scarcity in 4D, we introduce the PartDrag-4D dataset, providing multi-view observations of part-level dynamics across over 20,000 states. We enhance the model’s understanding of interaction conditions with a multi-scale drag embedding module that captures dynamics at varying granularities. To prevent catastrophic forgetting during fine-tuning, we implement a two-stage training process that focuses sequentially on motion and appearance learning. Experimental results show that PartRM establishes a new state-of-the-art in part-level motion learning and can be applied in manipulation tasks in robotics. Our code, data, and models are publicly available to facilitate future research.

We introduce DiffFNO, a novel diffusion framework for arbitrary-scale super-resolution strengthened by a Weighted Fourier Neural Operator (WFNO). Mode Re-balancing in WFNO effectively captures critical frequency components, significantly improving the reconstruction of high-frequency image details that are crucial for super-resolution tasks. Gated Fusion Mechanism (GFM) adaptively complements WFNO’s spectral features with spatial features from an Attention-based Neural Operator (AttnNO). This enhances the network’s capability to capture both global structures and local details. Adaptive Time-Step (ATS) ODE solver, a deterministic sampling strategy, accelerates inference without sacrificing output quality by dynamically adjusting integration step sizes ATS. Extensive experiments demonstrate that DiffFNO achieves state-of-the-art (SOTA) results, outperforming existing methods across various scaling factors by a margin of 2–4 dB in PSNR, including those beyond the training distribution. It also achieves this at lower inference time (Fig. 1 (a)). Our approach sets a new standard in super-resolution, delivering both superior accuracy and computational efficiency.

A high-performance image compression algorithm is crucial for real-time information transmission across numerous fields. Despite rapid progress in image compression, computational inefficiency and poor redundancy modeling still pose significant bottlenecks, limiting practical applications. Inspired by the effectiveness of state space models (SSMs) in capturing long-range dependencies, we leverage SSMs to address computational inefficiency in existing methods and improve image compression from multiple perspectives. In this paper, we integrate the advantages of SSMs for better efficiency-performance trade-off and propose an enhanced image compression approach through refined context modeling, which we term MambaIC. Specifically, we explore context modeling to adaptively refine the representation of hidden states. Additionally, we introduce window-based local attention into channel-spatial entropy modeling to reduce potential spatial redundancy during compression, thereby increasing efficiency. Comprehensive qualitative and quantitative results validate the effectiveness and efficiency of our approach, particularly for high-resolution image compression. Code is released at https://github.com/AuroraZengfh/MambaIC.

Fine-tuning helps large language models (LLM) recover degraded information and enhance task performance. Although Low-Rank Adaptation (LoRA) is widely used and effective for fine-tuning, we have observed that its scaling factor can limit or even reduce performance as the rank size increases. To address this issue, we propose RoRA (Rank-adaptive Reliability Optimization), a simple yet effective method for optimizing LoRA’s scaling factor. By replacing α/r with $\alpha /\sqrt r $, RoRA ensures improved performance as rank size increases. Moreover, RoRA enhances low-rank adaptation in fine-tuning uncompressed models and excels in the more challenging task of accuracy recovery when fine-tuning pruned models. Extensive experiments demonstrate the effectiveness of RoRA in fine-tuning both uncompressed and pruned models. RoRA surpasses the state-of-the-art (SOTA) in average accuracy and robustness on LLaMA-7B/13B, LLaMA2-7B, and LLaMA3-8B, specifically outperforming LoRA and DoRA by 6.5% and 2.9% on LLaMA-7B, respectively. In pruned model fine-tuning, RoRA shows significant advantages; for SHEARED-LLAMA-1.3, a LLaMA-7B with 81.4% pruning, RoRA achieves 5.7% higher average accuracy than LoRA and 3.9% higher than DoRA.

In recent years, the introduction of Multi-modal Large Language Models (MLLMs) into video understanding tasks has become increasingly prevalent. However, how to effectively integrate temporal information remains a critical research focus. Traditional approaches treat spatial and temporal information separately. Due to issues like motion blur, it is challenging to accurately represent the spatial information of rapidly moving objects. This can lead to temporally important regions being underemphasized during spatial feature extraction, which in turn hinders accurate spatio-temporal interaction and video understanding. To address this limitation, we propose an innovative video representation method called Dynamic-Image (DynImg). Specifically, we introduce a set of non-key frames as temporal prompts to highlight the spatial areas containing fast-moving objects. During the process of visual feature extraction, these prompts guide the model to pay additional attention to the fine-grained spatial features corresponding to these regions. Moreover, to maintain the correct sequence for DynImg, we employ a corresponding 4D video Rotary Position Embedding. This retains both the temporal and spatial adjacency of DynImg, helping MLLM understand the spatio-temporal order within this combined format. Experimental evaluations reveal that DynImg surpasses the state-of-the-art methods by approximately 2% across multiple video understanding benchmarks, proving the effectiveness of our temporal prompts in enhancing video comprehension.

Robot-assisted surgery has profoundly influenced current forms of minimally invasive surgery. However, in transurethral urological surgical robots, they need to work in a liquid environment. This causes vaporization of the liquid when shearing and heating is performed, resulting in bubble atomization that affects the visual perception of the robot. This can lead to the need for uninterrupted pauses in the surgical procedure, which makes the surgery take longer. To address the atomization characteristics of liquids under urological surgical robotic vision, we propose an unsupervised zero-shot dehaze method (RSF-Dehaze). Specifically, the proposed Region Similarity Filling Module (RSFM) of RSF-Dehaze significantly improves the recovery of blurred region tissues. In addition, we organize and propose a dehaze dataset for robotic vision in urological surgery (USRobot-Dehaze dataset). In particular, this dataset contains the three most common urological surgical robot operation scenarios. To the best of our knowledge, we are the first to organize and propose a publicly available dehaze dataset for urological surgical robot vision. The proposed RSF-Dehaze proves the effectiveness of our method in three urological surgical robot operation scenarios with extensive comparative experiments with 20 most classical and advanced dehazing and image recovery algorithms. The proposed source code and dataset are available at https://github.com/wurenkai/RSF-Dehaze.

Diffusion-based image super-resolution (SR) methods have demonstrated remarkable performance. Recent advancements have introduced deterministic sampling processes that reduce inference from 15 iterative steps to a single step, thereby significantly improving the inference speed of existing diffusion models. However, their efficiency remains limited when handling complex semantic regions due to the single-step inference. To address this limitation, we propose SAMSR, a semantic-guided diffusion framework that incorporates semantic segmentation masks into the sampling process. Specifically, we introduce the SAM-Noise Module, which refines Gaussian noise using segmentation masks to preserve spatial and semantic features. Furthermore, we develop a pixel-wise sampling strategy that dynamically adjusts the residual transfer rate and noise strength based on pixel-level semantic weights, prioritizing semantically rich regions during the diffusion process. To enhance model training, we also propose a semantic consistency loss, which aligns pixel-wise semantic weights between predictions and ground truth. Extensive experiments on both real-world and synthetic datasets demonstrate that SAMSR significantly improves perceptual quality and detail recovery, particularly in semantically complex images.

Low-rank Adaptation (LoRA) has demonstrated remarkable capabilities for task specific fine-tuning. However, in scenarios that involve multiple tasks, training a separate LoRA model for each one results in considerable inefficiency in terms of storage and inference. Moreover, existing parameter generation methods fail to capture the correlations among these tasks, making multi-task LoRA parameter generation challenging. To address these limitations, we propose In-Context Meta LoRA (ICM-LoRA), a novel approach that efficiently achieves task-specific customization of large language models (LLMs). Specifically, we use training data from all tasks to train a tailored generator, Conditional Variational Autoencoder (CVAE). CVAE takes task descriptions as inputs and produces task-aware LoRA weights as outputs. These LoRA weights are then merged with LLMs to create task-specialized models without the need for additional fine-tuning. Furthermore, we utilize in-context meta-learning for knowledge enhancement and task mapping, to capture the relationship between tasks and parameter distributions. As a result, our method achieves more accurate LoRA parameter generation for diverse tasks using CVAE. ICM-LoRA enables more accurate LoRA parameter reconstruction than current parameter reconstruction methods and is useful for implementing task-specific enhancements of LoRA parameters. At the same time, our method occupies 283MB, only 1% storage compared with the original LoRA. The code is available at https://github.com/YihuaJerry/ICM-LoRA.

Real‐world datasets usually contain multiple attributes, making it essential to ensure fairness across all of them simultaneously. However, different attributes may vary in difficulty, and no existing approaches have effectively addressed this issue. Consequently, an attribute‐adaptive strategy is needed to achieve fairness for all attributes. Multi‐task Learning (MTL) leverages shared information to optimize multiple tasks concurrently, while Sparsely‐Gated Mixture‐of‐Experts (SMoE) can dynamically allocate computational resources to the most needed tasks. In this work, we formulate multi‐attribute fairness issue as an MTL problem and employ SMoE to achieve desirable performance across all attributes simultaneously. We first analyze the feasibility and find the potentiality by formalizing multi-attribute fairness problem into a MTL problem and mitigating it by using SMoE. However, vanilla SMoE could lead to over-utilization problem which causes sub-optimal performance. We then proposed an innovative SMoE framework for multi-attribute fair image classification, which further improves multi-attribute fairness by redesigning the MoE layer and routing policy with fairness consideration. Extensive experiments demonstrated the effectiveness. Taking a DeiT-Small as the backbone, we achieve 77.25% and 86.01% accuracy on the ISIC2019 and CelebA dataset respectively with Multi-attribute Predictive Quality Disparity (PQD) score of 0.801 and 0.787, beating current state-of-the-art methods Muffin, InfoFair and MultiFair.

Data-Free Class Incremental Learning (DFCIL) aims to enable models to continuously learn new classes while retaining knowledge of old classes, even when the training data for old classes is unavailable. Although explored primarily with image datasets by researchers, this study focuses on investigating DFCIL for skeleton-based gesture classification due to its significant real-world implications, particularly considering gestures serve as one of the primary means of control and interaction in VR/AR, smart home systems and beyond. In this work, we made an intriguing observation: skeleton models trained with base classes(even very limited) demonstrate strong generalization capabilities to unseen classes without requiring additional training. Building on this insight, we developed a new model-updating paradigm - Synthetic Feature Replay (SFR) that directly synthesize fake features within the embedding space from saved class prototypes, instead of generating synthetic data, to replay for old classes and augment for new classes (under a few-shot setting). Our proposed method showcases significant advancements over the state-of-the-art, achieving up to 15% enhancements in mean accuracy across all steps and largely mitigating the accuracy imbalance between base classes and new classes.

With the increasing use of surgical robots in clinical practice, enhancing their ability to process multimodal medical images has become a key research challenge. Although traditional medical image fusion methods have made progress in improving fusion accuracy, they still face significant challenges in real-time performance, fine-grained feature extraction, and edge preservation. In this paper, we introduce TTTFusion, a Test-Time Training (TTT)-based image fusion strategy that dynamically adjusts model parameters during inference to efficiently fuse multimodal medical images. By adapting the model during the test phase, our method optimizes the parameters based on the input image data, leading to improved accuracy and better detail preservation in the fusion results. Experimental results demonstrate that TTTFusion significantly enhances the fusion quality of multimodal images compared to traditional fusion methods, particularly in fine-grained feature extraction and edge preservation. This approach not only improves image fusion accuracy but also offers a novel technical solution for real-time image processing in surgical robots.

Vehicle-to-everything (V2X) communication plays a crucial role in autonomous driving, enabling cooperation between vehicles and infrastructure. While simulation has significantly contributed to various autonomous driving tasks, its potential for data generation and augmentation in V2X scenarios remains underexplored. In this paper, we introduce CRUISE, a comprehensive reconstruction-and-synthesis framework designed for V2X driving environments. CRUISE employs decomposed Gaussian Splatting to accurately reconstruct real-world scenes while supporting flexible editing. By decomposing dynamic traffic participants into editable Gaussian representations, CRUISE allows for seamless modification and augmentation of driving scenes. Furthermore, the framework renders images from both ego-vehicle and infrastructure views, enabling large-scale V2X dataset augmentation for training and evaluation. Our experimental results demonstrate that: 1) CRUISE reconstructs real-world V2X driving scenes with high fidelity; 2) using CRUISE improves 3D detection across ego-vehicle, infrastructure, and cooperative views, as well as cooperative 3D tracking on the V2X-Seq benchmark; and 3) CRUISE effectively generates challenging corner cases. The code will be publicly available at https://github.com/SainingZhang/CRUISE.

In complex transportation systems, accurately sensing the surrounding environment and predicting the risk of potential accidents is crucial. Most existing accident prediction methods are based on temporal neural networks, such as RNN and LSTM. Recent multimodal fusion approaches improve vehicle localization through 3D target detection and assess potential risks by calculating inter-vehicle distances. However, these temporal networks and multimodal fusion methods suffer from limited detection robustness and high economic costs. To address these challenges, we propose AccidentBlip, a vision-only framework that employs our self-designed Motion Accident Transformer (MA-former) to process each frame of video. Unlike conventional self-attention mechanisms, MA-former replaces Q-former's self-attention with temporal attention, allowing the query corresponding to the previous frame to generate the query input for the next frame. Additionally, we introduce a residual module connection between queries of consecutive frames to enhance the model's temporal processing capabilities. For complex V2V and V2X scenarios, AccidentBlip adapts by concatenating queries from multiple cameras, effectively capturing spatial and temporal relationships. In particular, AccidentBlip achieves SOTA performance in both accident detection and prediction tasks on the DeepAccident dataset. It also outperforms current SOTA methods in V2V and V2X scenarios, demonstrating a superior capability to understand complex real-world environments.

Video Anomaly Detection (VAD) focuses on identifying anomalies within videos. Supervised methods require an amount of in-domain training data and often struggle to generalize to unseen anomalies. In contrast, training-free methods leverage the intrinsic world knowledge of large language models (LLMs) to detect anomalies but face challenges in localizing fine-grained visual transitions and diverse events. Therefore, we propose EventVAD, an event-aware video anomaly detection framework that combines tailored dynamic graph architectures and multimodal LLMs to perform fine-grained temporal-event reasoning. Specifically, EventVAD first employs dynamic spatiotemporal graph modeling with time-decay constraints to capture event-aware video features. Then, it performs adaptive noise filtering and uses signal ratio thresholding to detect event boundaries via unsupervised statistical features. Finally, it utilizes a hierarchical prompting strategy to guide MLLMs in performing reasoning and making final decisions. We conducted extensive experiments on the UCF-Crime and XD-Violence datasets. The results demonstrate that EventVAD with a 7B MLLM achieves state-of-the-art (SOTA) in training-free settings, outperforming strong baselines that use 7B or larger MLLMs. The code is available at https://github.com/YihuaJerry/EventVAD.

Xiaofeng Zhang, Yihao Quan, Chen Shen, Xiaosong Yuan, Shaotian Yan, Liang Xie, Wenxiao Wang, Chaochen Gu, Hao Tang, Jieping Ye. 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.

Deep neural networks are vulnerable to adversarial examples that exhibit transferability across various models. Numerous approaches are proposed to enhance the transferability of adversarial examples, including advanced optimization, data augmentation, and model modifications. However, these methods still show limited transferability, partiovovocularly in cross-architecture scenarios, such as from CNN to ViT. To achieve high transferability, we propose a technique termed Spatial Adversarial Alignment (SAA), which employs an alignment loss and leverages a witness model to fine-tune the surrogate model. Specifically, SAA consists of two key parts: spatial-aware alignment and adversarial-aware alignment. First, we minimize the divergences of features between the two models in both global and local regions, facilitating spatial alignment. Second, we introduce a self-adversarial strategy that leverages adversarial examples to impose further constraints, aligning features from an adversarial perspective. Through this alignment, the surrogate model is trained to concentrate on the common features extracted by the witness model. This facilitates adversarial attacks on these shared features, thereby yielding perturbations that exhibit enhanced transferability. Extensive experiments on various architectures on ImageNet show that aligned surrogate models based on SAA can provide higher transferable adversarial examples, especially in cross-architecture attacks.

Recent advances in large vision-language models (LVLMs) have demonstrated strong performance on general-purpose medical tasks. However, their effectiveness in specialized domains such as dentistry remains underexplored. In particular, panoramic X-rays, a widely used imaging modality in oral radiology, pose interpretative challenges due to dense anatomical structures and subtle pathological cues, which are not captured by existing medical benchmarks or instruction datasets. To this end, we introduce MMOral, the first large-scale multimodal instruction dataset and benchmark tailored for panoramic X-ray interpretation. MMOral consists of 20,563 annotated images paired with 1.3 million instruction-following instances across diverse task types, including attribute extraction, report generation, visual question answering, and image-grounded dialogue. In addition, we present MMOral-Bench, a comprehensive evaluation suite covering five key diagnostic dimensions in dentistry. We evaluate 64 LVLMs on MMOral-Bench and find that even the best-performing model, i.e., GPT-4o, only achieves 43.31% accuracy, revealing significant limitations of current models in this domain. To promote the progress of this specific domain, we provide the supervised fine-tuning (SFT) process utilizing our meticulously curated MMOral instruction dataset. Remarkably, a single epoch of SFT yields substantial performance enhancements for LVLMs, e.g., Qwen2.5-VL-7B demonstrates a 24.73% improvement. MMOral holds significant potential as a critical foundation for intelligent dentistry and enables more clinically impactful multimodal AI systems in the dental field.

