Layout-to-Image generation has significantly advanced content creation by enabling the rendering of visual text under predefined spatial layouts. Current approaches achieve training-free layout guidance by constructing attention-based energy functions to derive correction gradients. In this paper, we demonstrate that vanilla energy functions suffer from two limitations, resulting in imprecise layout control and visually unrealistic artifacts. First, the normalizing factor of the Boltzmann distribution defined by the energy functions is non-negligible when calculating correction gradients, yet current energy functions cannot compute this factor exactly. Furthermore, while attention varies over time during the denoising process, existing approaches employ a fixed formulation. To address these challenges, we introduce FreLay, a novel training-free approach equipped with a frequency-aware energy function. Our method first reformulates the energy function to handle the normalization factor, enabling accurate computation of correction gradients. Simultaneously, leveraging the prior knowledge that low-frequency information deteriorates slower during noise addition, we design a time-specific energy function for each timestep from a frequency-domain perspective. Experimental results demonstrate that FreLay consistently outperforms existing state-of-the-art training-free methods by a large margin both qualitatively and quantitatively across multiple datasets.

Hallucination in Large Vision-Language Models (LVLMs) remains a critical challenge, undermining their reliability in real-world applications. Existing studies have investigated the causes of hallucination at the modality level and proposed effective strategies. However, interaction patterns beyond the modality level remain insufficiently explored. In this paper, we conduct a token-level analysis and identify two key phenomena: (1) a small subset of textual tokens in LVLMs exert disproportionate influence in the visual-active layers, surpassing that of the visual modality and potentially misleading visual understanding; (2) while LVLMs can correctly identify key visual information, insufficient focus on these cues can sometimes lead to hallucinations. Based on such observation, we attribute hallucinations in LVLMs to two token-level causes: the disproportionate influence of certain textual tokens (phantom tokens) and the underutilization of critical visual cues (anchor tokens). To mitigate these issues, we introduce Token-Asymmetric Filtering (TAF)—a training-free, plug-and-play method that modulates intermediate attention maps in LVLMs. TAF isolates the influence of phantom tokens and emphasizes the influence of anchor tokens in the visual-active layers. Experimental results across multiple benchmarks demonstrate that TAF significantly mitigates hallucinations across a range of state-of-the-art LVLMs.

Diffusion Transformer (DiT)-based video diffusion models generate high-quality videos at scale but incur prohibitive processing latency and memory costs for long videos. To address this, we propose a novel distributed inference strategy, termed DualParal. The core idea is that, instead of generating an entire video on a single GPU, we parallelize computation by partitioning both video frames and model layers across multiple GPUs. However, a naive parallel implementation is not feasible. Because all frames need to share the same noise level, they can't be processed independently. Instead, every step must wait for all others to finish, which cancels out the speed benefits of parallel processing. We overcome this obstacle with a block-wise denoising scheme. Namely, we segment the video into sequential blocks, each with a different noise level. As a result, we process them in a pipeline across the GPUs. Each GPU, holding a subset of the model layers, processes a specific block of frames and passes the results to the next GPU, enabling asynchronous computation and communication. To further optimize performance, we incorporate two key enhancements. Firstly, each GPU uses a feature cache technique to reduce the overhead of smooth transitions by reusing only features involved in cross-frame computation from the prior block, minimizing inter-GPU communication and redundant computation. Secondly, we employ a coordinated noise initialization strategy, ensuring globally consistent temporal dynamics by sharing initial noise patterns across GPUs. Together, these enable fast, artifact-free, and infinitely long video generation. Applied to the latest diffusion transformer video generator, our method efficiently produces 1,025-frame videos with up to 6.54x lower latency and 1.48x lower memory cost on 8xRTX 4090 GPUs.

Blurry video super-resolution (BVSR) remains fundamentally ill-posed due to the simultaneous loss of high-frequency spatial details and reliable motion cues in blurry low-resolution frames. While cascade-based and joint BVSR methods struggle under severe blur, existing event-guided VSR approaches largely assume clean inputs and are ineffective against complex motion degradation. These methods fail to model blurry representations or leverage event signals for blur-aware motion cues, leading to sub-optimal performance. We propose BluR-EVSR, a unified framework that implicitly models Blurry Representations and leverages Event cameras to jointly address both blur and resolution degradation for VSR. The framework begins with a self-supervised degradation learning strategy guided by event streams and neighboring frames, enabling adaptive blur representation without requiring explicit supervision. A dynamic routing mechanism encodes spatially varying degradations, while a motion-saliency degradation-aware attention module injects motion saliency priors to facilitate efficient RGB-event fusion. Integrated into a bidirectional recurrent framework, BluR-EVSR enables temporally consistent and detail-preserving restoration with low computational cost. Extensive experiments across multiple benchmarks show that our method significantly outperforms prior BVSR and event-based approaches.

Event-guided video super-resolution (VSR) leverages high-temporal-resolution event streams to address motion blur, rapid dynamics, and poor illumination that challenge frame-only VSR methods. However, most existing approaches emphasize reconstruction quality while overlooking real-time performance and computational efficiency, limiting their deployment in latency-sensitive scenarios. To overcome these issues, we present E2VSR, a lightweight and Efficient Event-guided VSR framework tailored for real-time applications. Operating under a causal setting with only current and past observations, E2VSR is designed for low-latency event-guided VSR. We propose an event-confidence adaptive propagation strategy comprising two key modules: the Event-induced Feature Modulation (EvFM) block for robust cross-modal event-frame integration, and the Event-Confidence Feature Fusion (EvCFF) block, which exploits events as motion cues for adaptive inter-frame aggregation. This design improves motion-aware temporal aggregation in challenging dynamic conditions, where event cues may provide complementary temporal information. Furthermore, an Implicit Event Reconstruction (IER) technique leverages event information during training to enrich feature representations without adding inference-time cost, enhancing spatial and temporal fidelity. Experimental results demonstrate that E2VSR achieves superior quantitative and qualitative performance while maintaining a low parameter count and computational cost.

Rendering dynamic scenes from monocular videos is a crucial yet challenging task. The recent deformable Gaussian Splatting has emerged as a robust solution to represent real-world dynamic scenes. However, it often leads to heavily redundant Gaussians, attempting to fit every training view at various time steps, leading to slower rendering speeds. Additionally, the attributes of Gaussians in static areas are time-invariant, making it unnecessary to model every Gaussian, which can cause jittering in static regions. In practice, the primary bottleneck in rendering speed for dynamic scenes is the number of Gaussians. In response, we introduce Efficient Dynamic Gaussian Splatting (EDGS), which represents dynamic scenes via sparse time-variant attribute modeling. Our approach formulates dynamic scenes using a sparse anchor-grid representation, with the motion flow of dense Gaussians calculated via a classical kernel representation. Furthermore, we propose an unsupervised strategy to efficiently filter out anchors corresponding to static areas. Only anchors associated with deformable objects are input into MLPs to query time-variant attributes. Experiments on two real-world datasets demonstrate that our EDGS significantly improves the rendering speed with superior rendering quality compared to previous state-of-the-art methods.