Preference alignment in diffusion models has primarily focused on benign human preferences (e.g., aesthetic). In this paper, we propose a novel perspective: framing unrestricted adversarial example generation as a problem of aligning with adversary preferences. Unlike benign alignment, adversarial alignment involves two inherently conflicting preferences: visual consistency and attack effectiveness, which often lead to unstable optimization and reward hacking (e.g., reducing visual quality to improve attack success). To address this, we propose APA (Adversary Preferences Alignment), a two-stage framework that decouples conflicting preferences and optimizes each with differentiable rewards. In the first stage, APA fine-tunes LoRA to improve visual consistency using rule-based similarity reward. In the second stage, APA updates either the image latent or prompt embedding based on feedback from a substitute classifier, guided by trajectory-level and step-wise rewards. To enhance black-box transferability, we further incorporate a diffusion augmentation strategy. Experiments demonstrate that APA achieves significantly better attack transferability while maintaining high visual consistency, inspiring further research to approach adversarial attacks from an alignment perspective.

Generalist models have achieved remarkable success in both language and vision-language tasks, showcasing the potential of unified modeling. However, effectively integrating fine-grained perception tasks like detection and segmentation into these models remains a significant challenge. This is primarily because these tasks often rely heavily on task-specific designs and architectures that can complicate the modeling process. To address this challenge, we present UFO, a framework that unifies fine-grained visual perception tasks through an open-ended language interface. By transforming all perception targets into the language space, UFO unifies object-level detection, pixel-level segmentation, and image-level vision-language tasks into a single model. Additionally, we introduce a novel embedding retrieval approach that relies solely on the language interface to support segmentation tasks. Our framework bridges the gap between fine-grained perception and vision-language tasks, significantly simplifying architectural design and training strategies while achieving comparable or superior performance to methods with intricate task-specific designs. After multi-task training on five standard visual perception datasets, UFO outperforms the previous state-of-the-art generalist models by 12.3 mAP on COCO instance segmentation and 3.3 mIoU on ADE20K semantic segmentation. Furthermore, our method seamlessly integrates with existing MLLMs, effectively combining fine-grained perception capabilities with their advanced language abilities, thereby achieving superior performance on the challenging reasoning segmentation. Code and models are available at https://github.com/nnnth/UFO.

Fine-tuning pre-trained models with custom data leads to numerous expert models on specific tasks. Merging models into one universal model to empower multi-task ability refraining from data leakage has gained popularity. With the expansion in data and model size, parameter-efficient tuning becomes the common practice for obtaining task-specific models efficiently. However, few methods are dedicated to efficient merging, and existing methods designed for full fine-tuning merging fail under efficient merging. To address the issue, we analyze from low-rank decomposition and reveal that direction robustness during merging is crucial for merging efficient modules. We furthermore uncover that compensating for the gap between stark singular values contributes to direction robustness. Therefore, we propose RobustMerge, a training-free parameter-efficient merging method with complementary parameter adaptation to maintain direction robustness. Specifically, we (1) prune parameters and scale coefficients from inter-parameter relations for singular values to maintain direction stability away from task interference, and (2) perform cross-task normalization to enhance unseen task generalization. We establish a benchmark consisting of diverse multimodal tasks, on which we conduct experiments to certify the outstanding performance and generalizability of our method. Additional studies and extensive analyses further showcase the effectiveness.

Recent advancements in bird's-eye view (BEV) perception models have highlighted the superior performance of LiDAR-camera fusion systems over single-modality approaches, garnering considerable interest in the field. Despite the progress, the integration of temporal information, a technique that has considerably benefitted camera-only BEV models, remains underexplored for LiDAR-camera fusion. This paper presents Q-TempFusion, a novel approach for temporal multi-sensor fusion designed to enhance the BEV model's inference speed while keeping high predictive performance compared with the current state-of-the-art. Moreover, we are the first to make the multi-modality BEV model profiling on hardware devices. To address the challenges of substantial memory demands and non-trivial latency that hinder deployment in on-vehicle systems, particularly when temporal dynamics are incorporated into complex multi-sensor models, we introduce an activation-aware quantization framework to generate the fully 8-bit quantized Q-TempFusion model based on the profiling result, which can be directly deployed to target devices with negligible detection performance degradation. Our experiments show that our Q-TempFusion (8-bit) achieves 70.3% mAP and 72.7% NDS with 3×~18× FPS improvement over leading multi-modality baselines and the Q-TempFusion (32-bit) achieves 72.1% mAP and 74.8% NDS, comparable to SOTA multi-modality approaches. The results suggest that Q-TempFusion is a promising step toward real-time multi-sensor BEV applications, setting a new benchmark for efficient and reliable perception.

Abstract Despite their impressive performance on various tasks, vision transformers (ViTs) are heavy for mobile vision applications. Recent works have proposed combining the strengths of ViTs and convolutional neural networks (CNNs) to build lightweight networks. Still, these approaches rely on hand-designed architectures with a pre-determined number of parameters. In this work, we address the challenge of finding optimal light-weight ViTs given constraints on model size and computational cost using neural architecture search. We use a search algorithm that considers both model parameters and on-device deployment latency. This method analyzes network properties, hardware memory access pattern, and degree of parallelism to directly and accurately estimate the network latency. To prevent the need for extensive testing during the search process, we use a lookup table based on a detailed breakdown of the speed of each component and operation, which can be reused to evaluate the whole latency of each search structure. Our approach leads to improved efficiency compared to testing the speed of the whole model during the search process. Extensive experiments demonstrate that, under similar parameters and FLOPs, our searched lightweight ViTs achieve higher accuracy and lower latency than state-of-the-art models. For instance, on ImageNet-1K, AutoViT_XXS (71.3% Top-1 accuracy, 10.2ms latency) outperforms MobileViTv3_XXS (71.0% Top-1 accuracy, 12.5ms latency) with 0.3% higher accuracy and 2.3ms lower latency.

In this paper, we propose a novel cross-attention-based generative adversarial network (GAN) for the challenging person image generation task. Cross-attention is a novel and intuitive multi-modal fusion method in which an attention/correlation matrix is calculated between two feature maps of different modalities. Specifically, we propose the novel XingGAN (or CrossingGAN), which consists of two generation branches that capture the person's appearance and shape, respectively. Moreover, we propose two novel cross-attention blocks to effectively transfer and update the person's shape and appearance embeddings for mutual improvement. This has not been considered by any other existing GAN-based image generation work. To further learn the long-range correlations between different person poses at different scales and sub-regions, we propose two novel multi-scale cross-attention blocks. To tackle the issue of independent correlation computations within the cross-attention mechanism leading to noisy and ambiguous attention weights, which hinder performance improvements, we propose a module called enhanced attention (EA). Lastly, we introduce a novel densely connected co-attention module to fuse appearance and shape features at different stages effectively. Extensive experiments on two public datasets demonstrate that the proposed method outperforms current GAN-based methods and performs on par with diffusion-based methods. However, our method is significantly faster than diffusion-based methods in both training and inference.

We propose a novel spatial-temporal graph Mamba (STG-Mamba) for the music-guided dance video synthesis task, i.e., to translate the input music to a dance video. STG-Mamba consists of two translation mappings: music-to-skeleton translation and skeleton-to-video translation. In the music-to-skeleton translation, we introduce a novel spatial-temporal graph Mamba (STGM) block to effectively construct skeleton sequences from the input music, capturing dependencies between joints in both the spatial and temporal dimensions. For the skeleton-to-video translation, we propose a novel self-supervised regularization network to translate the generated skeletons, along with a conditional image, into a dance video. Lastly, we collect a new skeleton-to-video translation dataset from the Internet, containing 54,944 video clips. Extensive experiments demonstrate that STG-Mamba achieves significantly better results than existing methods.

Autonomous driving platforms encounter diverse driving scenarios, each with varying hardware resources and precision requirements. Given the computational limitations of embedded devices, it is crucial to consider computing costs when deploying on target platforms like the DRIVE PX 2. Our objective is to customize the semantic segmentation network according to the computing power and specific scenarios of autonomous driving hardware. We implement dynamic adaptability through a three-tier control mechanism—width multiplier, classifier depth, and classifier kernel—allowing fine-grained control over model components based on hardware constraints and task requirements. This adaptability facilitates broad model scaling, targeted refinement of the final layers, and scenario-specific optimization of kernel sizes, leading to improved resource allocation and performance. Additionally, we leverage Bayesian Optimization with surrogate modeling to efficiently explore hyperparameter spaces under tight computational budgets. Our approach addresses scenario-specific and task-specific requirements through automatic parameter search, accommodating the unique computational complexity and accuracy needs of autonomous driving. It scales its multiply-accumulate operations (MACs) for task-specific learning adaptation (TSLA), resulting in alternative configurations tailored to diverse self-driving tasks. These TSLA customizations maximize computational capacity and model accuracy, optimizing hardware utilization.

Self-supervised monocular depth estimation shows great promise since only a single camera is required. However, most existing methods fail to model the geometric structure of objects, leading to poor performance in object boundary depth estimation. To overcome these shortcomings, a dual attention guidance network (DAG-Net), containing two complementary modules termed depth-guided attention module (DAM) and semantic-guided multi-modal attention module (SAM), is proposed in this paper. The DAM utilizes depth features to guide semantic features through multi-head attention. When semantic features are well learned, they guide depth features to learn useful geometric representations through backpropagation. Besides, the SAM is proposed to incorporate multi-modal data from depth estimation and semantic segmentation predictions at different scales. To eliminate the mutual interference between DAM and SAM, we also propose a two-stage training strategy to adjust the convergence direction during the training process. The effectiveness of our proposed DAG-Net is qualitatively and quantitatively verified by various experiments on KITTI, Cityscapes, and Make3D datasets, showing outstanding performance compared with the state-of-the-art methods.

In this article, we present an innovative solution tailored for the intricate challenges of the virtual try-on task—our novel Hierarchical Cross-Attention Network, HCANet. HCANet is meticulously crafted with two primary stages: geometric matching and try-on, each playing a crucial role in delivering realistic and visually convincing virtual try-on outcomes. A distinctive feature of HCANet is the incorporation of a novel Hierarchical Cross-Attention (HCA) block into both stages, enabling the effective capture of long-range correlations between individual and clothing modalities. The HCA block functions as a cornerstone, enhancing the depth and robustness of the network. By adopting a hierarchical approach, it facilitates a nuanced representation of the interaction between the person and clothing, capturing intricate details essential for an authentic virtual try-on experience. Our extensive set of experiments establishes the prowess of HCANet. The results showcase its cutting-edge performance across both objective quantitative metrics and subjective evaluations of visual realism. HCANet stands out as a state-of-the-art solution, demonstrating its capability to generate virtual try-on results that not only excel in accuracy but also satisfy subjective criteria of realism. This marks a significant step forward in advancing the field of virtual try-on technologies.

Low-light image enhancement is a crucial visual task, and many unsupervised methods overlook the degradation of visible information in low-light scenes, adversely affecting the fusion of complementary information and hindering the generation of satisfactory results. To address this, we introduce Wakeup-Darkness, a multimodal enhancement framework that innovatively enriches user interaction through voice and textual commands. This approach signifies a technical leap and represents a paradigm shift in user engagement. We introduce a Cross-Modal Feature Fusion (CMFF) that synergizes semantic and depth context with low-light enhancement operations. Moreover, we propose a Gated Residual Block (GRB) and a channel-aware Look-Up Table (LUT) to adjust the intensity distribution of each channel. Crucially, the proposed Wakeup-Darkness scheme demonstrates remarkable generalization in unsupervised scenarios. The source code can be accessed from https://github.com/zhangbaijin/Wakeup-Dakness .

The Sign Language Production (SLP) project aims to automatically translate spoken languages into sign sequences. Our approach focuses on the transformation of sign gloss sequences into their corresponding sign pose sequences (G2P). In this paper, we present a novel solution for this task by converting the continuous pose space generation problem into a discrete sequence generation problem. We introduce the Pose-VQVAE framework, which combines Variational Autoencoders (VAEs) with vector quantization to produce a discrete latent representation for continuous pose sequences. Additionally, we propose the G2P-DDM model, a discrete denoising diffusion architecture for length-varied discrete sequence data, to model the latent prior. To further enhance the quality of pose sequence generation in the discrete space, we present the CodeUnet model to leverage spatial-temporal information. Lastly, we develop a heuristic sequential clustering method to predict variable lengths of pose sequences for corresponding gloss sequences. Our results show that our model outperforms state-of-the-art G2P models on the public SLP evaluation benchmark. For more generated results, please visit our project page: https://slpdiffusier.github.io/g2p-ddm.

3D pose transfer that aims to transfer the desired pose to a target mesh is one of the most challenging 3D generation tasks. Previous attempts rely on well-defined parametric human models or skeletal joints as driving pose sources. However, to obtain those clean pose sources, cumbersome but necessary pre-processing pipelines are inevitable, hindering implementations of the real-time applications. This work is driven by the intuition that the robustness of the model can be enhanced by introducing adversarial samples into the training, leading to a more invulnerable model to the noisy inputs, which even can be further extended to directly handling the real-world data like raw point clouds/scans without intermediate processing. Furthermore, we propose a novel 3D pose Masked Autoencoder (3D-PoseMAE), a customized MAE that effectively learns 3D extrinsic presentations (i.e., pose). 3D-PoseMAE facilitates learning from the aspect of extrinsic attributes by simultaneously generating adversarial samples that perturb the model and learning the arbitrary raw noisy poses via a multi-scale masking strategy. Both qualitative and quantitative studies show that the transferred meshes given by our network result in much better quality. Besides, we demonstrate the strong generalizability of our method on various poses, different domains, and even raw scans. Experimental results also show meaningful insights that the intermediate adversarial samples generated in the training can success-fully attack the existing pose transfer models.

Extracting keypoint locations from input hand frames, known as 3D hand pose estimation, is a critical task in various human-computer interaction applications. Essentially, the 3D hand pose estimation can be regarded as a 3D point subset generative problem conditioned on input frames. Thanks to the recent significant progress on diffusion-based generative models, hand pose estimation can also benefit from the diffusion model to estimate keypoint locations with high quality. However, directly deploying the existing diffusion models to solve hand pose estimation is non-trivial, since they cannot achieve the complex permutation mapping and precise localization. Based on this motivation, this paper proposes HandDiff, a diffusion-based hand pose estimation model that iteratively denoises accurate hand pose conditioned on hand-shaped image-point clouds. In order to recover keypoint permutation and accurate location, we further introduce Joint-wise condition and local detail condition. Experimental results demonstrate that the proposed HandDiff significantly outperforms the existing approaches on four challenging hand pose benchmark datasets. Codes and pre-trained models are publicly available at https://github.com/cwc1260/HandDiff.

Diffusion models have recently gained prominence as a novel category of generative models. Despite their success, these models face a notable drawback in terms of slow sampling speeds, requiring a high number of function evaluations (NFE) in the order of hundreds or thousands. In response, both learning-free and learning-based sampling strategies have been explored to expedite the sampling process. Learning-free sampling employs various ordinary differential equation (ODE) solvers based on the formulation of diffusion ODEs. However, it encounters challenges in faithfully tracking the true sampling trajectory, particularly for small NFE. Conversely, learning-based sampling methods, such as knowledge distillation, demand extensive additional training, limiting their practical applicability. To overcome these limitations, we introduce Distilled-ODE solvers (D-ODE solvers), a straightforward distillation approach grounded in ODE solver formulations. Our method seamlessly integrates the strengths of both learning-free and learning-based sampling. D-ODE solvers are constructed by introducing a single parameter adjustment to existing ODE solvers. Furthermore, we optimize D-ODE solvers with smaller steps using knowledge distillation from ODE solvers with larger steps across a batch of samples. Comprehensive experiments demonstrate the superior performance of D- ODE solvers compared to existing ODE solvers, including DDIM, PNDM, DPM-Solver, DEIS, and EDM, particularly in scenarios with fewer NFE. Notably, our method incurs negligible computational overhead compared to previous distillation techniques, facilitating straightforward and rapid integration with existing samplers. Qualitative analysis reveals that D-ODE solvers not only enhance image quality but also faithfully follow the target ODE trajectory.

While Transformers have rapidly gained popularity in various computer vision applications, post-hoc explanations of their internal mechanisms remain largely unexplored. Vision Transformers extract visual information by representing image regions as transformed tokens and integrating them via attention weights. However, existing post-hoc explanation methods merely consider these attention weights, neglecting crucial information from the transformed tokens, which fails to accurately illustrate the rationales behind the models' predictions. To incorporate the influence of token transformation into interpretation, we propose TokenTM, a novel post-hoc explanation method that utilizes our introduced measurement of token transformation effects. Specifically, we quantify token transformation effects by measuring changes in token lengths and correlations in their directions pre- and post-transformation. Moreover, we develop initialization and aggregation rules to integrate both attention weights and token transformation effects across all layers, capturing holistic token contributions throughout the model. Experimental results on segmentation and perturbation tests demonstrate the superiority of our proposed TokenTM compared to state-of-the-art Vision Transformer explanation methods.

To interpret Vision Transformers, post-hoc explanations assign salience scores to input pixels, providing human-understandable heatmaps. However, whether these inter-pretations reflect true rationales behind the model's output is still underexplored. To address this gap, we study the faithfulness criterion of explanations: the assigned salience scores should represent the influence of the corresponding input pixels on the model's predictions. To evaluate faithful-ness, we introduce Salience-guided Faithfulness Coefficient (SaCo), a novel evaluation metric leveraging essential in-formation of salience distribution. Specifically, we con-duct pair-wise comparisons among distinct pixel groups and then aggregate the differences in their salience scores, resulting in a coefficient that indicates the explanation's degree of faithfulness. Our explorations reveal that cur-rent metrics struggle to differentiate between advanced ex-planation methods and Random Attribution, thereby failing to capture the faithfulness property. In contrast, our pro-posed SaCo offers a reliable faithfulness measurement, es-tablishing a robust metric for interpretations. Furthermore, our SaCo demonstrates that the use of gradient and multi-layer aggregation can markedly enhance the faithfulness of attention-based explanation, shedding light on potential paths for advancing Vision Transformer explainability.

Recent advancements in subject-driven image generation have led to zero-shot generation, yet precise selection and focus on crucial subject representations remain challenging. Addressing this, we introduce the SSR-Encoder, a novel architecture designed for selectively capturing any subject from single or multiple reference images. It responds to various query modalities including text and masks, without necessitating test-time fine-tuning. The SSR-Encoder combines a Token-to-Patch Aligner that aligns query inputs with image patches and a Detail-Preserving Subject Encoder for extracting and preserving fine features of the subjects, thereby generating subject embeddings. These embeddings, used in conjunction with original text embeddings, condition the generation process. Characterized by its model generalizability and efficiency, the SSR-Encoder adapts to a range of custom models and control modules. Enhanced by the Embedding Consistency Regularization Loss for improved training, our extensive experiments demonstrate its effectiveness in versatile and high-quality image generation, indicating its broad applicability. Project page: ssr-encoder.github.io

Route planning for navigation under partial observability plays a crucial role in modern robotics and autonomous driving. Existing route planning approaches can be categorized into two main classes: traditional autoregressive and diffusion-based methods. The former often fails due to its myopic nature, while the latter either assumes full observability or struggles to adapt to unfamiliar scenarios, due to strong couplings with behavior cloning from experts. To address these deficiencies, we propose a versatile diffusion-based approach for both 2D and 3D route planning under partial observability. Specifically, our value-guided diffusion policy first generates plans to predict actions across various timesteps, providing ample foresight to the planning. It then employs a differentiable planner with state estimations to derive a value function, directing the agent's exploration and goal-seeking behaviors without seeking experts while explicitly addressing partial observability. During inference, our policy is further enhanced by a best-plan-selection strategy, substantially boosting the planning success rate. Moreover, we propose projecting point clouds, derived from RGB-D inputs, onto 2D grid-based bird-eye- view maps via semantic segmentation, generalizing to 3D environments. This simple yet effective adaption enables zero-shot transfer from 2D-trained policy to 3D, cutting across the laborious training for 3D policy, and thus certifying our versatility. Experimental results demonstrate our superior performance, particularly in navigating situations beyond expert demonstrations, surpassing state-of-the-art autoregressive and diffusion-based baselines for both 2D and 3D scenarios.

Knowledge distillation (KD), which distills knowledge from complex networks (teacher) to lightweight (student) networks, has been actively studied recently. Despite previous studies have proposed several advanced KD losses or intricate training strategies, the core concept of KD proves ineffective if the student model is too weak to mimic the teacher's performance. In this study, we aim to narrow the performance discrepancy between Transformer-based teacher and student models by incorporating the inductive biases of several heterogeneous student models. To this end, we put forward a novel cross-architecture knowledge distillation approach called Adaptive Cross-architecture Mutual Knowledge Distillation (ACMKD), which tries to mitigate the performance gap issue using a multi-students mutual learning strategy. Specifically, we utilize three mainstream models associated with various inductive biases (CNN, INN, and Transformer) as the student models. In addition, we propose an effective attention similarity mechanism to facilitate the student models in mimicking specific portions of the teacher model. Drawing inspiration from the Cannikin Law, we devise a unique second-stage KD process that dynamically enables the weakest student model to learn from other stronger student models again. We validate our proposed methods on ImageNet and CIFAR100 datasets, and the results confirm that our ACMKD method significantly narrows the performance gap compared to other KD methods.

Vision transformers (ViTs) have demonstrated their superior accuracy for computer vision tasks compared to convolutional neural networks (CNNs). However, ViT models are often computation-intensive for efficient deployment on resource-limited edge devices. This work proposes Quasar-ViT, a hardware-oriented quantization-aware architecture search framework for ViTs, to design efficient ViT models for hardware implementation while preserving the accuracy. First, Quasar-ViT trains a supernet using our row-wise flexible mixed-precision quantization scheme, mixed-precision weight entanglement, and supernet layer scaling techniques. Then, it applies an efficient hardware-oriented search algorithm, integrated with hardware latency and resource modeling, to determine a series of optimal subnets from supernet under different inference latency targets. Finally, we propose a series of model-adaptive designs on the FPGA platform to support the architecture search and mitigate the gap between the theoretical computation reduction and the practical inference speedup. Our searched models achieve 101.5, 159.6, and 251.6 frames-per-second (FPS) inference speed on the AMD/Xilinx ZCU102 FPGA with 80.4%, 78.6%, and 74.9% top-1 accuracy, respectively, for the ImageNet dataset, consistently outperforming prior works.

Whole-body pose estimation aims to regress human pose models that include the body, hand, and facial details from RGB images. While the task of whole-body mesh recovery has been extensively studied in recent literature, the focus has predominantly been on human mesh recovery for a single person, despite the frequent occurrence of multiple people in practical scenarios. Similar to body-only cases, such single-person whole-body pose estimation methods often fail in the multiple-people problem for two reasons: (i) Given the ambiguous bounding box, which could contain more than one instance, it is difficult for single-person-oriented methods to regress the body mesh model of the target person. (ii) Single-person pose estimation approaches neglect the person-person occlusions and the depth order among instances, thus generating interpenetrated models. In this paper, we propose the Multi-person Expressive POse (MEPO) model, which exploits expressive 3D human model reconstruction for multiple people. To our best knowledge, our model is the first multi-person whole-body mesh reconstruction model, which is intensified by heatmap, depthmap, and depth order loss. We propose the Heatmap Enhancement Net (HENet) to leverage the heatmap information to assist the model in concentrating on the target person in crowded multi-person cases, while the depthmap delivers depth information of the image. Furthermore, we impose a depth order loss to recover human mesh precisely for overlapped people. In our experiments, we evaluate our model on multiple challenging datasets, including AGORA, which consists of complex occlusions similar to real-world scenarios. Our method has a significant performance improvement compared with the state-of-the-art pose estimation methods.

Story visualization aims to generate realistic and coherent images based on multi-sentence stories. However, current methods face challenges in achieving high-quality image generation while maintaining lightweight models and a fast generation speed. The main issue lies in the two existing frameworks. The independent framework prioritizes speed but sacrifices image quality with the non-collaborative image generation process and basic GAN-based learning. The autoregressive framework modifies the large pretrained text-to-image model in an auto-regressive manner with additional history modules, leading to large model size, resource-intensive requirements, and slow generation speed. To address these issues, we propose a lightweight and effective framework, namely CoIn. Specifically, we introduce a Context-aware Story Generator to predict shared context semantics for each image generator. Additionally, we propose an Intra-Story Interchange module that allows each image generator to exchange visual information with other image generators. Furthermore, we incorporate DINOv2 into the story and image discriminators to assess the story image quality more accurately. Extensive experiments show that our CoIn keeps the model size and generation speed of the independent framework, while achieving promising story image quality.

Diffusion models have shown impressive potential on talking head generation. While plausible appearance and talking effect are achieved, these methods still suffer from temporal, 3D or expression inconsistency due to the error accumulation and inherent limitation of single-image generation ability. In this paper, we propose ConsistentAvatar, a novel framework for fully consistent and high-fidelity talking avatar generation. Instead of directly employing multi-modal conditions to the diffusion process, our method learns to first model the temporal representation for stability between adjacent frames. Specifically, we propose a Temporally-Sensitive Detail (TSD) map containing high-frequency feature and contours that vary significantly along the time axis. Using a temporal consistent diffusion module, we learn to align TSD of the initial result to that of the video frame ground truth. The final avatar is generated by a fully consistent diffusion module, conditioned on the aligned TSD, rough head normal, and emotion prompt embedding. We find that the aligned TSD, which represents the temporal patterns, constrains the diffusion process to generate temporally stable talking head. Further, its reliable guidance complements the inaccuracy of other conditions, suppressing the accumulated error while improving the consistency on various aspects. Extensive experiments demonstrate that ConsistentAvatar outperforms the state-of-the-art methods on the generated appearance, 3D, expression and temporal consistency.

Physical adversarial attacks can deceive deep neural networks (DNNs), leading to erroneous predictions in real-world scenarios. To uncover potential security risks, attacking the safety-critical task of person detection has garnered significant attention. However, we observe that existing attack methods overlook the pivotal role of the camera, involving capturing real-world scenes and converting them into digital images, in the physical adversarial attack workflow. This oversight leads to instability and challenges in reproducing these attacks. In this work, we revisit patch-based attacks against person detectors and introduce a camera-agnostic physical adversarial attack to mitigate this limitation. Specifically, we construct a differentiable camera Image Signal Processing (ISP) proxy network to compensate for the physical-to-digital transition gap. Furthermore, the camera ISP proxy network serves as a defense module, forming an adversarial optimization framework with the attack module. The attack module optimizes adversarial patches to maximize effectiveness, while the defense module optimizes the conditional parameters of the camera ISP proxy network to minimize attack effectiveness. These modules engage in an adversarial game, enhancing cross-camera stability. Experimental results demonstrate that our proposed Camera-Agnostic Patch (CAP) attack effectively conceals persons from detectors across various imaging hardware, including two distinct cameras and four smartphones.

The generation of natural human motion interactions is a hot topic in computer vision and computer animation. It is a challenging task due to the diversity of possible human motion interactions. Diffusion models, which have already shown remarkable generative capabilities in other domains, are a good candidate for this task. In this paper, we introduce a novel bipartite graph diffusion method (BiGraphDiff) to generate human motion interactions between two persons. Specifically, bipartite node sets are constructed to model the inherent geometric constraints between skeleton nodes during interactions. The interaction graph diffusion model is transformer-based, combining some state-of-theart motion methods. We show that the proposed achieves new state-of-the-art results on leading benchmarks for the human interaction generation task. Code, pre-trained models and additional results are available at https://github.com/CRISTAL-3DSAM/BiGraphDiff.

We introduce a novel weakly-supervised video actor-action segmentation (VAAS) framework, where only video-level tags are available. Previous VAAS methods follow a synthesize-and-refine scheme, i.e., they first synthesize the pseudo-segmentation and recursively refine the segmentation. However, this process requires significant time costs and heavily relies on the quality of the initial segmentation. Unlike existing works, our method hierarchically mines contrastive relations to supplement each other for learning a visually-plausible segmentation model. Specifically, three contrastive relations are abstracted from the pixel-level and frame-level, i.e., low-level edge-aware, class-activation map aware, and semantic tag-aware relations. Then, the discovered contrastive relations are unified into a universal objective for training the segmentation model, regardless of their heterogeneity. Moreover, we incorporate motion cues and unlabeled samples to increase the discriminative power and robustness of the segmentation model. Extensive experiments indicate that our proposed method produces reasonable segmentation.

Segment Anything Model (SAM), an advanced universal image segmentation model trained on an expansive visual dataset, has set a new benchmark in image segmentation and computer vision. However, it faced challenges when it came to distinguishing between shadows and their backgrounds. To address this, we proposed ShadClips, which consists of SAM-optimizer and SONet. It has dramatically enhanced SAM’s ability to segment shadow images, differentiating between the background and both soft and hard shadows adeptly. Due to its dependence on pixel point inputs, the SAM-Optimizer interface could do better. This method presents challenges, especially when dealing with long, extended shadows. To make the user experience more intuitive and effective, we incorporated the capabilities of CLIPs. Therefore, simple text descriptions like “A photo of a shadow” can be used to guide the SAM-Optimizer, allowing it to select the most relevant shadow mask from SAM’s comprehensive category list. Meanwhile, we introduce SONet to shadow removal. A large number of experiments on ISTD/SRD prove that the proposed method is effective and satisfactory. The source code of the ShadClips can be accessed from https://github.com/zhangbaijin/SAM-helps-Shadow .

We present a novel graph Transformer generative adversarial network (GTGAN) to learn effective graph node relations in an end-to-end fashion for challenging graph-constrained architectural layout generation tasks. The proposed graph-Transformer-based generator includes a novel graph Transformer encoder that combines graph convolutions and self-attentions in a Transformer to model both local and global interactions across connected and non-connected graph nodes. Specifically, the proposed connected node attention (CNA) and non-connected node attention (NNA) aim to capture the global relations across connected nodes and non-connected nodes in the input graph, respectively. The proposed graph modeling block (GMB) aims to exploit local vertex interactions based on a house layout topology. Moreover, we propose a new node classification-based discriminator to preserve the high-level semantic and discriminative node features for different house components. To maintain the relative spatial relationships between ground truth and predicted graphs, we also propose a novel graph-based cycle-consistency loss. Finally, we propose a novel self-guided pre-training method for graph representation learning. This approach involves simultaneous masking of nodes and edges at an elevated mask ratio (i.e., 40%) and their subsequent reconstruction using an asymmetric graph-centric autoencoder architecture. This method markedly improves the model's learning proficiency and expediency. Experiments on three challenging graph-constrained architectural layout generation tasks (i.e., house layout generation, house roof generation, and building layout generation) with three public datasets demonstrate the effectiveness of the proposed method in terms of objective quantitative scores and subjective visual realism. New state-of-the-art results are established by large margins on these three tasks.

Despite the impressive achievements of Deep Neural Networks (DNNs) in computer vision, their vulnerability to adversarial attacks remains a critical concern. Extensive research has demonstrated that incorporating sophisticated perturbations into input images can lead to a catastrophic degradation in DNNs' performance. This perplexing phenomenon not only exists in the digital space but also in the physical world. Consequently, it becomes imperative to evaluate the security of DNNs-based systems to ensure their safe deployment in real-world scenarios, particularly in security-sensitive applications. To facilitate a profound understanding of this topic, this paper presents a comprehensive overview of physical adversarial attacks. First, we distill four general steps for launching physical adversarial attacks. Building upon this foundation, we uncover the pervasive role of artifacts carrying adversarial perturbations in the physical world. These artifacts influence each step. To denote them, we introduce a new term: adversarial medium. Then, we take the first step to systematically evaluate the performance of physical adversarial attacks, taking the adversarial medium as a first attempt. Our proposed evaluation metric, hiPAA, comprises six perspectives: Effectiveness, Stealthiness, Robustness, Practicability, Aesthetics, and Economics. We also provide comparative results across task categories, together with insightful observations and suggestions for future research directions.

High Dynamic Range (HDR) images can be recovered from several Low Dynamic Range (LDR) images by existing Deep Neural Networks (DNNs) techniques. Despite the remarkable progress, DNN-based methods still generate ghosting artifacts when LDR images have saturation and large motion, which hinders potential applications in real-world scenarios. To address this challenge, we formulate the HDR deghosting problem as an image generation that leverages LDR features as the diffusion model’s condition, consisting of the feature condition generator and the noise predictor. Feature condition generator employs attention and Domain Feature Alignment (DFA) layer to transform the intermediate features to avoid ghosting artifacts. With the learned features as conditions, the noise predictor leverages a stochastic iterative denoising process for diffusion models to generate an HDR image by steering the sampling process. Furthermore, to mitigate semantic confusion caused by the saturation problem of LDR images, we design a sliding window noise estimator to sample smooth noise in a patch-based manner. In addition, an image space loss is proposed to avoid the color distortion of the estimated HDR results. We empirically evaluate our model on benchmark datasets for HDR imaging. The results demonstrate that our approach achieves state-of-the-art performances and well generalization to real-world images.

Vision Foundation Models (VFMs) such as the Segment Anything Model (SAM) allow zero-shot or interactive segmentation of visual contents, thus they are quickly applied in a variety of visual scenes. However, their direct use in many Remote Sensing (RS) applications is often unsatisfactory due to the special imaging properties of RS images. In this work, we aim to utilize the strong visual recognition capabilities of VFMs to improve change detection (CD) in very high-resolution (VHR) remote sensing images (RSIs). We employ the visual encoder of FastSAM, a variant of the SAM, to extract visual representations in RS scenes. To adapt FastSAM to focus on some specific ground objects in RS scenes, we propose a convolutional adaptor to aggregate the task-oriented change information. Moreover, to utilize the semantic representations that are inherent to SAM features, we introduce a task-agnostic semantic learning branch to model the semantic latent in bi-temporal RSIs. The resulting method, SAM-CD, obtains superior accuracy compared to the SOTA fully-supervised CD methods and exhibits a sample-efficient learning ability that is comparable to semi-supervised CD methods. To the best of our knowledge, this is the first work that adapts VFMs to CD in VHR RS images.