Recovering 4D world from monocular video is a crucial yet challenging task. Conventional methods usually rely on the assumptions of multi-view videos, known camera parameters, or static scenes. In this paper, we relax all these constraints and tackle a highly ambitious but practical task: With only one monocular video without camera parameters, we aim to recover the dynamic 3D world alongside the camera poses. To solve this, we introduce GFlow, a new framework that utilizes only 2D priors (depth and optical flow) to lift a video to a 4D scene, as a flow of 3D Gaussians through space and time. GFlow starts by segmenting the video into still and moving parts, then alternates between optimizing camera poses and the dynamics of the 3D Gaussian points. This method ensures consistency among adjacent points and smooth transitions between frames. Since dynamic scenes always continually introduce new visual content, we present prior-driven initialization and pixel-wise densification strategy for Gaussian points to integrate new content. By combining all those techniques, GFlow transcends the boundaries of 4D recovery from causal videos; it naturally enables tracking of points and segmentation of moving objects across frames. Additionally, GFlow estimates the camera poses for each frame, enabling novel view synthesis by changing camera pose. This capability facilitates extensive scene-level or object-level editing, highlighting GFlow's versatility and effectiveness.

Graph Neural Networks (GNNs) have become the preferred tool to process graph data, with their efficacy being boosted through graph data augmentation techniques. Despite the evolution of augmentation methods, issues like graph property distortions and restricted structural changes persist. This leads to the question: Is it possible to develop more property-conserving and structure-sensitive augmentation methods? Through a spectral lens, we investigate the interplay between graph properties, their augmentation, and their spectral behavior, and found that keeping the low-frequency eigenvalues unchanged can preserve the critical properties at a large scale when generating augmented graphs. These observations inform our introduction of the Dual-Prism (DP) augmentation method, comprising DP-Noise and DP-Mask, which adeptly retains essential graph properties while diversifying augmented graphs. Extensive experiments validate the efficiency of our approach, providing a new and promising direction for graph data augmentation.

As climate change reshapes global weather patterns, the increasing frequency and intensity of extreme rainfall events have amplified the safety imperatives for autonomous driving systems. During such events, rainfall can escalate from heavy to violent, as defined by the World Meteorological Organization, severely impairing images with diverse and significant degradations. Many existing semantic segmentation models perform well under light to heavy rain, but there is a notable absence of datasets addressing violent rain conditions for these models to validate and learn from. In this paper, we introduce the Extreme RainFall (ERF) dataset for semantic segmentation in both image and video tasks under violent rain conditions. Our dataset comprises 14,757 unlabeled frames and 100 labeled frames, all captured during four different violent rainfall periods. We use our dataset to evaluate the robustness of various methods against violent rainfall, focusing on four approaches: 1) image-based foundation models, 2) image-based domain generalization methods, 3) image-based domain adaptation methods, and 4) video-based methods. The results reveal that none of the existing models tested is capable of withstanding the extreme challenges posed by violent rainfall conditions. By analyzing the results, we offer insights and suggestions for developing more robust models under extreme rainfall events.

Chain-of-Thought significantly enhances a model’s reasoning capability, but it also comes with a considerable increase in inference costs due to long chains. With the observation that the reasoning path can be easily compressed under easy tasks but struggle on hard tasks, we explore the feasibility of elastically controlling the length of reasoning paths with only one model, thereby reducing the inference overhead of reasoning models dynamically based on task difficulty. We introduce a new tuning and inference strategy named CoT-Valve, designed to allow models to generate reasoning chains of varying lengths. To achieve this, we propose to identify a direction in the parameter space that, when manipulated, can effectively control the length of generated CoT. Moreover, we show that this property is valuable for compressing the reasoning chain. We construct datasets with chains from long to short for the same questions and explore two enhanced strategies for CoT-Valve: (1) a precise length-compressible CoT tuning method, and (2) a progressive chain length compression approach. Our experiments show that CoT-Valve successfully enables controllability and compressibility of the chain and shows better performance than the prompt-based control. We applied this method to QwQ-32B-Preview, reducing reasoning chains on GSM8K from 741 to 225 tokens with a minor performance drop (95.07% to 94.92%) and on AIME from 6827 to 4629 tokens, with only one additional incorrect answer.

Mamba, an architecture with RNN-like token mixer of state space model (SSM), was recently introduced to address the quadratic complexity of the attention mechanism and subsequently applied to vision tasks. Nevertheless, the performance of Mamba for vision is often underwhelming when compared with convolutional and attention-based models. In this paper, we delve into the essence of Mamba, and conceptually conclude that Mamba is ideally suited for tasks with long-sequence and autoregressive characteristics. For vision tasks, as image classification on ImageNet does not align with either characteristic, we hypothesize that Mamba is not necessary for this task; Detection and segmentation tasks on COCO or ADE20K are also not autoregressive, yet they adhere to the long-sequence characteristic, so we believe it is still worthwhile to explore Mamba’s potential for these tasks. To empirically verify our hypotheses, we construct a series of models named MambaOut through stacking Mamba blocks while removing their core token mixer, SSM. Experimental results strongly support our hypotheses. Specifically, our MambaOut model surpasses all visual Mamba models on ImageNet image classification, indicating that Mamba is indeed unnecessary for this task. As for detection and segmentation, MambaOut cannot match the performance of state-of-the-art visual Mamba models, demonstrating the potential of Mamba for long-sequence visual tasks.

Poster design is a critical medium for visual communication. Prior work has explored automatic poster design using deep learning techniques, but these approaches lack text accuracy, user customization, and aesthetic appeal, limiting their applicability in artistic domains such as movies and exhibitions, where both clear content delivery and visual impact are essential. To address these limitations, we present POSTA: a modular framework powered by diffusion models and multimodal large language models (MLLMs) for customized artistic poster generation. The framework consists of three modules. Background Diffusion creates a themed background based on user input. Design MLLM then generates layout and typography elements that align with and complement the background style. Finally, to enhance the poster’s aesthetic appeal, ArtText Diffusion applies additional stylization to key text elements. The final result is a visually cohesive and appealing poster, with a fully modular process that allows for complete customization. To train our models, we develop the PosterArt dataset, comprising high-quality artistic posters annotated with layout, typography, and pixel-level stylized text segmentation. Our comprehensive experimental analysis demonstrates POSTA’s exceptional controllability and design diversity, outperforming existing models in both text accuracy and aesthetic quality.

In the rapidly advancing field of image generation, Visual Auto-Regressive (VAR) modeling has garnered considerable attention for its innovative next-scale prediction approach. This paradigm offers substantial improvements in efficiency, scalability, and zero-shot generalization. Yet, the inherently coarse-to-fine nature of VAR introduces a prolonged token sequence, leading to prohibitive memory consumption and computational redundancies. To address these bottlenecks, we propose Collaborative Decoding (CoDe), a novel efficient decoding strategy tailored for the VAR framework. CoDe capitalizes on two critical observations: the substantially reduced parameter demands at larger scales and the exclusive generation patterns across different scales. Based on these insights, we partition the multi-scale inference process into a seamless collaboration between a large model and a small model. The large model serves as the ’drafter’, specializing in generating low-frequency content at smaller scales, while the smaller model serves as the ’refiner’, solely focusing on predicting high-frequency details at larger scales. This collaboration yields remarkable efficiency with minimal impact on quality: CoDe achieves a 1.7x speedup, slashes memory usage by around 50%, and preserves image quality with only a negligible FID increase from 1.95 to 1.98. When drafting steps are further decreased, CoDe can achieve an impressive 2.9x acceleration ratio, reaching 41 images/s at 256×256 resolution on a single NVIDIA 4090 GPU, while preserving a commendable FID of 2.27.

Diffusion Transformers have demonstrated remarkable capabilities in image generation but often come with excessive parameterization, resulting in considerable inference overhead in real-world applications. In this work, we present TinyFusion, a depth pruning method designed to remove redundant layers from diffusion transformers via end-to-end learning. The core principle of our approach is to create a pruned model with high recoverability, allowing it to regain strong performance after fine-tuning. To accomplish this, we introduce a differentiable sampling technique to make pruning learnable, paired with a co-optimized parameter to simulate future fine-tuning. While prior works focus on minimizing loss or error after pruning, our method explicitly models and optimizes the post-fine-tuning performance of pruned models. Experimental results indicate that this learnable paradigm offers substantial benefits for layer pruning of diffusion transformers, surpassing existing importance-based and error-based methods. Additionally, TinyFusion exhibits strong generalization across diverse architectures, such as DiTs, MARs, and SiTs. Experiments with DiT-XL show that TinyFusion can craft a shallow diffusion transformer at less than 7% of the pre-training cost, achieving a 2× speedup with an FID score of 2.86, outperforming competitors with comparable efficiency. Code is available at https://github.com/VainF/TinyFusion

Implicit Neural Representations (INRs) have emerged as a powerful framework for representing continuous signals. However, generating diverse INR weights remains challenging due to limited training data. We introduce Few-Shot Implicit Function Generation, a new problem setup that aims to generate diverse yet functionally consistent INR weights from only a few examples. This is challenging because even for the same signal, the optimal INRs can vary significantly depending on their initializations. To tackle this, we propose EquiGen, a framework that can generate new INRs from limited data. The core idea is that functionally similar networks can be transformed into one another through weight permutations, forming an equivariance group. By projecting these weights into an equivariant latent space, we enable diverse generation within these groups, even with few examples. EquiGen implements this through an equivariant encoder trained via contrastive learning and smooth augmentation, an equivariance-guided diffusion process, and controlled perturbations in the equivariant subspace. Experiments on 2D image and 3D shape INR datasets demonstrate that our approach effectively generates diverse INR weights while preserving their functional properties in few-shot scenarios. Code is available at https://github.com/JeanDiable/EquiGen.

Novel view synthesis from limited observations remains a significant challenge due to the lack of information in under-sampled regions, often resulting in noticeable artifacts. We introduce Generative Sparse-View Gaussian Splatting (GSGS), a general pipeline designed to enhance the rendering quality of 3D/4D Gaussian Splatting (GS) when training views are sparse. Our method generates unseen views using generative models, specifically leveraging pre-trained image diffusion models to iteratively refine view consistency and hallucinate additional images at pseudo views. This approach improves 3D/4D scene reconstruction by explicitly enforcing semantic correspondences during the generation of unseen views, thereby enhancing geometric consistency-unlike purely generative methods that often fail to maintain view consistency. Extensive evaluations on various 3D/4D datasets—including Blender, LLFF, Mip-NeRF360, and Neural 3D Video-Demonstrate that our GS-GS outperforms existing state-of-the-art methods in rendering quality without sacrificing efficiency.

Generating dense multiview images from text prompts is crucial for creating high-fidelity 3D assets. Nevertheless, existing methods struggle with space-view correspondences, resulting in sparse and low-quality outputs. In this paper, we introduce CoSER, a novel consistent dense Multiview Text-To-Image Generator for Text-To-3D, achieving both efficiency and quality by meticulously learning neighbor-view coherence and further alleviating ambiguity through the swift traversal of all views. For achieving neighbor-view consistency, each viewpoint densely interacts with adjacent viewpoints to perceive the global spatial structure, and aggregates information along motion paths explicitly defined by physical principles to refine details. To further enhance cross-view consistency and alleviate content drift, CoSER rapidly scan all views in spiral bidirectional manner to aware holistic information and then scores each point based on semantic material. Subsequently, we conduct weighted down-sampling along the spatial dimension based on scores, thereby facilitating prominent information fusion across all views with lightweight computation. Technically, the core module is built by integrating the attention mechanism with a selective state space model, exploiting the robust learning capabilities of the former and the low overhead of the latter. Extensive evaluation shows that CoSER is capable of producing dense, high-fidelity, content-consistent multiview images that can be flexibly integrated into various 3D generation models.

In this paper, we introduce a novel self-distillation paradigm for improving the performance of diffusion models. Previous studies have shown that introducing a teacher to distill the diffusion model can enhance its sampling efficiency. We raise an intriguing question: can the diffusion model itself serve as its teacher to further improve the performance of itself? To this end, we propose a new paradigm called Self Step-Distillation (SSD). The core idea of SSD is to integrate the predictions or the intermediate activations of the diffusion model at each timestep with its preceding timestep through a fusion mechanism. We propose two forms, explicit SSD and implicit SSD (iSSD), to perform N-step to N-step distillation from the diffusion model itself to achieve improved image quality. We further elucidate the relationship between SSD and high-order solver, highlighting their underlying relationship. The effectiveness of SSD is validated through extensive experiments on diffusion transformers of various sizes and across different sampling steps. Our results show that this novel self-distillation paradigm can significantly enhance performance. Additionally, our method is compatible with the distillation method designed for few-step inference. Notably, with iSSD trained less than one epoch, we obtain a 32-step DiT-XL/2 achieving an FID of 1.99, outperforming the original 250-step DiT-XL/2 with an FID of 2.26. We further validate the effectiveness of our method on text-to-image diffusion models, such as Stable Diffusion, and also observe notable improvement in image quality.

Self-supervised representation learning for point cloud has demonstrated effectiveness in improving pre-trained model performance across diverse tasks. However, as pre-trained models grow in complexity, fully fine-tuning them for downstream applications demands substantial computational and storage resources. Parameter-efficient fine-tuning (PEFT) methods offer a promising solution to mitigate these resource requirements, yet most current approaches rely on complex adapter and prompt mechanisms that increase tunable parameters. In this paper, we propose PointLoRA, a simple yet effective method that combines low-rank adaptation (LoRA) with multi-scale token selection to efficiently fine-tune point cloud models. Our approach embeds LoRA layers within the most parameter-intensive components of point cloud transformers, reducing the need for tunable parameters while enhancing global feature capture. Additionally, multi-scale token selection extracts critical local information to serve as prompts for downstream fine-tuning, effectively complementing the global context captured by LoRA. The experimental results across various pre-trained models and three challenging public datasets demonstrate that our approach achieves competitive performance with only 3.43% of the trainable parameters, making it highly effective for resource-constrained applications. Source code is available at: https://github.com/songw-zju/PointLoRA.

Exploiting temporal correlations is crucial for video super-resolution (VSR). Recent approaches enhance this by incorporating event cameras. In this paper, we introduce MamEVSR, a Mamba-based network for event-based VSR that leverages the selective state space model, Mamba. MamEVSR stands out by offering global receptive field coverage with linear computational complexity, thus addressing the limitations of convolutional neural networks and Transformers. The key components of MamEVSR include: (1) The interleaved Mamba (iMamba) block, which interleaves tokens from adjacent frames and applies multidirectional selective state space modeling, enabling efficient feature fusion and propagation across bi-directional frames while maintaining linear complexity. (2) The cross-modality Mamba (cMamba) block facilitates further interaction and aggregation between event information and the output from the iMamba block. The cMamba block can leverage complementary spatio-temporal information from both modalities and allows MamEVSR to capture finer motion details. Experimental results show that the proposed MamEVSR achieves superior performance on various datasets quantitatively and qualitatively.