The 2D image-based virtual try-on has aroused increased interest from the multimedia and computer vision fields due to its enormous commercial value. Nevertheless, most existing image-based virtual try-on approaches directly combine the person-identity representation and the in-shop clothing items without taking their mutual correlations into consideration. Moreover, these methods are commonly established on pure convolutional neural networks (CNNs) architectures which are not simple to capture the long-range correlations among the input pixels. As a result, it generally results in inconsistent results. To alleviate these issues, in this article, we propose a novel two-stage cloth interactive transformer (CIT) method for the virtual try-on task. During the first stage, we design a CIT matching block, aiming at precisely capturing the long-range correlations between the cloth-agnostic person information and the in-shop cloth information. Consequently, it makes the warped in-shop clothing items look more natural in appearance. In the second stage, we put forth a CIT reasoning block for establishing global mutual interactive dependencies among person representation, the warped clothing item, and the corresponding warped cloth mask. The empirical results, based on mutual dependencies, demonstrate that the final try-on results are more realistic. Substantial empirical results on a public fashion dataset illustrate that the suggested CIT attains competitive virtual try-on performance.

Vision transformers (ViTs) have recently obtained success in many applications, but their intensive computation and heavy memory usage at both training and inference time limit their generalization. Previous compression algorithms usually start from the pre-trained dense models and only focus on efficient inference, while time-consuming training is still unavoidable. In contrast, this paper points out that the million-scale training data is redundant, which is the fundamental reason for the tedious training. To address the issue, this paper aims to introduce sparsity into data and proposes an end-to-end efficient training framework from three sparse perspectives, dubbed Tri-Level E-ViT. Specifically, we leverage a hierarchical data redundancy reduction scheme, by exploring the sparsity under three levels: number of training examples in the dataset, number of patches (tokens) in each example, and number of connections between tokens that lie in attention weights. With extensive experiments, we demonstrate that our proposed technique can noticeably accelerate training for various ViT architectures while maintaining accuracy. Remarkably, under certain ratios, we are able to improve the ViT accuracy rather than compromising it. For example, we can achieve 15.2% speedup with 72.6% (+0.4) Top-1 accuracy on Deit-T, and 15.7% speedup with 79.9% (+0.1) Top-1 accuracy on Deit-S. This proves the existence of data redundancy in ViT. Our code is released at https://github.com/ZLKong/Tri-Level-ViT

The research in real-time segmentation mainly focuses on desktop GPUs. However, autonomous driving and many other applications rely on real-time segmentation on the edge, and current arts are far from the goal. In addition, recent advances in vision transformers also inspire us to re-design the network architecture for dense prediction task. In this work, we propose to combine the self attention block with lightweight convolutions to form new building blocks, and employ latency constraints to search an efficient sub-network. We train an MLP latency model based on generated architecture configurations and their latency measured on mobile devices, so that we can predict the latency of subnets during search phase. To the best of our knowledge, we are the first to achieve over 74% mIoU on Cityscapes with semi-real-time inference (over 15 FPS) on mobile GPU from an off-the-shelf phone.

Text-guided image editing models have shown remarkable results. However, there remain two problems. First, they employ fixed manipulation modules for various editing requirements (e.g., color changing, texture changing, content adding and removing), which results in over-editing or insufficient editing. Second, they do not clearly distinguish between text-required and text-irrelevant parts, which leads to inaccurate editing. To solve these limitations, we propose: (i) a Dynamic Editing Block (DEBlock) that composes different editing modules dynamically for various editing requirements. (ii) a Composition Predictor (Comp-Pred), which predicts the composition weights for DEBlock according to the inference on target texts and source images. (iii) a Dynamic text-adaptive Convolution Block (DCBlock) that queries source image features to distinguish text-required parts and text-irrelevant parts. Extensive experiments demonstrate that our DE-Net achieves excellent performance and manipulates source images more correctly and accurately.

Adversarial attacks on thermal infrared imaging expose the risk of related applications. Estimating the security of these systems is essential for safely deploying them in the real world. In many cases, realizing the attacks in the physical space requires elaborate special perturbations. These solutions are often impractical and attention-grabbing. To address the need for a physically practical and stealthy adversarial attack, we introduce HotCold Block, a novel physical attack for infrared detectors that hide persons utilizing the wearable Warming Paste and Cooling Paste. By attaching these readily available temperature-controlled materials to the body, HotCold Block evades human eyes efficiently. Moreover, unlike existing methods that build adversarial patches with complex texture and structure features, HotCold Block utilizes an SSP-oriented adversarial optimization algorithm that enables attacks with pure color blocks and explores the influence of size, shape, and position on attack performance. Extensive experimental results in both digital and physical environments demonstrate the performance of our proposed HotCold Block. Code is available: https://github.com/weihui1308/HOTCOLDBlock.

We present a novel graph Transformer generative adversarial network (GTGAN) to learn effective graph node relations in an end-to-end fashion for the challenging graph-constrained house generation task. The proposed graph-Transformer-based generator includes a novel graph Transformer encoder that combines graph convolutions and self-attentions in a Transformer to model both local and global interactions across connected and non-connected graph nodes. Specifically, the proposed connected node attention (CNA) and non-connected node attention (NNA) aim to capture the global relations across connected nodes and non-connected nodes in the input graph, respectively. The proposed graph modeling block (GMB) aims to exploit local vertex interactions based on a house layout topology. More-over, we propose a new node classification-based discriminator to preserve the high-level semantic and discriminative node features for different house components. Finally, we propose a novel graph-based cycle-consistency loss that aims at maintaining the relative spatial relationships between ground truth and predicted graphs. Experiments on two challenging graph-constrained house generation tasks (i.e., house layout and roof generation) with two public datasets demonstrate the effectiveness of GTGAN in terms of objective quantitative scores and subjective visual realism. New state-of-the-art results are established by large margins on both tasks.

Deep point cloud registration methods face challenges to partial overlaps and rely on labeled data. To address these issues, we propose UDPReg, an unsupervised deep probabilistic registration framework for point clouds with partial overlaps. Specifically, we first adopt a network to learn posterior probability distributions of Gaussian mixture models (GMMs) from point clouds. To handle partial point cloud registration, we apply the Sinkhorn algorithm to predict the distribution-level correspondences under the constraint of the mixing weights of GMMs. To enable unsupervised learning, we design three distribution consistency-based losses: self-consistency, cross-consistency, and local contrastive. The self-consistency loss is formulated by encouraging GMMs in Euclidean and feature spaces to share identical posterior distributions. The cross-consistency loss derives from the fact that the points of two partially overlapping point clouds belonging to the same clusters share the cluster centroids. The cross-consistency loss allows the network to flexibly learn a transformation-invariant posterior distribution of two aligned point clouds. The local contrastive loss facilitates the network to extract discriminative local features. Our UDPReg achieves competitive performance on the 3DMatch/3DLoMatch and ModelNet/ModelLoNet benchmarks.

The rapid advances in Vision Transformer (ViT) refresh the state-of-the-art performances in various vision tasks, overshadowing the conventional CNN-based models. This ignites a few recent striking-back research in the CNN world showing that pure CNN models can achieve as good performance as ViT models when carefully tuned. While encouraging, designing such high-performance CNN models is challenging, requiring non-trivial prior knowledge of network design. To this end, a novel framework termed Mathematical Architecture Design for Deep CNN (Deep-MAD11Source codes are available at https://github.com/alibaba/lightweight-neural-architecture-search) is proposed to design high-performance CNN models in a principled way. In DeepMAD, a CNN network is modeled as an information processing system whose expressiveness and effectiveness can be analytically formulated by their structural parameters. Then a constrained mathematical programming (MP) problem is proposed to optimize these structural parameters. The MP problem can be easily solved by off-the-shelf MP solvers on CPUs with a small memory footprint. In addition, DeepMAD is a pure mathematical framework: no GPU or training data is required during network design. The superiority of DeepMAD is validated on multiple large-scale computer vision benchmark datasets. Notably on ImageNet-1k, only using conventional convolutional layers, DeepMAD achieves 0.7% and 1.5% higher top-1 accuracy than ConvNeXt and Swin on Tiny level, and 0.8% and 0.9% higher on Small level.

Transformer-based models achieve favorable performance in artistic style transfer recently thanks to its global receptive field and powerful multi-head/layer attention operations. Nevertheless, the over-paramerized multi-layer structure increases parameters significantly and thus presents a heavy burden for training. Moreover, for the task of style transfer, vanilla Transformer that fuses content and style features by residual connections is prone to content-wise distortion. In this paper, we devise a novel Transformer model termed as Master specifically for style transfer. On the one hand, in the proposed model, different Transformer layers share a common group of parameters, which (1) reduces the total number of parameters, (2) leads to more robust training convergence, and (3) is readily to control the degree of stylization via tuning the number of stacked layers freely during inference. On the other hand, different from the vanilla version, we adopt a learnable scaling operation on content features before content-style feature interaction, which better preserves the original similarity between a pair of content features while ensuring the stylization quality. We also propose a novel meta learning scheme for the proposed model so that it can not only work in the typical setting of arbitrary style transfer, but also adaptable to the few-shot setting, by only fine-tuning the Transformer encoder layer in the few-shot stage for one specific style. Text-guided few-shot style transfer is firstly achieved with the proposed framework. Extensive experiments demonstrate the superiority of Master under both zero-shot and few-shot style transfer settings.

Synthesizing high-fidelity complex images from text is challenging. Based on large pretraining, the autoregressive and diffusion models can synthesize photo-realistic images. Although these large models have shown notable progress, there remain three flaws. 1) These models require tremendous training data and parameters to achieve good performance. 2) The multi-step generation design slows the image synthesis process heavily. 3) The synthesized visual features are challenging to control and require delicately designed prompts. To enable high-quality, efficient, fast, and controllable text-to-image synthesis, we propose Generative Adversarial CLIPs, namely GALIP. GALIP leverages the powerful pretrained CLIP model both in the discriminator and generator. Specifically, we propose a CLIP-based discriminator. The complex scene understanding ability of CLIP enables the discriminator to accurately assess the image quality. Furthermore, we propose a CLIP-empowered generator that induces the visual concepts from CLIP through bridge features and prompts. The CLIP-integrated generator and discriminator boost training efficiency, and as a result, our model only requires about 3% training data and 6% learnable parameters, achieving comparable results to large pretrained autoregressive and diffusion models. Moreover, our model achieves ~120×faster synthesis speed and inherits the smooth latent space from GAN. The extensive experimental results demonstrate the excellent performance of our GALIP. Code is available at https://github.com/tobran/GALIP.

Generating a high-quality High Dynamic Range (HDR) image from dynamic scenes has recently been extensively studied by exploiting Deep Neural Networks (DNNs). Most DNNs-based methods require a large amount of training data with ground truth, requiring tedious and time-consuming work. Few-shot HDR imaging aims to generate satisfactory images with limited data. However, it is difficult for modern DNNs to avoid overfitting when trained on only a few images. In this work, we propose a novel semi-supervised approach to realize few-shot HDR imaging via two stages of training, called SSHDR. Unlikely previous methods, directly recovering content and removing ghosts simultaneously, which is hard to achieve optimum, we first generate content of saturated regions with a self-supervised mechanism and then address ghosts via an iterative semi-supervised learning framework. Concretely, considering that saturated regions can be regarded as masking Low Dynamic Range (LDR) input regions, we design a Saturated Mask AutoEncoder (SMAE) to learn a robust feature representation and reconstruct a non-saturated HDR image. We also propose an adaptive pseudo-label selection strategy to pick high-quality HDR pseudo-labels in the second stage to avoid the effect of mislabeled samples. Experiments demonstrate that SSHDR outperforms state-of-the-art methods quantitatively and qualitatively within and across different datasets, achieving appealing HDR visualization with few labeled samples.

With the ever-increasing popularity of edge devices, it is necessary to implement real-time segmentation on the edge for autonomous driving and many other applications. Vision Transformers (ViTs) have shown considerably stronger results for many vision tasks. However, ViTs with the fullattention mechanism usually consume a large number of computational resources, leading to difficulties for real- time inference on edge devices. In this paper, we aim to derive ViTs with fewer computations and fast inference speed to facilitate the dense prediction of semantic segmentation on edge devices. To achieve this, we propose a pruning parameterization method to formulate the pruning problem of semantic segmentation. Then we adopt a bi-level optimization method to solve this problem with the help of implicit gradients. Our experimental results demonstrate that we can achieve 38.9 mIoU on ADE20K val with a speed of 56.5 FPS on Samsung S21, which is the highest mIoU under the same computation constraint with real-time inference.

For semantic-guided cross-view image translation, it is crucial to learn where to sample pixels from the source view image and where to reallocate them guided by the target view semantic map, especially when there is little overlap or drastic view difference between the source and target images. Hence, one not only needs to encode the long- range dependencies among pixels in both the source view image and target view semantic map but also needs to translate these learned dependencies. To this end, we propose a novel generative adversarial network, PI-Trans, which mainly consists of a novel Parallel-ConvMLP module and an Implicit Transformation module at multiple semantic levels. Extensive experimental results show that PI-Trans achieves the best qualitative and quantitative performance by a large margin compared to the state-of-the-art methods on two challenging datasets. The source code is available at https://github.com/Amazingren/PI-Trans.

Brain vessel image segmentation can be used as a promising biomarker for better prevention and treatment of different diseases. One successful approach is to consider the segmentation as an image-to-image translation task and perform a conditional Generative Adversarial Network (cGAN) to learn a transformation between two distributions. In this paper, we present a novel multi-view approach, MLP-GAN, which splits a 3D volumetric brain vessel image into three different dimensional 2D images (i.e., sagittal, coronal, axial) and then feed them into three different 2D cGANs. The proposed MLP-GAN not only alleviates the memory issue which exists in the original 3D neural networks but also retains 3D spatial information. Specifically, we utilize U-Net as the backbone for our generator and redesign the pattern of skip connection integrated with the MLP-Mixer [1] which has attracted lots of attention recently. Our model obtains the ability to capture cross-patch information to learn global information with the MLP-Mixer. Extensive experiments are performed on the public brain vessel dataset [2] that show our MLP-GAN outperforms other state-of-the-art methods.

This paper introduces an effective and tiny model for real-time Camouflaged Object Detection (COD) named Tiny-COD. It achieves high performance with very low costs (Parameters < 5M, FLOPs < 1.5G), which can be applied on mobile devices. Specifically, we introduce a simple but effective Adjacent Scale Features Fusion module (ASFF), which can significantly enhance the representation ability of features from a lightweight backbone. Besides, as the edge areas of the camouflaged object often blend into the background, we carefully design an Edge Area Focus module (EAF) to solve this problem. Experimental results on COD datasets prove that the proposed method achieves state-of-the-art performance compared with other methods.

Jointly processing information from multiple sensors is crucial to achieving accurate and robust perception for reliable autonomous driving systems. However, current 3D perception research follows a modality-specific paradigm, leading to additional computation overheads and inefficient collaboration between different sensor data. In this paper, we present an efficient multi-modal backbone for outdoor 3D perception named UniTR, which processes a variety of modalities with unified modeling and shared parameters. Unlike previous works, UniTR introduces a modality-agnostic transformer encoder to handle these view-discrepant sensor data for parallel modal-wise representation learning and automatic cross-modal interaction without additional fusion steps. More importantly, to make full use of these complementary sensor types, we present a novel multi-modal integration strategy by both considering semantic-abundant 2D perspective and geometry-aware 3D sparse neighborhood relations. UniTR is also a fundamentally task-agnostic backbone that naturally supports different 3D perception tasks. It sets a new state-of-the-art performance on the nuScenes benchmark, achieving +1.1 NDS higher for 3D object detection and +12.0 higher mIoU for BEV map segmentation with lower inference latency. Code will be available at https://github.com/Haiyang-W/UniTR.

Owing to the large distribution gap between the heterogeneous data in Visible-Infrared Person Re-identification (VI Re-ID), we point out that existing paradigms often suffer from the inter-modal semantic misalignment issue and thus fail to align and compare local details properly. In this paper, we present Concordant Attention Learning (CAL), a novel framework that learns semantic-aligned representations for VI Re-ID. Specifically, we design the Target-aware Concordant Alignment paradigm, which allows target-aware attention adaptation when aligning heterogeneous samples (i.e., adaptive attention adjustment according to the target image being aligned). This is achieved by exploiting the discriminative clues from the modality counterpart and designing effective modality-agnostic correspondence searching strategies. To ensure semantic concordance during the cross-modal retrieval stage, we further propose MatchDistill, which matches the attention patterns across modalities and learns their underlying semantic correlations by bipartite-graph-based similarity modeling and cross-modal knowledge exchange. Extensive experiments on VI Re-ID benchmark datasets demonstrate the effectiveness and superiority of the proposed CAL.

Vision Transformers (ViTs) have continuously achieved new milestones in object detection. However, the considerable computation and memory burden compromise their efficiency and generalization of deployment on resource-constraint devices. Besides, efficient transformer-based detectors designed by existing works can hardly achieve a realistic speedup, especially on multi-core processors (e.g., GPUs). The main issue is that the current literature solely concentrates on building algorithms with minimal computation, oblivious that the practical latency can also be affected by the memory access cost and the degree of parallelism. Therefore, we propose SpeedDETR, a novel speed-aware transformer for end-to-end object detectors, achieving high-speed inference on multiple devices. Specifically, we design a latency prediction model which can directly and accurately estimate the network latency by analyzing network properties, hardware memory access pattern, and degree of parallelism. Following the effective local-to-global visual modeling process and the guidance of the latency prediction model, we build our hardware-oriented architecture design and develop a new family of SpeedDETR. Experiments on the MS COCO dataset show SpeedDETR outperforms current DETR-based methods on Tesla V100. Even acceptable speed inference can be achieved on edge GPUs.