The quality of 3D generative modeling has been notably improved by the adoption of 2D diffusion models. Despite this progress, the cumbersome optimization process per se presents a critical problem to efficiency. In this paper, we introduce Hash3D, a universal acceleration for 3D score distillation sampling (SDS) without model training. Central to Hash3D is the observation that images rendered from similar camera positions and diffusion time-steps often have redundant feature maps. By hashing and reusing these feature maps across nearby timesteps and camera angles, Hash3D eliminates unnecessary calculations. We implement this through an adaptive grid-based hashing. As a result, it largely speeds up the process of 3D generation. Surprisingly, this feature-sharing mechanism not only makes generation faster but also improves the smoothness and view consistency of the synthesized 3D objects. Our experiments covering 5 text-to-3D and 3 image-to-3D models, demonstrate Hash3D’s versatility to speed up optimization, enhancing efficiency by 1.5 ~ 4×. Additionally, Hash3D’s integration with 3D Gaussian splatting largely speeds up 3D model creation, reducing text-to-3D conversion to about 10 minutes and image-to-3D conversion to 30 seconds. The project page is https://adamdad.github.io/hash3D/.

We present OminiControl, a novel approach that rethinks how image conditions are integrated into Diffusion Transformer (DiT) architectures. Current image conditioning methods either introduce substantial parameter overhead or handle only specific control tasks effectively, limiting their practical versatility. OminiControl addresses these limitations through three key innovations: (1) a minimal architectural design that leverages the DiT's own VAE encoder and transformer blocks, requiring just 0.1% additional parameters; (2) a unified sequence processing strategy that combines condition tokens with image tokens for flexible token interactions; and (3) a dynamic position encoding mechanism that adapts to both spatially-aligned and non-aligned control tasks. Our extensive experiments show that this streamlined approach not only matches but surpasses the performance of specialized methods across multiple conditioning tasks. To overcome data limitations in subject-driven generation, we also introduce Subjects200K, a large-scale dataset of identity-consistent image pairs synthesized using DiT models themselves. This work demonstrates that effective image control can be achieved without architectural complexity, opening new possibilities for efficient and versatile image generation systems.

While Text-to-Image (T2I) diffusion models excel at generating visually appealing images of individual instances, they struggle to accurately position and control the features generation of multiple instances. The Layout-to-Image (L2I) task was introduced to address the positioning challenges by incorporating bounding boxes as spatial control signals, but it still falls short in generating precise instance features. In response, we propose the Instance Feature Generation (IFG) task, which aims to ensure both positional accuracy and feature fidelity in generated instances. To address the IFG task, we introduce the Instance Feature Adapter (IFAdapter). The IFAdapter enhances feature depiction by incorporating additional appearance tokens and utilizing an Instance Semantic Map to align instance-level features with spatial locations. The IFAdapter guides the diffusion process as a plug-and-play module, making it adaptable to various community models. For evaluation, we contribute an IFG benchmark and develop a verification pipeline to objectively compare models' abilities to generate instances with accurate positioning and features. Experimental results demonstrate that IFAdapter outperforms other models in both quantitative and qualitative evaluations.

Dataset distillation or condensation aims to condense a large-scale training dataset into a much smaller synthetic one such that the training performance of distilled and original sets on neural networks are similar. Although the number of training samples can be reduced substantially, current state-of-the-art methods heavily rely on enormous soft labels to achieve satisfactory performance. As a result, the required storage can be comparable even to original datasets, especially for large-scale ones. To solve this problem, instead of storing these heavy labels, we propose a novel label-lightening framework termed HeLlO aiming at effective image-to-label projectors, with which synthetic labels can be directly generated online from synthetic images. Specifically, to construct such projectors, we leverage prior knowledge in open-source foundation models, e.g., CLIP, and introduce a LoRA-like fine-tuning strategy to mitigate the gap between pre-trained and target distributions, so that original models for soft-label generation can be distilled into a group of low-rank matrices. Moreover, an effective image optimization method is proposed to further mitigate the potential error between the original and distilled label generators. Extensive experiments demonstrate that with only about 0.003% of the original storage required for a complete set of soft labels, we achieve comparable performance to current state-of-the-art dataset distillation methods on large-scale datasets. Our code will be available.

Graph neural networks (GNNs) are conventionally trained on a per-domain, per-task basis. It creates a significant barrier in transferring the acquired knowledge to different, heterogeneous data setups. This paper introduces **GraphBridge**, a novel framework to enable knowledge transfer across disparate tasks and domains in GNNs, circumventing the need for modifications to task configurations or graph structures. Specifically, GraphBridge allows for the augmentation of any pre-trained GNN with prediction heads and a bridging network that connects the input to the output layer. This architecture not only preserves the intrinsic knowledge of the original model but also supports outputs of arbitrary dimensions. To mitigate the negative transfer problem, GraphBridge merges the source model with a concurrently trained model, thereby reducing the source bias when applied to the target domain. Our method is thoroughly evaluated across diverse transfer learning scenarios, including Graph2Graph, Node2Node, Graph2Node, and graph2point-cloud. Empirical validation, conducted over 16 datasets representative of these scenarios, confirms the framework's capacity for task- and domain-agnostic transfer learning within graph-like data, marking a significant advancement in the field of GNNs. Code is available at https://github.com/jujulili888/GraphBridge.

3D Gaussian splatting (3DGS), known for its groundbreaking performance and efficiency, has become a dominant 3D representation and brought progress to many 3D vision tasks. However, in this work, we reveal a significant security vulnerability that has been largely overlooked in 3DGS: the computation cost of training 3DGS could be maliciously tampered by poisoning the input data. By developing an attack named Poison-splat, we reveal a novel attack surface where the adversary can poison the input images to drastically increase the computation memory and time needed for 3DGS training, pushing the algorithm towards its worst computation complexity. In extreme cases, the attack can even consume all allocable memory, leading to a Denial-of-Service (DoS) that disrupts servers, resulting in practical damages to real-world 3DGS service vendors. Such a computation cost attack is achieved by addressing a bi-level optimization problem through three tailored strategies: attack objective approximation, proxy model rendering, and optional constrained optimization. These strategies not only ensure the effectiveness of our attack but also make it difficult to defend with simple defensive measures. We hope the revelation of this novel attack surface can spark attention to this crucial yet overlooked vulnerability of 3DGS systems. Our code is available at https://github.com/jiahaolu97/poison-splat .

The "pre-train, prompt-tuning'' paradigm has demonstrated impressive performance for tuning pre-trained heterogeneous graph neural networks (HGNNs) by mitigating the gap between pre-trained models and downstream tasks. However, most prompt-tuning-based works may face at least two limitations: (i) the model may be insufficient to fit the graph structures well as they are generally ignored in the prompt-tuning stage, increasing the training error to decrease the generalization ability; and (ii) the model may suffer from the limited labeled data during the prompt-tuning stage, leading to a large generalization gap between the training error and the test error to further affect the model generalization. To alleviate the above limitations, we first derive the generalization error bound for existing prompt-tuning-based methods, and then propose a unified framework that combines two new adapters with potential labeled data extension to improve the generalization of pre-trained HGNN models. Specifically, we design dual structure-aware adapters to adaptively fit task-related homogeneous and heterogeneous structural information. We further design a label-propagated contrastive loss and two self-supervised losses to optimize dual adapters and incorporate unlabeled nodes as potential labeled data. Theoretical analysis indicates that the proposed method achieves a lower generalization error bound than existing methods, thus obtaining superior generalization ability. Comprehensive experiments demonstrate the effectiveness and generalization of the proposed method on different downstream tasks.

Transformers stand as the cornerstone of mordern deep learning. Traditionally, these models rely on multi-layer perceptron (MLP) layers to mix the information between channels. In this paper, we introduce the Kolmogorov–Arnold Transformer (KAT), a novel architecture that replaces MLP layers with Kolmogorov-Arnold Network (KAN) layers to enhance the expressiveness and performance of the model. Integrating KANs into transformers, however, is no easy feat, especially when scaled up. Specifically, we identify three key challenges: (C1) Base function. The standard B-spline function used in KANs is not optimized for parallel computing on modern hardware, resulting in slower inference speeds. (C2) Parameter and Computation Inefficiency. KAN requires a unique function for each input-output pair, making the computation extremely large. (C3) Weight initialization. The initialization of weights in KANs is particularly challenging due to their learnable activation functions, which are critical for achieving convergence in deep neural networks. To overcome the aforementioned challenges, we propose three key solutions: (S1) Rational basis. We replace B-spline functions with rational functions to improve compatibility with modern GPUs. By implementing this in CUDA, we achieve faster computations. (S2) Group KAN. We share the activation weights through a group of neurons, to reduce the computational load without sacrificing performance. (S3) Variance-preserving initialization. We carefully initialize the activation weights to make sure that the activation variance is maintained across layers. With these designs, KAT scales effectively and readily outperforms traditional MLP-based transformers. We demonstrate the advantages of KAT across various tasks, including image recognition, object detection, and semantic segmentation. It consistently enhances performance over the standard transformer architectures of different model sizes.

Given a collection of neural networks, can we determine which are parent models and which are child models fine-tuned from the parents? In this work, we strive to answer this question via introducing a new task termed as neural phylogeny detection, aimed at identifying the existence and direction of the fine-tuning relationship. Specifically, neural phylogeny detection attempts to identify all parent-child model pairs and determine, within each pair, which model is the parent and which is the child. We present two approaches for neural phylogeny detection: a learning-free method and a learning-based method. First, we propose a metric that leverages the distance from network parameters to a fake initialization to infer fine-tuning directions. By integrating this metric with traditional clustering algorithms, we propose a series of efficient, learning-free neural phylogeny detection methods. Second, we introduce a transformer-based neural phylogeny detector, which significantly enhances detection accuracy through a learning-based manner. Extensive experiments, ranging from shallow fully-connected networks to open-sourced Stable Diffusion and LLaMA models, progressively validate the effectiveness of both methods. The results demonstrate the reliability of both the learning-free and the learning-based approaches across various learning tasks and network architectures, as well as their ability to detect cross-generational phylogeny between ancestor models and their fine-tuned descendants.

This study introduces dataset distillation (DD) tailored for 3D data, particularly point clouds. DD aims to substitute large-scale real datasets with a small set of synthetic samples while preserving model performance. Existing methods mainly focus on structured data such as images. However, adapting DD for unstructured point clouds poses challenges due to their diverse orientations and resolutions in 3D space. To address these challenges, we theoretically demonstrate the importance of matching rotation-invariant features between real and synthetic data for 3D distillation. We further propose a plug-and-play point cloud rotator to align the point cloud to a canonical orientation, facilitating the learning of rotation-invariant features by all point cloud models. Furthermore, instead of optimizing fixed-size synthetic data directly, we devise a point-wise generator to produce point clouds at various resolutions based on the sampled noise amount. Compared to conventional DD methods, the proposed approach, termed DD3D, enables efficient training on low-resolution point clouds while generating high-resolution data for evaluation, thereby significantly reducing memory requirements and enhancing model scalability. Extensive experiments validate the effectiveness of DD3D in shape classification and part segmentation tasks across diverse scenarios, such as cross-architecture and cross-resolution settings.

Layout-to-Image generation aims to create complex scenes with precise control over the placement and arrangement of subjects. Existing works have demonstrated that pre-trained Text-to-Image diffusion models can achieve this goal without training on any specific data; however, they often face challenges with imprecise localization and unrealistic artifacts. Focusing on these drawbacks, we propose a novel training-free method, WinWinLay. At its core, WinWinLay presents two key strategies—Non-local Attention Energy Function and Adaptive Update—that collaboratively enhance control precision and realism. On one hand, we theoretically demonstrate that the commonly used attention energy function introduces inherent spatial distribution biases, hindering objects from being uniformly aligned with layout instructions. To overcome this issue, non-local attention prior is explored to redistribute attention scores, facilitating objects to better conform to the specified spatial conditions. On the other hand, we identify that the vanilla backpropagation update rule can cause deviations from the pre-trained domain, leading to out-of-distribution artifacts. We accordingly introduce a Langevin dynamics-based adaptive update scheme as a remedy that promotes in-domain updating while respecting layout constraints. Extensive experiments demonstrate that WinWinLay excels in controlling element placement and achieving photorealistic visual fidelity, outperforming the current state-of-the-art methods.

Text-to-image diffusion models have achieved remarkable progress in recent years. However, training models for high-resolution image generation remains challenging, particularly when training data and computational resources are limited. In this paper, we explore this practical problem from two key perspectives: data and parameter efficiency, and propose a set of key guidelines for ultra-resolution adaptation termed URAE. For data efficiency, we theoretically and empirically demonstrate that synthetic data generated by some teacher models can significantly promote training convergence. For parameter efficiency, we find that tuning minor components of the weight matrices outperforms widely-used low-rank adapters when synthetic data are unavailable, offering substantial performance gains while maintaining efficiency. Additionally, for models leveraging guidance distillation, such as FLUX, we show that disabling classifier-free guidance, i.e., setting the guidance scale to 1 during adaptation, is crucial for satisfactory performance. Extensive experiments validate that URAE achieves comparable 2K-generation performance to state-of-the-art closed-source models like FLUX1.1 [Pro] Ultra with only 3K samples and 2K iterations, while setting new benchmarks for 4K-resolution generation. Codes are available [here](https://github.com/Huage001/URAE).

Large Reasoning Models (LRMs) have garnered considerable attention for their ability to tackle complex tasks through the Chain-of-Thought (CoT) approach. However, their tendency toward overthinking results in unnecessarily lengthy reasoning chains, dramatically increasing the inference costs. To mitigate this issue, we introduce VeriThinker, a novel approach for CoT compression. Unlike conventional methods that fine-tune LRMs directly on the original reasoning task using synthetic concise CoT data, we innovatively fine-tune the model solely through an auxiliary verification task. By training LRMs to accurately verify the correctness of CoT solutions, the LRMs inherently become more discerning about the necessity of subsequent self-reflection steps, thereby effectively suppressing overthinking. Extensive experiments validate that VeriThinker substantially reduces reasoning chain lengths while maintaining or even slightly improving accuracy. When applied to DeepSeek-R1-Distill-Qwen-7B, our approach reduces reasoning tokens on MATH500 from 3790 to 2125 while improving accuracy by 0.8% (94.0% to 94.8%), and on AIME25, tokens decrease from 14321 to 10287 with a 2.1% accuracy gain (38.7% to 40.8%). Additionally, our experiments demonstrate that VeriThinker can also be zero-shot generalized to speculative reasoning.