The long-tail effect is a common issue that limits the performance of deep learning models on real-world datasets. Character image datasets are also affected by such unbalanced data distribution due to differences in character usage frequency. Thus, current character recognition methods are limited when applied in the real world, especially for the categories in the tail that lack training samples, e.g., uncommon characters. In this paper, we propose a zero-shot character recognition framework via radical-based reasoning, called RZCR, to improve the recognition performance of few-sample character categories in the tail. Specifically, we exploit radicals, the graphical units of characters, by decomposing and reconstructing characters according to orthography. RZCR consists of a visual semantic fusion-based radical information extractor (RIE) and a knowledge graph character reasoner (KGR). RIE aims to recognize candidate radicals and their possible structural relations from character images in parallel. The results are then fed into KGR to recognize the target character by reasoning with a knowledge graph. We validate our method on multiple datasets, and RZCR shows promising experimental results, especially on few-sample character datasets.

The conventional lottery ticket hypothesis (LTH) claims that there exists a sparse subnetwork within a dense neural network and a proper random initialization method, called the winning ticket, such that it can be trained from scratch to almost as good as the dense counterpart. Meanwhile, the research of LTH in vision transformers (ViTs) is scarcely evaluated. In this paper, we first show that the conventional winning ticket is hard to find at weight level of ViTs by existing methods. Then, we generalize the LTH for ViTs to input data consisting of image patches inspired by the input dependence of ViTs. That is, there exists a subset of input image patches such that a ViT can be trained from scratch by using only this subset of patches and achieve similar accuracy to the ViTs trained by using all image patches. We call this subset of input patches the winning tickets, which represent a significant amount of information in the input data. We use a ticket selector to generate the winning tickets based on the informativeness of patches for various types of ViT, including DeiT, LV-ViT, and Swin Transformers. The experiments show that there is a clear difference between the performance of models trained with winning tickets and randomly selected subsets, which verifies our proposed theory. We elaborate the analogical similarity between our proposed Data-LTH-ViTs and the conventional LTH for further verifying the integrity of our theory. The Source codes are available at https://github.com/shawnricecake/vit-lottery-ticket-input.

3D motion transfer aims at transferring the motion from a dynamic input sequence to a static 3D object and outputs an identical motion of the target with high-fidelity and realistic visual effects. In this work, we propose a novel 3D Transformer framework called LART for 3D motion transfer. With carefully-designed architectures, LART is able to implicitly learn the correspondence via a flexible geometry perception. Thus, unlike other existing methods, LART does not require any key point annotations or pre-defined correspondence between the motion source and target meshes and can also handle large-size full-detailed unseen 3D targets. Besides, we introduce a novel latent metric regularization on the Transformer for better motion generation. Our rationale lies in the observation that the decoded motions can be approximately expressed as linearly geometric distortion at the frame level. The metric preservation of motions could be translated to the formation of linear paths in the underlying latent space as a rigorous constraint to control the synthetic motions occurring in the construction of the latent space. The proposed LART shows a high learning efficiency with the need for a few samples from the AMASS dataset to generate motions with plausible visual effects. The experimental results verify the potential of our generative model in applications of motion transfer, content generation, temporal interpolation, and motion denoising. The code is made available: https://github.com/mikecheninoulu/LART.

The bird's-eye-view (BEV) perception plays a critical role in autonomous driving systems, involving the accurate and efficient detection and tracking of objects from a top-down perspective. To achieve real-time decision-making in self-driving scenarios, low-latency computation is essential. While recent approaches to BEV detection have focused on improving detection precision using Lift-Splat-Shoot (LSS)-based or transformer-based schemas, the substantial computational and memory burden of these approaches increases the risk of system crashes when multiple on-vehicle tasks run simultaneously. Unfortunately, there is a dearth of literature on efficient BEV detector paradigms, let alone achieving realistic speedups. Unlike existing works that focus on reducing computation costs, this paper focuses on developing an efficient model design that prioritizes actual on-device latency. To achieve this goal, we propose a latency-aware design methodology that considers key hardware properties, such as memory access cost and degree of parallelism. Given the prevalence of GPUs as the main computation platform for autonomous driving systems, we develop a theoretical latency prediction model and introduce efficient building operators. By leveraging these operators and following an effective local-to-global visual modeling process, we propose a hardware-oriented backbone that is also optimized for strong feature capturing and fusing. Using these insights, we present a new hardware-oriented framework for efficient yet accurate camera-view BEV detectors. Experiments show that HotBEV achieves a 2\%$\sim$23\% NDS gain, and 2\%$\sim$7.8\% mAP gain with a 1.1$\times$$\sim$3.4$\times$ speedups compared to existing works on V100; On multiple GPU devices such as GPU GTX 2080 and the low-end GTX 1080, HotBEV achieves 1.1$\times$$\sim$6.3$\times$ faster than others.

While Vision Transformers (ViTs) have undoubtedly made impressive strides in computer vision (CV), their intricate network structures necessitate substantial computation and memory resources. A decision-making process for CV tasks typically entails performing computations with low latency, which is a tricky problem for ViT models. Model quantization is a widely-used technique to optimize the hardware efficiency of deep neural networks. Full quantization under Sub-8-bit precision, in particular, is a promising solution to reduce inference latency significantly. Unfortunately, current commodity hardware, such as CPUs and GPUs, still struggles to efficiently execute these sub-8-bit quantized networks, as their SIMD instructions only support a granularity of 8 bits or wider. Also, there is a scarcity of literature that presents a full quantization paradigm for ViTs. In this paper, we propose an activation-aware fully sub-8-bit quantization-aware training (QAT) framework called PackQViT for efficient yet accurate ViT acceleration on mobile devices to facilitate real-time AI-powered decision-making. Specifically, in revisiting data activation within the ViT dataflow, two characteristics are relevant to quantization strategy and precision: the long-tailed distribution and systematic channel-wise outliers. In response, we employ either log2 quantization or clipping to address the long-tailed distribution and incorporate outlier-aware training for residual link quantization to regulate the various channel-wise outliers more consistently. Notably, due to the systematic fixed pattern, outlier-aware training approach can predict the channel indices and regularized scales of outliers in advance, thus avoiding the runtime data-adaptive selection during inference. Furthermore, we employ Int-$2^{n}$-Softmax, Int-LayerNorm, and Integer GELU to enable integer-only computation flow. Finally, we develop a SIMD-based 4-bit packed multiplier to achieve end-to-end ViT acceleration on mobile phones. Compared to prior studies on ViT quantization using 8-bit precision, PackQViT surpasses other works by an improved accuracy ranging from 0.4\% to 17.9\% for various widely used ViTs on ImageNet dataset; under 4-bit precision, PackQViT demonstrates 0.4%$\sim$2.8% higher accuracy. Compared to the baseline multiplier, our implementations on the Realme GT Android smartphone with Snapdragon 870 SoC CPU achieve 2.6x$\sim$3.7x speedup under 8-bit scenario and 3.8x$\sim$5.9x speedup under 4-bit which ensures practical real-time performance.

Training on large-scale graphs has achieved remarkable results in graph representation learning, but its cost and storage have attracted increasing concerns. Existing graph condensation methods primarily focus on optimizing the feature matrices of condensed graphs while overlooking the impact of the structure information from the original graphs. To investigate the impact of the structure information, we conduct analysis from the spectral domain and empirically identify substantial Laplacian Energy Distribution (LED) shifts in previous works. Such shifts lead to poor performance in cross-architecture generalization and specific tasks, including anomaly detection and link prediction. In this paper, we propose a novel Structure-broadcasting Graph Dataset Distillation (\textbf{SGDD}) scheme for broadcasting the original structure information to the generation of the synthetic one, which explicitly prevents overlooking the original structure information. Theoretically, the synthetic graphs by SGDD are expected to have smaller LED shifts than previous works, leading to superior performance in both cross-architecture settings and specific tasks. We validate the proposed SGDD~across 9 datasets and achieve state-of-the-art results on all of them: for example, on YelpChi dataset, our approach maintains 98.6\% test accuracy of training on the original graph dataset with 1,000 times saving on the scale of the graph. Moreover, we empirically evaluate there exist 17.6\% $\sim$ 31.4\% reductions in LED shift crossing 9 datasets. Extensive experiments and analysis verify the effectiveness and necessity of the proposed designs. The code will be made public.

Recently, due to the increasing requirements of medical imaging applications and the professional requirements of annotating medical images, few-shot learning has gained increasing attention in the medical image semantic segmentation field. To perform segmentation with limited number of labeled medical images, most existing studies use Prototypical Networks (PN) and have obtained compelling success. However, these approaches overlook the query imagefeatures extracted from the proposed representation network, failing to preserving the spatial connection between query and support images. In this paper, we propose a novel self-supervised few-shot medical image segmentation network and introduce a novel Cycle-Resemblance Attention (CRA) module to fully leverage the pixel-wise relation between query and support medical images. Notably, we first line up multiple attention blocks to refine more abundant relation information. Then, we present CRAPNet by integrating the CRA module with a classic prototype network, where pixel-wise relations between query and support features are well recaptured for segmentation. Extensive experiments on two different medical image datasets, e.g., abdomen MRI and abdomen CT, demonstrate the superiority of our model over existing state-of-the-art methods.

In the last few years, Neural Painting (NP) techniques became capable of producing extremely realistic artworks. This paper advances the state of the art in this emerging research domain by proposing the first approach for Interactive NP. Considering a setting where a user looks at a scene and tries to reproduce it on a painting, our objective is to develop a computational framework to assist the user’s creativity by suggesting the next strokes to paint, that can be possibly used to complete the artwork. To accomplish such a task, we propose I-Paint, a novel method based on a conditional transformer Variational AutoEncoder (VAE) architecture with a two-stage decoder. To evaluate the proposed approach and stimulate research in this area, we also introduce two novel datasets. Our experiments show that our approach provides good stroke suggestions and compares favorably to the state of the art.

In this paper, we address the task of semantic-guided image generation. One challenge common to most existing image-level generation methods is the difficulty in generating small objects and detailed local textures. To address this, in this work we consider generating images using local context. As such, we design a local class-specific generative network using semantic maps as guidance, which separately constructs and learns subgenerators for different classes, enabling it to capture finer details. To learn more discriminative class-specific feature representations for the local generation, we also propose a novel classification module. To combine the advantages of both global image-level and local class-specific generation, a joint generation network is designed with an attention fusion module and a dual-discriminator structure embedded. Lastly, we propose a novel semantic-aware upsampling method, which has a larger receptive field and can take far-away pixels that are semantically related for feature upsampling, enabling it to better preserve semantic consistency for instances with the same semantic labels. Extensive experiments on two image generation tasks show the superior performance of the proposed method. State-of-the-art results are established by large margins on both tasks and on nine challenging public benchmarks. The source code and trained models are available at https://github.com/Ha0Tang/LGGAN.

We propose a noveledge guidedgenerativeadversarialnetwork withcontrastive learning (ECGAN) for the challenging semantic image synthesis task. Although considerable improvements have been achieved by the community in the recent period, the quality of synthesized images is far from satisfactory due to three largely unresolved challenges. 1) The semantic labels do not provide detailed structural information, making it challenging to synthesize local details and structures; 2) The widely adopted CNN operations such as convolution, down-sampling, and normalization usually cause spatial resolution loss and thus cannot fully preserve the original semantic information, leading to semantically inconsistent results (e.g., missing small objects); 3) Existing semantic image synthesis methods focus on modeling “local” semantic information from a single input semantic layout. However, they ignore “global” semantic information of multiple input semantic layouts, i.e., semantic cross-relations between pixels across different input layouts. To tackle 1), we propose to use the edge as an intermediate representation which is further adopted to guide image generation via a proposed attention guided edge transfer module. Edge information is produced by a convolutional generator and introduces detailed structure information. To tackle 2), we design an effective module to selectively highlight class-dependent feature maps according to the original semantic layout to preserve the semantic information. To tackle 3), inspired by current methods in contrastive learning, we propose a novel contrastive learning method, which aims to enforce pixel embeddings belonging to the same semantic class to generate more similar image content than those from different classes. We further propose a novel multi-scale contrastive learning method that aims to push same-class features from different scales closer together being able to capture more semantic relations by explicitly exploring the structures of labeled pixels from multiple input semantic layouts from different scales. Experiments on three challenging datasets show that our methods achieve significantly better results than state-of-the-art approaches. The source code is available athttps://github.com/Ha0Tang/ECGAN.

We propose a novel model named Multi-Channel Attention Selection Generative Adversarial Network (SelectionGAN) for guided image-to-image translation, where we translate an input image into another while respecting an external semantic guidance. The proposed SelectionGAN explicitly utilizes the semantic guidance information and consists of two stages. In the first stage, the input image and the conditional semantic guidance are fed into a cycled semantic-guided generation network to produce initial coarse results. In the second stage, we refine the initial results by using the proposed multi-scale spatial pooling & channel selection module and the multi-channel attention selection module. Moreover, uncertainty maps automatically learned from attention maps are used to guide the pixel loss for better network optimization. Exhaustive experiments on four challenging guided image-to-image translation tasks (face, hand, body, and street view) demonstrate that our SelectionGAN is able to generate significantly better results than the state-of-the-art methods. Meanwhile, the proposed framework and modules are unified solutions and can be applied to solve other generation tasks such as semantic image synthesis. The code is available at https://github.com/Ha0Tang/SelectionGAN.

Despite significant progress, estimating 3D human poses from monocular videos remains a challenging task due to depth ambiguity and self-occlusion. Most existing works attempt to solve both issues by exploiting spatial and temporal relationships. However, those works ignore the fact that it is an inverse problem where multiple feasible solutions ( i.e. , hypotheses) exist. To relieve this limitation, we propose a Multi-Hypothesis Transformer that learns spatio-temporal representations of multiple plausible pose hypotheses. In order to effectively model multi-hypothesis dependencies and build strong relationships across hypothesis features, we introduce a one-to-many-to-one three-stage framework: (i) Generate multiple initial hypothesis representations; (ii) Model self-hypothesis communication, merge multiple hypotheses into a single converged representation and then partition it into several diverged hypotheses; (iii) Learn cross-hypothesis communication and aggregate the multi-hypothesis features to synthesize the final 3D pose. Through the above processes, the final representation is enhanced and the synthesized pose is much more accurate. Extensive experiments show that the proposed method achieves state-of-the-art results on two challenging datasets: Human3.6M and MPI-INF-3DHP. The code and models are available at https://github.com/Vegetebird/MHFormer .

Arbitrary-oriented object detection (AOOD) is a challenging task to detect objects in the wild with arbitrary orientations and cluttered arrangements. Existing approaches are mainly based on anchor-based boxes or dense points, which rely on complicated hand-designed processing steps and inductive bias, such as anchor generation, transformation, and non-maximum suppression reasoning. Recently, the emerging transformer-based approaches view object detection as a direct set prediction problem that effectively removes the need for hand-designed components and inductive biases. In this paper, we propose an Arbitrary-Oriented Object DEtection TRansformer framework, termed AO2-DETR, which comprises three dedicated components. More precisely, an oriented proposal generation mechanism is proposed to explicitly generate oriented proposals, which provides better positional priors for pooling features to modulate the cross-attention in the transformer decoder. An adaptive oriented proposal refinement module is introduced to extract rotation-invariant region features and eliminate the misalignment between region features and objects. And a rotation-aware set matching loss is used to ensure the one-to-one matching process for direct set prediction without duplicate predictions. Our method considerably simplifies the overall pipeline and presents a new AOOD paradigm. Comprehensive experiments on several challenging datasets show that our method achieves superior performance on the AOOD task.

Camouflaged Object Detection (COD) aims to detect objects well hidden in the environment. The main challenges of COD come from the high degree of texture and color overlapping between the objects and their surroundings. Inspired by that humans tend to go closer to the object and magnify it to recognize ambiguous objects more clearly, we propose a novel three-stage architecture called Search-Amplify-Recognize and design a network SARNet to address the challenges. Specifically, In the Search part, we utilize an attention-based backbone to locate the object. In the Amplify part, to obtain rich searched features and fine segmentation, we design Object Area Amplification modules (OAA) to perform cross-level and adjacent-level feature fusion and amplifying operations on feature maps. Besides, the OAA can be regarded as a simple and effective plug-in module to integrate and amplify the feature maps. The main components of the Recognize part are the Figure-Ground Conversion modules (FGC). The FGC modules alternately pay attention to the foreground and background to precisely separate the highly similar foreground and background. Extensive experiments on benchmark datasets show that our model outperforms other SOTA methods not only on COD tasks but also in COD downstream tasks, such as polyp segmentation and video camouflaged object detection. Source codes will be available athttps://github.com/Haozhe-Xing/SARNet.