Reasoning Language Models, capable of extended chain-of-thought reasoning, have demonstrated remarkable performance on tasks requiring complex logical inference. However, applying elaborate reasoning for all queries often results in substantial computational inefficiencies, particularly when many problems admit straightforward solutions. This motivates an open question: Can LLMs learn when to think? To answer this, we propose Thinkless, a learnable framework that empowers an LLM to adaptively select between short-form and long-form reasoning, based on both task complexity and the model's ability. Thinkless is trained under a reinforcement learning paradigm and employs two control tokens, \<short\> for concise responses and \<think\> for detailed reasoning. At the core of our method is a Decoupled Group Relative Policy Optimization (DeGRPO) algorithm, which decomposes the learning objective of hybrid reasoning into two components: (1) a control token loss that governs the selection of the reasoning mode, and (2) a response loss that improves the accuracy of the generated answers. This decoupled formulation enables fine-grained control over the contributions of each objective, stabilizing training and effectively preventing the collapse observed in vanilla GRPO. Empirically, on several benchmarks such as Minerva Algebra, MATH-500, and GSM8K, Thinkless is able to reduce the usage of long-chain thinking by 50% - 90%, significantly improving the efficiency of Reasoning Language Models. The code is available at \url{https://github.com/VainF/Thinkless}

While diffusion models have achieved remarkable success in text-to-image generation, they encounter significant challenges with instruction-driven image editing. Our research highlights a key challenge: these models particularly struggle with structurally-inconsistent edits that involve substantial layout changes. To address this gap, we introduce Image Editing As Programs (IEAP), a unified image editing framework built upon the Diffusion Transformer (DiT) architecture. Specifically, IEAP deals with complex instructions by decomposing them into a sequence of programmable atomic operations. Each atomic operation manages a specific type of structurally consistent edit; when sequentially combined, IEAP enables the execution of arbitrary and structurally-inconsistent transformations. This reductionist approach enables IEAP to robustly handle a wide spectrum of edits, encompassing both structurally-consistent and inconsistent changes. Extensive experiments demonstrate that IEAP significantly outperforms state-of-the-art methods on standard benchmarks across various editing scenarios. In these evaluations, our framework delivers superior accuracy and semantic fidelity, particularly for complex, multi-step instructions. Codes are available at https://github.com/YujiaHu1109/IEAP.

Multimodal large language models (MLLMs) demonstrates remarkable capabilities in handling complex multimodal tasks and are increasingly adopted in video understanding applications. However, their rapid advancement raises serious data privacy concerns, particularly given the potential inclusion of sensitive video content, such as personal recordings and surveillance footage, in their training datasets. Determining improperly used videos during training remains a critical and unresolved challenge. Despite considerable progress on membership inference attacks (MIAs) for text and image data in MLLMs, existing methods fail to generalize effectively to the video domain. These methods suffer from poor scalability as more frames are sampled and generally achieve negligible true positive rates at low false positive rates (TPR@Low FPR), mainly due to their failure to capture the inherent temporal variations of video frames and to account for model behavior differences as the number of frames varies. To address these challenges, we introduce Vid-SME (**Vid**eo **S**harma–**M**ittal **E**ntropy), the first membership inference method tailored for video data used in video understanding LLMs (VULLMs). Vid-SME leverages the confidence of model output and integrates adaptive parameterization to compute Sharma–Mittal entropy (SME) for video inputs. By leveraging the SME difference between natural and temporally-reversed video frames, Vid-SME derives robust membership scores to determine whether a given video is part of the model's training set. Experiments on various self-trained and open-sourced VULLMs demonstrate the strong effectiveness of Vid-SME.

Diffusion Transformers (DiT) have become a leading architecture in image generation. However, the quadratic complexity of attention mechanisms, which are responsible for modeling token-wise relationships, results in significant latency when generating high-resolution images. To address this issue, we aim at a linear attention mechanism in this paper that reduces the complexity of pre-trained DiTs to linear. We begin our exploration with a comprehensive summary of existing efficient attention mechanisms and identify four key factors crucial for successful linearization of pre-trained DiTs: locality, formulation consistency, high-rank attention maps, and feature integrity. Based on these insights, we introduce a convolution-like local attention strategy termed CLEAR, which limits feature interactions to a local window around each query token, and thus achieves linear complexity. Our experiments indicate that, by fine-tuning the attention layer on merely 10K self-generated samples for 10K iterations, we can effectively transfer knowledge from a pre-trained DiT to a student model with linear complexity, yielding results comparable to the teacher model. Simultaneously, it reduces attention computations by 99.5% and accelerates generation by 6.3 times for generating 8K-resolution images. Furthermore, we investigate favorable properties in the distilled attention layers, such as zero-shot generalization across various models and plugins, and improved support for multi-GPU parallel inference. Models and codes will be available.

Diffusion Language Models (DLMs) have been seen as a promising competitor for autoregressive language models (ARs). However, diffusion language models have long been constrained by slow inference. A core challenge is that their non‑autoregressive architecture and bidirectional attention preclude the key–value cache that accelerates decoding. We address this bottleneck by proposing a KV-cache-like mechanism, **d**elayed **KV-Cache**, for the denoising process of DLMs. Our approach is motivated by the observation that different tokens have distinct representation dynamics throughout the diffusion process. Accordingly, we propose a delayed and conditioned caching strategy for key and value states. We design two complementary variants to cache key and value step‑by‑step: (1) dKV-Cache-Decode, which provides almost lossless acceleration, and even improves performance on long sequences, suggesting that existing DLMs may under‑utilise contextual information during inference. (2) dKV-Cache‑Greedy, which has aggressive caching with reduced lifespan, achieving higher speed-ups with quadratic time complexity at the cost of some performance degradation. dKV-Cache, in final, achieves from 2-10$\times$ speedup in inference, largely narrowing the gap between ARs and DLMs. We evaluate our dKV-Cache on several benchmarks, delivering acceleration across general language understanding, mathematical, and code‑generation benchmarks. Experiments demonstrate that cache can also be used in DLMs, even in a training-free manner from current DLMs.

Dynamic 3D scene reconstruction from multi-view videos demands representation to model complex deformations at scale. Current Gaussian Splatting based methods often either suffer from significant computation cost due to dense MLP-based modeling or explicit modeling deformation of each Gaussian independently. However, the dynamics of objects within a scene are typically hierarchical and exhibit structural correlations. To leverage these structural priors into the representation, we introduce **TreeSplat**, a **Tree** data structure for deformable Gaussian **Splat**ting. In TreeSplat, as the name suggests, motions of Gaussian are represented hierarchically within a tree. Each node learns coefficients for time-varying basis functions, defining a part of the motion. The full motion for any given Gaussian is then determined by accumulating these transformations along the tree path from its leaf node to the root node. This tree isn't predefined; instead, it is constructed adaptively alongside Gaussian densification, where cloning or splitting a Gaussian correspondingly creates new leaf nodes. One central property of TreeSplat is its mergeability; after optimization during training, the hierarchical motion parameters for each Gaussian can be efficiently consolidated. By performing this merging step before test time, we eliminate the need to traverse the tree explicitly for each Gaussian during rendering. This results in dramatically faster rendering over 200 FPS and compact storage, while maintaining state-of-the-art rendering quality. Experiments on diverse synthetic and real-world datasets validate these advantages.