Deep learning-based synthetic aperture radar (SAR) automatic target recognition (ATR) algorithms have achieved outstanding performance under the condition of hundreds or thousands of training samples in recent years. Nevertheless, it is often rare to acquire great quantities of target samples in real SAR application scenarios. This article proposes a novel ATR method called transductive prototypical attention reasoning network (TPARN) to solve the problem of SAR target recognition with only a few training samples. To be specific, a region awareness-based feature extraction model is first developed, which can effectively focus on the target region of interest and suppress the background clutter by embedding direction-aware and position-sensitive information to extract more transferable knowledge. To heighten the discrimination of the sample features, a cross-feature spatial attention module is then proposed following the feature embedding model. Finally, a transductive prototype reasoning method is presented to realize the identity reasoning of the target, which can continuously update each class prototype with training samples and test samples together, thereby improving the classification accuracy. In addition, a marginal adaptive hybrid loss is proposed to obtain a discriminative feature embedding space with intra-class compactness and inter-class divergence, aiming to facilitate subsequent target identity reasoning. Extensive experiments on the moving and stationary target acquisition and recognition (MSTAR) benchmark dataset reveal that the proposed method outperforms some state-of-the-arts under different few-shot SAR ATR tasks.

Data-driven automatic target recognition (ATR) methods have become the mainstream in the synthetic aperture radar (SAR) community at this stage. However, in real SAR application scenarios, the scarcity of training samples is a common problem. Especially in military application scenarios, only a small number of samples of each type of target are usually available. In the case of limited available samples, it is bound to bring challenges to the feature extraction model and classifier inference learning. In this paper, we propose a novel method called adaptive convolutional subspace reasoning network (ACSRNet) to address few-shot SAR target recognition tasks. To be specific, we first present a dynamic-aware convolutional feature embedding network based on the siamese architecture, which can not only learn more transferable knowledge for the few-shot tasks, but also dynamically adjust the convolution kernel according to the input data to extract more discriminative features. To effectively achieve target identity reasoning, we then resort to high-order information of samples, i.e., the idea of adaptive subspace learning, to develop a few-shot subspace classification module, which can online reason the identity of the unknown target through the spanning subspace of training samples. Meanwhile, a dual-loss is designed to train a feature embedding space with intra-class compactness and inter-class divergence, aiming to facilitate subsequent classification. Meta-learning integrating random sampling episode way is introduced into the process of model training to realize few-shot SAR ATR tasks by emulating the human cognitive process. Experimental results on the moving and stationary target acquisition recognition (MSTAR) dataset demonstrate that the proposed method is competitive with some state-of-the-arts for few-shot SAR ATR tasks.

We address the challenging task of human reaction generation, which aims to generate a corresponding reaction based on an input action. Most of the existing works do not focus on generating and predicting the reaction and cannot generate the motion when only the action is given as input. To address this limitation, we propose a novel interaction Transformer (InterFormer) consisting of a Transformer network with both temporal and spatial attention. Specifically, temporal attention captures the temporal dependencies of the motion of both characters and of their interaction, while spatial attention learns the dependencies between the different body parts of each character and those which are part of the interaction. Moreover, we propose using graphs to increase the performance of spatial attention via an interaction distance module that helps focus on nearby joints from both characters. Extensive experiments on the SBU interaction, K3HI, and DuetDance datasets demonstrate the effectiveness of InterFormer. Our method is general and can be used to generate more complex and long-term interactions. We also provide videos of generated reactions and the code with pre-trained models athttps://github.com/CRISTAL-3DSAM/InterFormer

In this paper, we tackle the problem of ground-view panorama image generation conditioning on top-view aerial image, which is a challenging problem due to large gap between image domains associated with different view-points. Instead of learning the underlying cross-view mapping by a feedforward network as previous methods, we propose a novel adversarial feedback GAN framework named PanoGAN consisting of two key components: an adversarial feedback module and a dual branch discrimination strategy. First, aerial image is fed into the generator to produce panorama image and segmentation map, which facilitates the use of semantic layout information for model training. Second, the discriminator's feature responses of model outputs are encoded by the adversarial feedback module and then fed back to the generator for the next round of generation. Continual improvement of generated image quality is achieved through an iterative generation process. Third, to pursue high-fidelity and semantic consistency of the generated panorama image, we propose a pixel-segmentation alignment mechanism under the dual branch discrimiantion strategy that promotes the cooperation between generator and discriminator. Extensive experimental results on two challenging cross-view image datasets show that the proposed PanoGAN generates high-quality panorama images with more convincing details than state-of-the-art methods. The source code and trained models are available at https://github.com/sswuai/PanoGAN.

International audience

State-of-the-art methods in the image-to-image translation are capable of learning a mapping from a source domain to a target domain with unpaired image data. Though the existing methods have achieved promising results, they still produce visual artifacts, being able to translate low-level information but not high-level semantics of input images. One possible reason is that generators do not have the ability to perceive the most discriminative parts between the source and target domains, thus making the generated images low quality. In this article, we propose a new Attention-Guided Generative Adversarial Networks (AttentionGAN) for the unpaired image-to-image translation task. AttentionGAN can identify the most discriminative foreground objects and minimize the change of the background. The attention-guided generators in AttentionGAN are able to produce attention masks, and then fuse the generation output with the attention masks to obtain high-quality target images. Accordingly, we also design a novel attention-guided discriminator which only considers attended regions. Extensive experiments are conducted on several generative tasks with eight public datasets, demonstrating that the proposed method is effective to generate sharper and more realistic images compared with existing competitive models. The code is available at https://github.com/Ha0Tang/AttentionGAN.

Video processing and analysis have become an urgent task, as a huge amount of videos (e.g., YouTube, Hulu) are uploaded online every day. The extraction of representative key frames from videos is important in video processing and analysis since it greatly reduces computing resources and time. Although great progress has been made recently, large-scale video classification remains an open problem, as the existing methods have not well balanced the performance and efficiency simultaneously. To tackle this problem, this work presents an unsupervised method to retrieve the key frames, which combines the convolutional neural network and temporal segment density peaks clustering. The proposed temporal segment density peaks clustering is a generic and powerful framework, and it has two advantages compared with previous works. One is that it can calculate the number of key frames automatically. The other is that it can preserve the temporal information of the video. Thus, it improves the efficiency of video classification. Furthermore, a long short-term memory network is added on the top of the convolutional neural network to further elevate the performance of classification. Moreover, a weight fusion strategy of different input networks is presented to boost performance. By optimizing both video classification and key frame extraction simultaneously, we achieve better classification performance and higher efficiency. We evaluate our method on two popular datasets (i.e., HMDB51 and UCF101), and the experimental results consistently demonstrate that our strategy achieves competitive performance and efficiency compared with the state-of-the-art approaches.

Salient object detection has been long studied to identify the most visually attractive objects in images/videos. Recently, a growing amount of approaches have been proposed, all of which rely on the contour/edge information to improve detection performance. The edge labels are either put into the loss directly or used as extra supervision. The edge and body can also be learned separately and then fused afterward. Both methods either lead to high prediction errors near the edge or cannot be trained in an end-to-end manner. Another problem is that existing methods may fail to detect objects of various sizes due to the lack of efficient and effective feature fusion mechanisms. In this work, we propose to decompose the saliency detection task into two cascaded sub-tasks, i.e., detail modeling and body filling. Specifically, detail modeling focuses on capturing the object edges by supervision of explicitly decomposed detail label that consists of the pixels that are nested on the edge and near the edge. Then the body filling learns the body part that will be filled into the detail map to generate more accurate saliency map. To effectively fuse the features and handle objects at different scales, we have also proposed two novel multi-scale detail attention and body attention blocks for precise detail and body modeling. Experimental results show that our method achieves state-of-the-art performances on six public datasets.

The mainstream motion prediction methods usually focus on short-term prediction, and their predicted long-term motions often fall into an average pose, i.e., the freezing forecasting problem [ 27 ]. To mitigate this problem, we propose a novel Bidirectional Transformer-based Generative Adversarial Network (BiTGAN) for long-term human motion prediction. The bidirectional setup leads to consistent and smooth generation in both forward and backward directions. Besides, to make full use of the history motions, we split them into two parts. The first part is fed to the Transformer encoder in our BiTGAN while the second part is used as the decoder input. This strategy can alleviate the exposure problem [ 37 ]. Additionally, to better maintain both the local (i.e., frame-level pose) and global (i.e., video-level semantic) similarities between the predicted motion sequence and the real one, the soft dynamic time warping (Soft-DTW) loss is introduced into the generator. Finally, we utilize a dual-discriminator to distinguish the predicted sequence at both frame and sequence levels. Extensive experiments on the public Human3.6M dataset demonstrate that our proposed BiTGAN achieves state-of-the-art performance on long-term (4 s ) human motion prediction, and reduces the average error of all actions by 4%.

We present a customized 3D mesh Transformer model for the pose transfer task. As the 3D pose transfer essentially is a deformation procedure dependent on the given meshes, the intuition of this work is to perceive the geometric inconsistency between the given meshes with the powerful self-attention mechanism. Specifically, we propose a novel geometry-contrastive Transformer that has an efficient 3D structured perceiving ability to the global geometric inconsistencies across the given meshes. Moreover, locally, a simple yet efficient central geodesic contrastive loss is further proposed to improve the regional geometric-inconsistency learning. At last, we present a latent isometric regularization module together with a novel semi-synthesized dataset for the cross-dataset 3D pose transfer task towards unknown spaces. The massive experimental results prove the efficacy of our approach by showing state-of-the-art quantitative performances on SMPL-NPT, FAUST and our new proposed dataset SMG-3D datasets, as well as promising qualitative results on MG-cloth and SMAL datasets. It's demonstrated that our method can achieve robust 3D pose transfer and be generalized to challenging meshes from unknown spaces on cross-dataset tasks. The code and dataset are made available. Code is available: https://github.com/mikecheninoulu/CGT.

Multi-modal fusion is proven to be an effective method to improve the accuracy and robustness of speaker tracking, especially in complex scenarios. However, how to combine the heterogeneous information and exploit the complementarity of multi-modal signals remains a challenging issue. In this paper, we propose a novel Multi-modal Perception Tracker (MPT) for speaker tracking using both audio and visual modalities. Specifically, a novel acoustic map based on spatial-temporal Global Coherence Field (stGCF) is first constructed for heterogeneous signal fusion, which employs a camera model to map audio cues to the localization space consistent with the visual cues. Then a multi-modal perception attention network is introduced to derive the perception weights that measure the reliability and effectiveness of intermittent audio and video streams disturbed by noise. Moreover, a unique cross-modal self-supervised learning method is presented to model the confidence of audio and visual observations by leveraging the complementarity and consistency between different modalities. Experimental results show that the proposed MPT achieves 98.6% and 78.3% tracking accuracy on the standard and occluded datasets, respectively, which demonstrates its robustness under adverse conditions and outperforms the current state-of-the-art methods.

This paper presents a novel Physically-guided Disentangled Implicit Rendering (PhyDIR) framework for highfidelity 3D face modeling. The motivation comes from two observations: Widely-used graphics renderers yield excessive approximations against photo-realistic imaging, while neural rendering methods produce superior appearances but are highly entangled to perceive 3D-aware operations. Hence, we learn to disentangle the implicit rendering via explicit physical guidance, while guaranteeing the properties of: (1) 3D-aware comprehension and (2) high-reality image formation. For the former one, PhyDIR explicitly adopts 3D shading and rasterizing modules to control the renderer, which disentangles the light, facial shape, and viewpoint from neural reasoning. Specifically, PhyDIR proposes a novel multi-image shading strategy to compensate for the monocular limitation, so that the lighting variations are accessible to the neural renderer. For the latter, PhyDIR learns the face-collection implicit texture to avoid ill-posed intrinsic factorization, then leverages a series of consistency losses to constrain the rendering robustness. With the disentangled method, we make 3D face modeling benefit from both kinds of rendering strategies. Extensive experiments on benchmarks show that PhyDIR obtains superior performance than state-of-the-art explicit/implicit methods on geometry/texture modeling.

In-the-wild 3D face modelling is a challenging problem as the predicted facial geometry and texture suffer from a lack of reliable clues or priors, when the input images are degraded. To address such a problem, in this paper we propose a novel Learning to Restore (L2R) 3D face framework for unsupervised high-quality face reconstruction from low-resolution images. Rather than directly refining 2D image appearance, L2R learns to recover fine-grained 3D details on the proxy against degradation via extracting generative facial priors. Concretely, L2R proposes a novel albedo restoration network to model high-quality 3D facial texture, in which the diverse guidance from the pre-trained Generative Adversarial Networks (GANs) is leveraged to complement the lack of input facial clues. With the finer details of the restored 3D texture, L2R then learns displacement maps from scratch to enhance the significant facial structure and geometry. Both of the procedures are mutually optimized with a novel 3D-aware adversarial loss, which further improves the modelling performance and suppresses the potential uncertainty. Extensive experiments on benchmarks show that L2R outperforms state-of-the-art methods under the condition of low-quality inputs, and obtains superior performances than 2D pre-processed modelling approaches with limited 3D proxy.

Estimating 3D human poses from monocular videos is a challenging task due to depth ambiguity and self-occlusion. Most existing works attempt to solve both issues by exploiting spatial and temporal relationships. However, those works ignore the fact that it is an inverse problem where multiple feasible solutions (i.e., hypotheses) exist. To relieve this limitation, we propose a Multi-Hypothesis Transformer (MHFormer) that learns spatio-temporal representations of multiple plausible pose hypotheses. In order to effectively model multi-hypothesis dependencies and build strong relationships across hypothesis features, the task is decomposed into three stages: (i) Generate multiple initial hypothesis representations; (ii) Model self-hypothesis communication, merge multiple hypotheses into a single converged representation and then partition it into several diverged hypotheses; (iii) Learn cross-hypothesis communication and aggregate the multi-hypothesis features to synthesize the final 3D pose. Through the above processes, the final representation is enhanced and the synthesized pose is much more accurate. Extensive experiments show that MHFormer achieves state-of-the-art results on two challenging datasets: Human3.6M and MPI-INF-3DHP. Without bells and whistles, its performance surpasses the previous best result by a large margin of 3% on Human3.6M. Code and models are available at https://github.com/Vegetebird/MHFormer.

Synthesizing high-quality realistic images from text descriptions is a challenging task. Existing text-to-image Generative Adversarial Networks generally employ a stacked architecture as the backbone yet still remain three flaws. First, the stacked architecture introduces the entanglements between generators of different image scales. Second, existing studies prefer to apply and fix extra networks in adversarial learning for text-image semantic consistency, which limits the supervision capability of these networks. Third, the cross-modal attention-based text-image fusion that widely adopted by previous works is limited on several special image scales because of the computational cost. To these ends, we propose a simpler but more effective Deep Fusion Generative Adversarial Networks (DF-GAN). To be specific, we propose: (i) a novel one-stage text-to-image backbone that directly synthesizes high-resolution images without entanglements between different generators, (ii) a novel Target-Aware Discriminator composed of Matching-Aware Gradient Penalty and One-Way Output, which enhances the text-image semantic consistency without introducing extra networks, (iii) a novel deep text-image fusion block, which deepens the fusion process to make a full fusion between text and visual features. Compared with current state-of-the-art methods, our proposed DF-GAN is simpler but more efficient to synthesize realistic and text-matching images and achieves better performance on widely used datasets. Code is available at https://github.com/tobran/DF-GAN.

To achieve disentangled image manipulation, previous works depend heavily on manual annotation. Meanwhile, the available manipulations are limited to a pre-defined set the models were trainedfor. We propose a novelframework, i.e., Predict, Prevent, and Evaluate (PPE), for disentangled text-driven image manipulation that requires little manual annotation while being applicable to a wide variety of ma-nipulations. Our method approaches the targets by deeply exploiting the power of the large-scale pre-trained vision-language model CLIP [32]. Concretely, we firstly Predict the possibly entangled attributes for a given text command. Then, based on the predicted attributes, we introduce an entanglement loss to Prevent entanglements during training. Finally, we propose a new evaluation metric to Evaluate the disentangled image manipulation. We verify the effectiveness of our method on the challenging face editing task. Extensive experiments show that the proposed PPE frame-work achieves much better quantitative and qualitative re-sults than the up-to-date StyleCLIP [31] baseline. Code is available at https://github.com/zipengxuc/PPE.

Vision transformers (ViTs) are emerging with significantly improved accuracy in computer vision tasks. However, their complex architecture and enormous computation/storage demand impose urgent needs for new hardware accelerator design methodology. This work proposes an FPGA-aware automatic ViT acceleration framework based on the proposed mixed-scheme quantization. To the best of our knowledge, this is the first FPGA-based ViT acceleration framework exploring model quantization. Compared with state-of-the-art ViT quantization work (algorithmic approach only without hardware acceleration), our quantization achieves 0.31% to 1.25% higher Top-1 accuracy under the same bit-width. Compared with the 32-bit floating-point baseline FPGA accelerator, our accelerator achieves around 5.6× improvement on the frame rate (i.e., 56.4 FPS vs. 10.0 FPS) with 0.83% accuracy drop for DeiT-base.

Vision transformers (ViTs) are emerging with significantly improved accuracy in computer vision tasks. However, their complex architecture and enormous computation/storage demand impose urgent needs for new hardware accelerator design methodology. This work proposes an FPGA-aware automatic ViT acceleration framework based on the proposed mixed-scheme quantization. To the best of our knowledge, this is the first FPGA-based ViT acceleration framework exploring model quantization. Compared with state-of-the-art ViT quantization work (algorithmic approach only without hardware acceleration), our quantization achieves 0.47% to 1.36% higher Top-l accuracy under the same bit-width. Compared with the 32-bit floating-point baseline FPGA accelerator, our accelerator achieves around 5.6x improvement on the frame rate (i.e., 56.8 FPS vs. 10.0 FPS) with 0.71% accuracy drop on ImageNet dataset for DeiT-base.

Unsupervised domain adaptation (UDA) person re-identification (re-ID) aims to transfer knowledge learned from labeled source domain to unlabeled target domain and has been successfully applied into a wide range of real-world scenarios. However, existing methods are mainly ineffective at handling domain shift as well as being sensitive to camera styles due to the unannotated target domain. In this paper, we pro-pose a Camera-style Separation and Uncertainty Estimation (CSUE) model to address the problem from two perspectives. To alleviate the negative effect of cross-camera variation, we introduce the Camera-aware Style Decoupling module to im-pose inter-and-intra camera constraints on the feature extracting stage. It can better mine and describe the latent camera invariant features. Moreover, to avoid the inherent defect of clustering, an Uncertainty Modeling module is constructed via estimating the certainty, which helps progressively refine the pseudo labels. Extensive experiments on widely used datasets demonstrate the state-of-the-art performance of our model under the UDA re-ID setting.

Cloth-changing person re-identification (CC-ReID), which aims to match person identities under clothing changes, is a new rising research topic in recent years. However, typical biometrics-based CC-ReID methods often require cumber-some pose or body part estimators to learn cloth-irrelevant features from human biometric traits, which comes with high computational costs. Besides, the performance is significantly limited due to the resolution degradation of surveillance images. To address the above limitations, we propose an effective Identity-Sensitive Knowledge Propagation framework (DeSKPro) for CC-ReID. Specifically, a Cloth-irrelevant Spatial Attention module is introduced to eliminate the distraction of clothing appearance by acquiring knowledge from the human parsing module. To mitigate the resolution degradation issue and mine identity-sensitive cues from human faces, we propose to restore the missing facial details using prior facial knowledge, which is then propagated to a smaller network. After training, the extra computations for human parsing or face restoration are no longer required. Extensive experiments show that our framework outperforms state-of-the-art methods by a large margin. Our code is available at https://github.com/KimbingNg/DeskPro.

Cloth-changing person re-identification (CC-ReID) is a challenging task that aims at retrieving the target person across large spatial and temporal spans, with a high probability of changing clothes. To explicitly alleviate the impact of the person changing clothes on re-identification, this paper presents a cloth-irrelevant harmonious attention network (CIHANet) that learns cloth-irrelevant knowledge. Firstly, with the help of human parsing, the color information of human clothing is removed to generate black clothes images. Secondly, the raw person images are used to learn features with more color-based appearance knowledge, while the black clothes images are used to learn features with more cloth-irrelevant knowledge. Then, to fuse the knowledge of two distinct streams, we propose the harmonious attention module, including mutual learning attention and salience guided attention mechanisms. The mutual learning attention mechanism adaptively selects identity-relevant features across feature channels to make two streams interact with each other. The salience guided attention mechanism highlights the cloth-irrelevant areas by transferring the spatial knowledge from the black clothes stream to the raw images stream. Finally, quantitative and qualitative results on three CC-ReID datasets validate the superiority of our method on the CC-ReID task.

In this work we demonstrate real-time portrait stylization, specifically, translating self-portrait into cartoon or anime style on mobile devices. We propose a latency-driven differentiable architecture search method, maintaining realistic generative quality. With our framework, we obtain 10× computation reduction on the generative model and achieve real-time video stylization on off-the-shelf smartphone using mobile GPUs.

Degraded images commonly exist in the general sources of character images, leading to unsatisfactory character recognition results. Existing methods have dedicated efforts to restoring degraded character images. However, the denoising results obtained by these methods do not appear to improve character recognition performance. This is mainly because current methods only focus on pixel-level information and ignore critical features of a character, such as its glyph, resulting in character-glyph damage during the denoising process. In this paper, we introduce a novel generic framework based on glyph fusion and attention mechanisms, i.e., CharFormer, for precisely recovering character images without changing their inherent glyphs. Unlike existing frameworks, CharFormer introduces a parallel target task for capturing additional information and injecting it into the image denoising backbone, which will maintain the consistency of character glyphs during character image denoising. Moreover, we utilize attention-based networks for global-local feature interaction, which will help to deal with blind denoising and enhance denoising performance. We compare CharFormer with state-of-the-art methods on multiple datasets. The experimental results show the superiority of CharFormer quantitatively and qualitatively.

Constructing high-quality character image datasets is challenging because real-world images are often affected by image degradation. There are limitations when applying current image restoration methods to such real-world character images, since (i) the categories of noise in character images are different from those in general images; (ii) real-world character images usually contain more complex image degradation, e.g., mixed noise at different noise levels. To address these problems, we propose a real-world character restoration network (RCRN) to effectively restore degraded character images, where character skeleton information and scale-ensemble feature extraction are utilized to obtain better restoration performance. The proposed method consists of a skeleton extractor (SENet) and a character image restorer (CiRNet). SENet aims to preserve the structural consistency of the character and normalize complex noise. Then, CiRNet reconstructs clean images from degraded character images and their skeletons. Due to the lack of benchmarks for real-world character image restoration, we constructed a dataset containing 1,606 character images with real-world degradation to evaluate the validity of the proposed method. The experimental results demonstrate that RCRN outperforms state-of-the-art methods quantitatively and qualitatively.

We propose a simple yet powerful Landmark guided Generative Adversarial Network (LandmarkGAN) for the facial expression-to-expression translation using a single image, which is an important and challenging task in computer vision since the expression-to-expression translation is a non-linear and non-aligned problem. Moreover, it requires a high-level semantic understanding between the input and output images since the objects in images can have arbitrary poses, sizes, locations, backgrounds, and self-occlusions. To tackle this problem, we propose utilizing facial landmark information explicitly. Since it is a challenging problem, we split it into two sub-tasks, (i) category-guided landmark generation, and (ii) landmark-guided expression-to-expression translation. Two sub-tasks are trained in an end-to-end fashion that aims to enjoy the mutually improved benefits from the generated landmarks and expressions. Compared with current keypoint-guided approaches, the proposed LandmarkGAN only needs a single facial image to generate various expressions. Extensive experimental results on four public datasets demonstrate that the proposed LandmarkGAN achieves better results compared with state-of-the-art approaches only using a single image.

Long-range contextual information is crucial for the semantic segmentation of high-resolution (HR) remote sensing images (RSIs). However, image cropping operations, commonly used for training neural networks, limit the perception of long-range contexts in large RSIs. To overcome this limitation, we propose a wide-context network (WiCoNet) for the semantic segmentation of HR RSIs. Apart from extracting local features with a conventional convolutional neural network (CNN), the WiCoNet has an extra context branch to aggregate information from a larger image area. Moreover, we introduce a context transformer to embed contextual information from the context branch and selectively project it onto the local features. The context transformer extends the vision transformer, an emerging kind of neural networks, to model the dual-branch semantic correlations. It overcomes the locality limitation of CNNs and enables the WiCoNet to see the bigger picture before segmenting the land-cover/land-use (LCLU) classes. Ablation studies and comparative experiments conducted on several benchmark datasets demonstrate the effectiveness of the proposed method. In addition, we present a new Beijing Land-Use (BLU) dataset. This is a large-scale HR satellite dataset with high-quality and fine-grained reference labels, which can facilitate future studies in this field.

Ship detection in optical remote sensing images plays a significant role in a wide range of civilian and military tasks. However, it is still a challenging issue owing to complex environmental interferences and a large variety of target scales and positions. To overcome these limitations, we propose a supervised multi-scale attention-guided detection framework, which can effectively detect ships of different scales both in complex pure ocean and port scenes. Specifically, a multi-scale supervision module is first proposed to adjust the semantic consistency of different feature levels, obtaining extracted features with small semantic gaps. Next, an attention-guided module is utilized to aggregate context information from both spatial and channel dimensions by calculating map correlations, adaptively enhancing the feature representation. Moreover, to preserve the attribute and spatial relationship of the optimized features, we adopt a capsule-based module as the classifier and obtain satisfactory classification performance. Experimental results conducted on two public high-quality datasets demonstrate that the proposed method obtains state-of-the-art performance in comparison with several advanced methods.

Building extraction in VHR RSIs remains a challenging task due to occlusion and boundary ambiguity problems. Although conventional convolutional neural networks (CNNs) based methods are capable of exploiting local texture and context information, they fail to capture the shape patterns of buildings, which is a necessary constraint in the human recognition. To address this issue, we propose an adversarial shape learning network (ASLNet) to model the building shape patterns that improve the accuracy of building segmentation. In the proposed ASLNet, we introduce the adversarial learning strategy to explicitly model the shape constraints, as well as a CNN shape regularizer to strengthen the embedding of shape features. To assess the geometric accuracy of building segmentation results, we introduced several object-based quality assessment metrics. Experiments on two open benchmark datasets show that the proposed ASLNet improves both the pixel-based accuracy and the object-based quality measurements by a large margin. The code is available at: https://github.com/ggsDing/ASLNet.

Modern saliency detection models are based on the encoder-decoder framework and they use different strategies to fuse the multi-level features between the encoder and decoder to boost representation power. Motivated by recent work in implicit modelling, we propose to introduce an implicit function to simulate the equilibrium state of the feature pyramid at infinite depths. We question the existence of the ideal equilibrium and thus propose a quasi-equilibrium model by taking the first-order derivative into the black-box root solver using Taylor expansion. It models more realistic convergence states and significantly improves the network performance. We also propose a differentiable edge extractor that directly extracts edges from the saliency masks. By optimizing the extracted edges, the generated saliency masks are naturally optimized on contour constraints and the non-deterministic predictions are removed. We evaluate the proposed methodology on five public datasets and extensive experiments show that our method achieves new state-of-the-art performances on six metrics across datasets.

This paper proposes a gaze correction and animation method for high-resolution, unconstrained portrait images, which can be trained without the gaze angle and the head pose annotations. Common gaze-correction methods usually require annotating training data with precise gaze, and head pose information. Solving this problem using an unsupervised method remains an open problem, especially for high-resolution face images in the wild, which are not easy to annotate with gaze and head pose labels. To address this issue, we first create two new portrait datasets: CelebGaze ( 256 ×256 ) and high-resolution CelebHQGaze ( 512 ×512 ). Second, we formulate the gaze correction task as an image inpainting problem, addressed using a Gaze Correction Module (GCM) and a Gaze Animation Module (GAM). Moreover, we propose an unsupervised training strategy, i.e., Synthesis-As-Training, to learn the correlation between the eye region features and the gaze angle. As a result, we can use the learned latent space for gaze animation with semantic interpolation in this space. Moreover, to alleviate both the memory and the computational costs in the training and the inference stage, we propose a Coarse-to-Fine Module (CFM) integrated with GCM and GAM. Extensive experiments validate the effectiveness of our method for both the gaze correction and the gaze animation tasks in both low and high-resolution face datasets in the wild and demonstrate the superiority of our method with respect to the state of the art.

We propose a novel and unified Cycle In Cycle Generative Adversarial Network (C2GAN) for generating human faces, hands, bodies, and natural scenes. Our proposed C2GAN is a cross-modal model exploring a joint exploitation of the input image data and guidance data in an interactive manner. C2GAN contains two different generators, i.e., an image-generation generator and a guidance-generation generator. Both generators are mutually connected and trained in an end-to-end fashion and explicitly form three cycled subnets, i.e., one image generation cycle and two guidance generation cycles. Each cycle aims at reconstructing the input domain and simultaneously produces a useful output involved in the generation of another cycle. In this way, the cycles constrain each other implicitly providing complementary information from both image and guidance modalities and bringing an extra supervision gradient across the cycles, facilitating a more robust optimization of the whole model. Extensive experimental results on four guided image-to-image translation subtasks, i.e., person image generation, facial expression generation, hand gesture-to-gesture translation, and cross-view image translation, demonstrate that the proposed C2GAN is effective in generating more realistic images compared with state-of-the-art models.

With the strength of deep generative models, 3D pose transfer regains intensive research interests in recent years. Existing methods mainly rely on a variety of constraints to achieve the pose transfer over 3D meshes, e.g., the need for manually encoding for shape and pose disentanglement. In this paper, we present an unsupervised approach to conduct the pose transfer between any arbitrate given 3D meshes. Specifically, a novel Intrinsic-Extrinsic Preserved Generative Adversarial Network (IEP-GAN) is presented for both intrinsic (i.e., shape) and extrinsic (i.e., pose) information preservation. Extrinsically, we propose a co-occurrence discriminator to capture the structural/pose invariance from distinct Laplacians of the mesh. Meanwhile, intrinsically, a local intrinsic-preserved loss is introduced to preserve the geodesic priors while avoiding heavy computations. At last, we show the possibility of using IEP-GAN to manipulate 3D human meshes in various ways, including pose transfer, identity swapping and pose interpolation with latent code vector arithmetic. The extensive experiments on various 3D datasets of humans, animals and hands qualitatively and quantitatively demonstrate the generality of our approach. Our proposed model produces better results and is substantially more efficient compared to recent state-of-the-art methods. Code is available: https://github.com/mikecheninoulu/Unsupervised_IEPGAN

While convolutional neural networks have shown a tremendous impact on various computer vision tasks, they generally demonstrate limitations in explicitly modeling long-range dependencies due to the intrinsic locality of the convolution operation. Initially designed for natural language processing tasks, Transformers have emerged as alternative architectures with innate global self-attention mechanisms to capture long-range dependencies. In this paper, we propose TransDepth, an architecture that benefits from both convolutional neural networks and transformers. To avoid the network losing its ability to capture locallevel details due to the adoption of transformers, we propose a novel decoder that employs attention mechanisms based on gates. Notably, this is the first paper that applies transformers to pixel-wise prediction problems involving continuous labels (i.e., monocular depth prediction and surface normal estimation). Extensive experiments demonstrate that the proposed TransDepth achieves state-of-theart performance on three challenging datasets. Our code is available at: https://github.com/ygjwd12345/TransDepth.

Cross-view video synthesis task seeks to generate video sequences of one view from another dramatically different view. In this paper, we investigate the exocentric (third-person) view to egocentric (first-person) view video generation task. This is challenging because egocentric view sometimes is remarkably different from the exocentric view. Thus, transforming the appearances across the two different views is a non-trivial task. Particularly, we propose a novel Bi-directional Spatial Temporal Attention Fusion Generative Adversarial Network (STA-GAN) to learn both spatial and temporal information to generate egocentric video sequences from the exocentric view. The proposed STA-GAN consists of three parts: temporal branch, spatial branch, and attention fusion. First, the temporal and spatial branches generate a sequence of fake frames and their corresponding features. The fake frames are generated in both downstream and upstream directions for both temporal and spatial branches. Next, the generated four different fake frames and their corresponding features (spatial and temporal branches in two directions) are fed into a novel multi-generation attention fusion module to produce the final video sequence. Meanwhile, we also propose a novel temporal and spatial dual-discriminator for more robust network optimization. Extensive experiments on the Side2Ego and Top2Ego datasets show that the proposed STA-GAN significantly outperforms the existing methods.

The major challenge in audio-visual event localization task lies in how to fuse information from multiple modalities effectively. Recent works have shown that the attention mechanism is beneficial to the fusion process. In this paper, we propose a novel joint attention mechanism with multi-modal fusion methods for audio-visual event localization. Particularly, we present a concise yet valid architecture that effectively learns representations from multiple modalities in a joint manner. Initially, visual features are combined with auditory features and then turned into joint representations. Next, we make use of the joint representations to attend to visual features and auditory features, respectively. With the help of this joint co-attention, new visual and auditory features are produced, and thus both features can enjoy the mutually improved benefits from each other. It is worth noting that the joint co-attention unit is recursive meaning that it can be performed multiple times for obtaining better joint representations progressively. Extensive experiments on the public AVE dataset have shown that the proposed method achieves significantly better results than the state-of-the-art methods.

The trade-off between feature representation power and spatial localization accuracy is crucial for the dense classification/semantic segmentation of remote sensing images (RSIs). High-level features extracted from the late layers of a neural network are rich in semantic information, yet have blurred spatial details; low-level features extracted from the early layers of a network contain more pixel-level information but are isolated and noisy. It is therefore difficult to bridge the gap between high- and low-level features due to their difference in terms of physical information content and spatial distribution. In this article, we contribute to solve this problem by enhancing the feature representation in two ways. On the one hand, a patch attention module (PAM) is proposed to enhance the embedding of context information based on a patchwise calculation of local attention. On the other hand, an attention embedding module (AEM) is proposed to enrich the semantic information of low-level features by embedding local focus from high-level features. Both proposed modules are lightweight and can be applied to process the extracted features of convolutional neural networks (CNNs). Experiments show that, by integrating the proposed modules into a baseline fully convolutional network (FCN), the resulting local attention network (LANet) greatly improves the performance over the baseline and outperforms other attention-based methods on two RSI data sets.