Recent advances in generative models trained on large-scale datasets have enabled high-quality 3D synthesis across various domains. However, these models also raise critical privacy concerns. Unlike 2D image synthesis, where risks typically involve the leakage of visual features or identifiable patterns, 3D generation introduces additional challenges, as reconstructed shapes, textures, and spatial structures may inadvertently expose proprietary designs, biometric data, or other sensitive geometric information. This paper presents the first exploration of machine unlearning in 3D generation tasks. We investigate different unlearning objectives, including re-targeting and partial unlearning, and propose a novel framework that does not require full supervision of the unlearning target. To enable a more efficient unlearning process, we introduce a skip-acceleration mechanism, which leverages the similarity between multi-view generated images to bypass redundant computations. By establishing coherence across viewpoints during acceleration, our framework not only reduces computation but also enhances unlearning effectiveness, outperforming the non-accelerated baseline in both accuracy and efficiency. We conduct extensive experiments on the typical 3D generation models (Zero123 and Zero123XL), demonstrating that our approach achieves a 30\% speedup, while effectively unlearning target concepts without compromising generation quality. Our framework provides a scalable and practical solution for privacy-preserving 3D generation, ensuring responsible AI deployment in real-world applications. The code is available at: https://github.com/sxxsxw/Fast-3D-Unlearn-with-Skip-acceleration

Modern smartphones often feature asymmetric dual-lens systems, capturing wide-angle and ultra-wide views with complementary perspectives and details. Motion and shake can blur the wide lens, while the ultra-wide lens, despite lower resolution, retains sharper details. This natural complementarity offers valuable cues for video deblurring. However, existing methods focus mainly on single-camera inputs or symmetric stereo pairs, neglecting the cross-lens redundancy in mobile dual-camera systems. In this paper, we propose a practical video deblurring method, AsLeD-Net, which recurrently aligns and propagates temporal reference features from ultra-wide views fused with features extracted from wide-angle blurry frames. AsLeD-Net consists of two key modules: the adaptive local matching (ALM) module, which refines blurry features using $K$-nearest neighbor reference features, and the difference compensation (DC) module, which ensures spatial consistency and reduces misalignment. Additionally, AsLeD-Net uses the reference-guided motion compensation (RMC) module for temporal alignment, further improving frame-to-frame consistency in the deblurring process. We validate the effectiveness of AsLeD-Net through extensive experiments, benchmarking it against potential solutions for asymmetric lens deblurring.

Dense matching methods like DUSt3R regress pairwise pointmaps for 3D reconstruction. However, the reliance on pairwise prediction and the limited generalization capability inherently restrict the global geometric consistency. In this work, we introduce \textbf{Test3R}, a surprisingly simple test-time learning technique that significantly boosts geometric accuracy. Using image triplets ($I_1,I_2,I_3$), Test3R generates reconstructions from pairs ($I_1,I_2$) and ($I_1,I_3$). The core idea is to optimize the network at test time via a self-supervised objective: maximizing the geometric consistency between these two reconstructions relative to the common image $I_1$. This ensures the model produces cross-pair consistent outputs, regardless of the inputs. Extensive experiments demonstrate that our technique significantly outperforms previous state-of-the-art methods on the 3D reconstruction and multi-view depth estimation tasks. Moreover, it is universally applicable and nearly cost-free, making it easily applied to other models and implemented with minimal test-time training overhead and parameter footprint. Code is available at https://github.com/nopQAQ/Test3R.

4D Gaussian Splatting (4DGS) has recently gained considerable attention as a method for reconstructing dynamic scenes. Despite achieving superior quality, 4DGS typically requires substantial storage and suffers from slow rendering speed. In this work, we delve into these issues and identify two key sources of temporal redundancy. (Q1) \textbf{Short-Lifespan Gaussians}: 4DGS uses a large portion of Gaussians with short temporal span to represent scene dynamics, leading to an excessive number of Gaussians. (Q2) \textbf{Inactive Gaussians}: When rendering, only a small subset of Gaussians contributes to each frame. Despite this, all Gaussians are processed during rasterization, resulting in redundant computation overhead. To address these redundancies, we present \textbf{4DGS-1K}, which runs at over 1000 FPS on modern GPUs. For Q1, we introduce the Spatial-Temporal Variation Score, a new pruning criterion that effectively removes short-lifespan Gaussians while encouraging 4DGS to capture scene dynamics using Gaussians with longer temporal spans. For Q2, we store a mask for active Gaussians across consecutive frames, significantly reducing redundant computations. Compared to vanilla 4DGS, our method achieves a $41\times$ reduction in storage and $9\times$ faster rasterization on complex dynamic scenes, while maintaining comparable visual quality.

Prompt learning has been recently introduced into the adaption of pre-trained vision-language models (VLMs) by tuning a set of trainable tokens to replace hand-crafted text templates. Despite the encouraging results achieved, existing methods largely rely on extra annotated data for training. In this paper, we investigate a more realistic scenario, where only the unlabeled test data is available. Existing test-time prompt learning methods often separately learn a prompt for each test sample. However, relying solely on a single sample heavily limits the performance of the learned prompts, as it neglects the task-level knowledge that can be gained from multiple samples. To that end, we propose a novel test-time prompt learning method of VLMs, called Task-to-Instance PromPt LEarning (TIPPLE), which adopts a two-stage training strategy to leverage both task- and instance-level knowledge. Specifically, we reformulate the effective online pseudo-labeling paradigm along with two tailored components: an auxiliary text classification task and a diversity regularization term, to serve the task-oriented prompt learning. After that, the learned task-level prompt is further combined with a tunable residual for each test sample to integrate with instance-level knowledge. We demonstrate the superior performance of TIPPLE on 15 downstream datasets, e.g., the average improvement of 1.87% over the state-of-the-art method, using ViT-B/16 visual backbone. Our code is open-sourced at https://github.com/zhiheLu/TIPPLE.

Most existing anomaly detection (AD) methods require a dedicated model for each category. Such a paradigm, despite its promising results, is computationally expensive and inefficient, thereby failing to meet the requirements for real-world applications. Inspired by how humans detect anomalies, by comparing a query image to known normal ones, this article proposes a novel few-shot AD (FSAD) framework. Using a training set of normal images from various categories, registration, aiming to align normal images of the same categories, is leveraged as the proxy task for self-supervised category-agnostic representation learning. At test time, an image and its corresponding support set, consisting of a few normal images from the same category, are supplied, and anomalies are identified by comparing the registered features of the test image to its corresponding support image features. Such a setup enables the model to generalize to novel test categories. It is, to our best knowledge, the first FSAD method that requires no model fine-tuning for novel categories: enabling a single model to be applied to all categories. Extensive experiments demonstrate the effectiveness of the proposed method. Particularly, it improves the current state-of-the-art (SOTA) for FSAD by 11.3% and 8.3% on the MVTec and MPDD benchmarks, respectively. The source code is available at https://github.com/Haoyan-Guan/CAReg.

Adversarial attacks constitute a notable threat to machine learning systems, given their potential to induce erroneous predictions and classifications. However, within real-world contexts, the essential specifics of the deployed model are frequently treated as a black box, consequently mitigating the vulnerability to such attacks. Thus, enhancing the transferability of the adversarial samples has become a crucial area of research, which heavily relies on selecting appropriate surrogate models. To address this challenge, we propose a novel approach that generates adversarial attacks in a mutual-modality optimization scheme. Our approach is accomplished by leveraging the pre-trained CLIP model. Firstly, we conduct a visual attack on the clean image that causes semantic perturbations on the aligned embedding space with the other textual modality. Then, we apply the corresponding defense on the textual modality by updating the prompts, which forces the re-matching on the perturbed embedding space. Finally, to enhance the attack transferability, we utilize the iterative training strategy on the visual attack and the textual defense, where the two processes optimize from each other. We evaluate our approach on several benchmark datasets and demonstrate that our mutual-modal attack strategy can effectively produce high-transferable attacks, which are stable regardless of the target networks. Our approach outperforms state-of-the-art attack methods and can be readily deployed as a plug-and-play solution.