In this paper, we address the task of layout-to-image translation, which aims to translate an input semantic layout to a realistic image. One open challenge widely observed in existing methods is the lack of effective semantic constraints during the image translation process, leading to models that cannot preserve the semantic information and ignore the semantic dependencies within the same object. To address this issue, we propose a novel Double Pooling GAN (DPGAN) for generating photo-realistic and semantically-consistent results from the input layout. We also propose a novel Double Pooling Module (DPM), which consists of the Square-shape Pooling Module (SPM) and the Rectangle-shape Pooling Module (RPM). Specifically, SPM aims to capture short-range semantic dependencies of the input layout with different spatial scales, while RPM aims to capture long-range semantic dependencies from both horizontal and vertical directions. We then effectively fuse both outputs of SPM and RPM to further enlarge the receptive field of our generator. Extensive experiments on five popular datasets show that the proposed DPGAN achieves better results than state-of-the-art methods. Finally, both SPM and SPM are general and can be seamlessly integrated into any GAN-based architectures to strengthen the feature representation. The code is available at https://github.com/Ha0Tang/DPGAN.

We present a new deep dictionary learning and coding network (DDLCN) for image-recognition tasks with limited data. The proposed DDLCN has most of the standard deep learning layers (e.g., input/output, pooling, and fully connected), but the fundamental convolutional layers are replaced by our proposed compound dictionary learning and coding layers. The dictionary learning learns an overcomplete dictionary for input training data. At the deep coding layer, a locality constraint is added to guarantee that the activated dictionary bases are close to each other. Then, the activated dictionary atoms are assembled and passed to the compound dictionary learning and coding layers. In this way, the activated atoms in the first layer can be represented by the deeper atoms in the second dictionary. Intuitively, the second dictionary is designed to learn the fine-grained components shared among the input dictionary atoms; thus, a more informative and discriminative low-level representation of the dictionary atoms can be obtained. We empirically compare DDLCN with several leading dictionary learning methods and deep learning models. Experimental results on five popular data sets show that DDLCN achieves competitive results compared with state-of-the-art methods when the training data are limited. Code is available at https://github.com/Ha0Tang/DDLCN.

In this paper, we address the task of semantic-guided scene generation. One open challenge widely observed in global image-level generation methods is the difficulty of generating small objects and detailed local texture. To tackle this issue, in this work we consider learning the scene generation in a local context, and correspondingly design a local class-specific generative network with semantic maps as a guidance, which separately constructs and learns sub-generators concentrating on the generation of different classes, and is able to provide more scene details. To learn more discriminative class-specific feature representations for the local generation, a novel classification module is also proposed. To combine the advantage of both global image-level and the local class-specific generation, a joint generation network is designed with an attention fusion module and a dual-discriminator structure embedded. Extensive experiments on two scene image generation tasks show superior generation performance of the proposed model. State-of-the-art results are established by large margins on both tasks and on challenging public benchmarks. The source code and trained models are available at https://github.com/Ha0Tang/LGGAN.

Cross-view image generation has been recently proposed to generate images of one view from another dramatically different view. In this paper, we investigate exocentric (third-person) view to egocentric (first-person) view image generation. This is a challenging task since egocentric view sometimes is remarkably different from exocentric view. Thus, transforming the appearances across the two views is a nontrivial task. To this end, we propose a novel Parallel Generative Adversarial Network (P-GAN) with a novel cross-cycle loss to learn the shared information for generating egocentric images from exocentric view. We also incorporate a novel contextual feature loss in the learning procedure to capture the contextual information in images. Extensive experiments on the Exo-Ego datasets [1] show that our model outperforms the state-of-the-art approaches.

Since we were babies, we intuitively develop the ability to correlate the input from different cognitive sensors such as vision, audio, and text. However, in machine learning, this cross-modal learning is a nontrivial task because different modalities have no homogeneous properties. Previous works discover that there should be bridges among different modalities. From a neurology and psychology perspective, humans have the capacity to link one modality with another one, e.g., associating a picture of a bird with the only hearing of its singing and vice versa. Is it possible for machine learning algorithms to recover the scene given the audio signal? In this paper, we propose a novel Cascade Attention-Guided Residue GAN (CAR-GAN), aiming at reconstructing the scenes given the corresponding audio signals. Particularly, we present a residue module to mitigate the gap between different modalities progressively. Moreover, a cascade attention guided network with a novel classification loss function is designed to tackle the cross-modal learning task. Our model keeps consistency in the high-level semantic label domain and is able to balance two different modalities. The experimental results demonstrate that our model achieves the state-of-the-art cross-modal audio-visual generation on the challenging Sub-URMP dataset.

In this paper, we focus on the semantic image synthesis task that aims at transferring semantic label maps to photo-realistic images. Existing methods lack effective semantic constraints to preserve the semantic information and ignore the structural correlations in both spatial and channel dimensions, leading to unsatisfactory blurry and artifact-prone results. To address these limitations, we propose a novel Dual Attention GAN (DAGAN) to synthesize photo-realistic and semantically-consistent images with fine details from the input layouts without imposing extra training overhead or modifying the network architectures of existing methods. We also propose two novel modules, i.e., position-wise Spatial Attention Module (SAM) and scale-wise Channel Attention Module (CAM), to capture semantic structure attention in spatial and channel dimensions, respectively. Specifically, SAM selectively correlates the pixels at each position by a spatial attention map, leading to pixels with the same semantic label being related to each other regardless of their spatial distances. Meanwhile, CAM selectively emphasizes the scale-wise features at each channel by a channel attention map, which integrates associated features among all channel maps regardless of their scales. We finally sum the outputs of SAM and CAM to further improve feature representation. Extensive experiments on four challenging datasets show that DAGAN achieves remarkably better results than state-of-the-art methods, while using fewer model parameters.

We address the problem of unsupervised gaze correction in the wild, presenting a solution that works without the need of precise annotations of the gaze angle and the head pose. We created a new dataset called CelebAGaze consisting of two domains X, Y, where the eyes are either staring at the camera or somewhere else. Our method consists of three novel modules: the Gaze Correction module(GCM), the Gaze Animation module(GAM), and the Pretrained Autoencoder module (PAM). Specifically, GCM and GAM separately train a dual in-painting network using data from the domain X for gaze correction and data from the domain Y for gaze animation. Additionally, a Synthesis-As-Training method is proposed when training GAM to encourage the features encoded from the eye region to be correlated with the angle information, resulting in gaze animation achieved by interpolation in the latent space. To further preserve the identity information e.g., eye shape, iris color, we propose the PAM with an Autoencoder, which is based on Self-Supervised mirror learning where the bottleneck features are angle-invariant and which works as an extra input to the dual in-painting models. Extensive experiments validate the effectiveness of the proposed method for gaze correction and gaze animation in the wild and demonstrate the superiority of our approach in producing more compelling results than state-of-the-art baselines. Our code, the pretrained models and supplementary results are available at:https://github.com/zhangqianhui/GazeAnimation.

We propose a unified Generative Adversarial Network (GAN) for controllable image-to-image translation, i.e., transferring an image from a source to a target domain guided by controllable structures. In addition to conditioning on a reference image, we show how the model can generate images conditioned on controllable structures, e.g., class labels, object keypoints, human skeletons, and scene semantic maps. The proposed model consists of a single generator and a discriminator taking a conditional image and the target controllable structure as input. In this way, the conditional image can provide appearance information and the controllable structure can provide the structure information for generating the target result. Moreover, our model learns the image-to-image mapping through three novel losses, i.e., color loss, controllable structure guided cycle-consistency loss, and controllable structure guided self-content preserving loss. Also, we present the Fr´echet ResNet Distance (FRD) to evaluate the quality of the generated images. Experiments on two challenging image translation tasks, i.e., hand gesture-to-gesture translation and cross-view image translation, show that our model generates convincing results, and significantly outperforms other state-of-the-art methods on both tasks. Meanwhile, the proposed framework is a unified solution, thus it can be applied to solving other controllable structure guided image translation tasks such as landmark guided facial expression translation and keypoint guided person image generation. To the best of our knowledge, we are the first to make one GAN framework work on all such controllable structure guided image translation tasks. Code is available at https://github.com/Ha0Tang/GestureGAN.

Cross-view image translation is challenging because it involves images with drastically different views and severe deformation. In this paper, we propose a novel approach named Multi-Channel Attention SelectionGAN (SelectionGAN) that makes it possible to generate images of natural scenes in arbitrary viewpoints, based on an image of the scene and a novel semantic map. The proposed SelectionGAN explicitly utilizes the semantic information and consists of two stages. In the first stage, the condition image and the target semantic map are fed into a cycled semantic-guided generation network to produce initial coarse results. In the second stage, we refine the initial results by using a multi-channel attention selection mechanism. Moreover, uncertainty maps automatically learned from attentions are used to guide the pixel loss for better network optimization. Extensive experiments on Dayton, CVUSA and Ego2Top datasets show that our model is able to generate significantly better results than the state-of-the-art methods. The source code, data and trained models are available at https://github.com/Ha0Tang/SelectionGAN.

Facial attributes are important since they provide a detailed description and determine the visual appearance of human faces. In this paper, we aim at converting a face image to a sketch while simultaneously generating facial attributes. To this end, we propose a novel Attribute-Guided Sketch Generative Adversarial Network (ASGAN) which is an end-to-end framework and contains two pairs of generators and discriminators, one of which is used to generate faces with attributes while the other one is employed for image-to-sketch translation. The two generators form a W-shaped network (W-net) and they are trained jointly with a weight-sharing constraint. Additionally, we also propose two novel discriminators, the residual one focusing on attribute generation and the triplex one helping to generate realistic looking sketches. To validate our model, we have created a new large dataset with 8,804 images, named the Attribute Face Photo & Sketch (AFPS) dataset which is the first dataset containing attributes associated to face sketch images. The experimental results demonstrate that the proposed network (i) generates more photo-realistic faces with sharper facial attributes than baselines and (ii) has good generalization capability on different generative tasks.

In this paper, we focus on the facial expression translation task and propose a novel Expression Conditional GAN (ECGAN) which can learn the mapping from one image domain to another one based on an additional expression attribute. The proposed ECGAN is a generic framework and is applicable to different expression generation tasks where specific facial expression can be easily controlled by the conditional attribute label. Besides, we introduce a novel face mask loss to reduce the influence of background changing. Moreover, we propose an entire framework for facial expression generation and recognition in the wild, which consists of two modules, i.e., generation and recognition. Finally, we evaluate our framework on several public face datasets in which the subjects have different races, illumination, occlusion, pose, color, content and background conditions. Even though these datasets are very diverse, both the qualitative and quantitative results demonstrate that our approach is able to generate facial expressions accurately and robustly.

Multi-view subspace clustering aims to divide a set of multisource data into several groups according to their underlying subspace structure. Although the spectral clustering based methods achieve promotion in multi-view clustering, their utility is limited by the separate learning manner in which affinity matrix construction and cluster indicator estimation are isolated. In this paper, we propose to jointly learn the self-representation, continue and discrete cluster indicators in an unified model. Our model can explore the subspace structure of each view and fusion them to facilitate clustering simultaneously. Experimental results on two benchmark datasets demonstrate that our method outperforms other existing competitive multi-view clustering methods.

The state-of-the-art approaches in Generative Adversarial Networks (GANs) are able to learn a mapping function from one image domain to another with unpaired image data. However, these methods often produce artifacts and can only be able to convert low-level information, but fail to transfer high-level semantic part of images. The reason is mainly that generators do not have the ability to detect the most discriminative semantic part of images, which thus makes the generated images with low-quality. To handle the limitation, in this paper we propose a novel Attention-Guided Generative Adversarial Network (AGGAN), which can detect the most discriminative semantic object and minimize changes of unwanted part for semantic manipulation problems without using extra data and models. The attention-guided generators in AGGAN are able to produce attention masks via a built-in attention mechanism, and then fuse the input image with the attention mask to obtain a target image with high-quality. Moreover, we propose a novel attention-guided discriminator which only considers attended regions. The proposed AGGAN is trained by an end-to-end fashion with an adversarial loss, cycle-consistency loss, pixel loss and attention loss. Both qualitative and quantitative results demonstrate that our approach is effective to generate sharper and more accurate images than existing models.

In this work, we propose a novel Cycle In Cycle Generative Adversarial Network (C2GAN) for the task of keypoint-guided image generation. The proposed C2GAN is a cross-modal framework exploring a joint exploitation of the keypoint and the image data in an interactive manner. C2GAN contains two different types of generators, i.e., keypoint-oriented generator and image-oriented generator. Both of them are mutually connected in an end-to-end learnable fashion and explicitly form three cycled sub-networks, i.e., one image generation cycle and two keypoint generation cycles. Each cycle not only aims at reconstructing the input domain, and also produces useful output involving in the generation of another cycle. By so doing, the cycles constrain each other implicitly, which provides complementary information from the two different modalities and brings extra supervision across cycles, thus facilitating more robust optimization of the whole network. Extensive experimental results on two publicly available datasets, i.e., Radboud Faces and Market-1501, demonstrate that our approach is effective to generate more photo-realistic images compared with state-of-the-art models.

In this paper, we propose a novel Deep Micro-Dictionary Learning and Coding Network (DDLCN). DDLCN has most of the standard deep learning layers (pooling, fully, connected, input/output, etc.) but the main difference is that the fundamental convolutional layers are replaced by novel compound dictionary learning and coding layers. The dictionary learning layer learns an over-complete dictionary for the input training data. At the deep coding layer, a locality constraint is added to guarantee that the activated dictionary bases are close to each other. Next, the activated dictionary atoms are assembled together and passed to the next compound dictionary learning and coding layers. In this way, the activated atoms in the first layer can be represented by the deeper atoms in the second dictionary. Intuitively, the second dictionary is designed to learn the fine-grained components which are shared among the input dictionary atoms. In this way, a more informative and discriminative low-level representation of the dictionary atoms can be obtained. We empirically compare the proposed DDLCN with several dictionary learning methods and deep learning architectures. The experimental results on four popular benchmark datasets demonstrate that the proposed DDLCN achieves competitive results compared with state-of-the-art approaches.

Recent works have shown the benefit of integrating Conditional Random Fields (CRFs) models into deep architectures for improving pixel-level prediction tasks. Following this line of research, in this paper we introduce a novel approach for monocular depth estimation. Similarly to previous works, our method employs a continuous CRF to fuse multi-scale information derived from different layers of a front-end Convolutional Neural Network (CNN). Differently from past works, our approach benefits from a structured attention model which automatically regulates the amount of information transferred between corresponding features at different scales. Importantly, the proposed attention model is seamlessly integrated into the CRF, allowing end-to-end training of the entire architecture. Our extensive experimental evaluation demonstrates the effectiveness of the proposed method which is competitive with previous methods on the KITTI benchmark and outperforms the state of the art on the NYU Depth V2 dataset.

Hand gesture-to-gesture translation in the wild is a challenging task since hand gestures can have arbitrary poses, sizes, locations and self-occlusions. Therefore, this task requires a high-level understanding of the mapping between the input source gesture and the output target gesture. To tackle this problem, we propose a novel hand Gesture Generative Adversarial Network (GestureGAN). GestureGAN consists of a single generator G and a discriminator D, which takes as input a conditional hand image and a target hand skeleton image. GestureGAN utilizes the hand skeleton information explicitly, and learns the gesture-to-gesture mapping through two novel losses, the color loss and the cycle-consistency loss. The proposed color loss handles the issue of "channel pollution" while back-propagating the gradients. In addition, we present the Frechet ResNet Distance (FRD) to evaluate the quality of generated images. Extensive experiments on two widely used benchmark datasets demonstrate that the proposed GestureGAN achieves state-of-the-art performance on the unconstrained hand gesture-to-gesture translation task. Meanwhile, the generated images are in high-quality and are photo-realistic, allowing them to be used as data augmentation to improve the performance of a hand gesture classifier. Our model and code are available at https://github.com/Ha0Tang/GestureGAN.

Depth map has shown promising capability in human action recognition, however it always be auxiliary of RGB features in previous work. As to sufficiently exploring depth map, we propose an innovative descriptor for human action recognition using solo depth data. First, Salient Depth Map (SDM) is calculated between two consecutive depth frames, which is superior for action description as it is located on salient moving objects. Moreover, Binary Shape Map (BSM) is proposed to depict the silhouettes induced by the lateral component of the scene action parallel to the image plane. Then, for implementation, a new framework as Bag-of-Map-Words is employed after concatenating SDM and BSM feature vectors. Experiments on NHA database demonstrate the superiority and high efficiency of the proposed method. We also give detailed comparisons with other features and analysis for parameters as a guidance of further applications.

This paper presents a pioneering study on gender classification from unconstrained back-facing video sequences in natural scenes. In many cases, classifying gender simply via faces or other biometric cues may fail when the video only contains back-facing people. To address this problem, we propose a novel approach to classify the gender according to back-facing video sequences. For this task, a novel Pyramid Segmentation approach is proposed to divide video sequence into a suite of equal time-length sleeves with different scales. Moreover, a heuristic approach is used to compute weights for different features from each sleeve. Finally, a framework of gender classification based on video sequences is presented. To validate our approach, we introduce a new dataset, called BackFacing dataset, featured by 720 annotated back-facing human video sequences. To our knowledge, this is the first dataset only containing back-facing video shots. Experiments demonstrate that the proposed approach achieves competitive results on VidTIMIT, Cohn-Kanade, CASIA Gait and BackFacing datasets.