Recent advancements in large-scale visual-language pre-trained models have led to significant progress in zero/few-shot anomaly detection within natural image domains. However, the substantial domain divergence between natural and medical images limits the effectiveness of these methodologies in medical anomaly detection. This paper introduces a novel lightweight multi-level adaptation and comparison framework to repurpose the CLIP model for medical anomaly detection. Our approach integrates multiple residual adapters into the pre-trained visual encoder, enabling a stepwise enhancement of visual features across different levels. This multi-level adaptation is guided by multi-level, pixel-wise visual-language feature alignment loss functions, which recalibrate the model's focus from object semantics in natural imagery to anomaly identification in medical images. The adapted features exhibit improved generalization across various medical data types, even in zero-shot scenarios where the model encounters unseen medical modalities and anatomical regions during training. Our experiments on medical anomaly detection benchmarks demonstrate that our method significantly surpasses current state-of-the-art models, with an average AUC improvement of 6.24% and 7.33% for anomaly classification, 2.03% and 2.37% for anomaly segmentation, under the zero-shot and few-shot settings, respectively. Source code is available at: https://github.com/MediaBrain-SJTU/MVFA-AD

Foundation segmentation models, while powerful, pose a significant risk: they enable users to effortlessly extract any objects from any digital content with a single click, potentially leading to copyright infringement or malicious misuse. To mitigate this risk, we introduce a new task “Anything Unsegmentable” to grant any image “the right to be unsegmented”. The ambitious pursuit of the task is to achieve highly transferable adversarial attack against all prompt-based segmentation models, regardless of model parameterizations and prompts. We highlight the non-transferable and heterogeneous nature of promptspecific adversarial noises. Our approach focuses on disrupting image encoder features to achieve prompt-agnostic attacks. Intriguingly, targeted feature attacks exhibit better transferability compared to untargeted ones, suggesting the optimal update direction aligns with the image manifold. Based on the observations, we design a novel attack named Unsegment Anything by Simulating Deformation (UAD). Our attack optimizes a differentiable deformation function to create a target deformed image, which alters structural information while preserving achievable feature distance by adversarial example. Extensive experiments verify the effectiveness of our approach, compromising a variety of promptable segmentation models with different architectures and prompt interfaces. We release the code at https://github.com/ijahaolu97/anything-unsegmentable.

Diffusion models have recently gained unprecedented attention in the field of image synthesis due to their re-markable generative capabilities. Notwithstanding their prowess, these models often incur substantial computational costs, primarily attributed to the sequential denoising process and cumbersome model size. Traditional methods for compressing diffusion models typically involve extensive retraining, presenting cost and feasibility challenges. In this paper, we introduce DeepCache, a novel training-free paradigm that accelerates diffusion models from the per-spective of model architecture. DeepCache capitalizes on the inherent temporal redundancy observed in the sequential denoising steps of diffusion models, which caches and retrieves features across adjacent denoising stages, thereby curtailing redundant computations. Utilizing the property of the U-Net, we reuse the high-level features while updating the low-level features in a very cheap way. This innovative strategy, in turn, enables a speedup factor of 2.3× for Stable Diffusion v1.5 with only a 0.05 decline in CLIP Score, and 4.1 x for LDM-4-G with a slight de-crease of 0.22 in FID on ImageNet. Our experiments also demonstrate DeepCache's superiority over existing pruning and distillation methods that necessitate retraining and its compatibility with current sampling techniques. Furthermore, we find that under the same throughput, Deep-Cache effectively achieves comparable or even marginally improved results with DDIM or PLMS. Code is available at https://github.com/horseee/DeepCache.

Brain decoding, a pivotal field in neuroscience, aims to reconstruct stimuli from acquired brain signals, primarily utilizing functional magnetic resonance imaging (fMRI). Currently, brain decoding is confined to a per-subject-per-model paradigm, limiting its applicability to the same individual for whom the decoding model is trained. This constraint stems from three key challenges: 1) the inherent variability in input dimensions across subjects due to differences in brain size; 2) the unique intrinsic neural patterns, influencing how different individuals perceive and process sensory information; 3) limited data availability for new subjects in real-world scenarios hampers the performance of decoding models. In this paper, we present a novel approach, MindBridge, that achieves cross-subject brain decoding by employing only one model. Our proposed framework establishes a generic paradigm capable of addressing these challenges by introducing biological-inspired aggregation function and novel cyclic fMRI reconstruction mechanism for subject-invariant representation learning. Notably, by cycle re-construction of fMRI, MindBridge can enable novel fMRI synthesis, which also can serve as pseudo data augmentation. Within the framework, we also devise a novel reset-tuning method for adapting a pretrained model to a new subject. Experimental results demonstrate MindBridge's ability to reconstruct images for multiple subjects, which is competitive with dedicated subject-specific models. Fur-thermore, with limited data for a new subject, we achieve a high level of decoding accuracy, surpassing that of subject-specific models. This advancement in cross-subject brain decoding suggests promising directions for wider applications in neuroscience and indicates potential for more efficient utilization of limited fMRI data in real-world scenarios. Project page: https://littlepure2333.github.io/MindBridge

Despite their exceptional generative abilities, large T2I diffusion models, much like skilled but careless artists, often struggle with accurately depicting visual relationships between objects. This issue, as we uncover through careful analysis, arises from a misaligned text encoder that struggles to interpret specific relationships and differentiate the logical order of associated objects. To resolve this, we in-troduce a novel task termed Relation Rectification, aiming to refine the model to accurately represent a given relationship it initially fails to generate. To address this, we propose an innovative solution utilizing a Heterogeneous Graph Convolutional Network (HGCN). It models the di-rectional relationships between relation terms and corre-sponding objects within the input prompts. Specifically, we optimize the HGCN on a pair of prompts with identical relational words but reversed object orders, supplemented by a few reference images. The lightweight HGCN adjusts the text embeddings generated by the text encoder, ensuring accurate reflection of the textual relation in the em-bedding space. Crucially, our method retains the parameters of the text encoder and diffusion model, preserving the model's robust performance on unrelated descriptions. We validated our approach on a newly curated dataset of di-verse relational data, demonstrating both quantitative and qualitative enhancements in generating images with precise visual relations. Project page: https://wuyinwei-hah.github.io/rrnet.github.io/.

The training of contemporary deep learning models heavily relies on publicly available data, posing a risk of unauthorized access to online data and raising concerns about data privacy. Current approaches to creating unlearnable data involve incorporating small, specially designed noises, but these methods strictly limit data usability, overlooking its potential usage in authorized scenarios. In this paper, we extend the concept of unlearnable data to conditional data learnability and introduce UnGeneralizable Examples (UGEs). UGEs exhibit learnability for authorized users while maintaining unlearnability for potential hackers. The protector defines the authorized network and optimizes UGEs to match the gradients of the original data and its ungeneralizable version, ensuring learnability. To prevent unauthorized learning, UGEs are trained by maximizing a designated distance loss in a common feature space. Additionally, to further safeguard the authorized side from potential attacks, we introduce additional undistillation optimization. Experimental results on multiple datasets and various networks demonstrate that the proposed UGEs framework preserves data usability while reducing training performance on hacker networks, even under different types of attacks.

The proliferation of large-scale AI models trained on extensive datasets has revolutionized machine learning. With these models taking on increasingly central roles in various applications, the need to understand their behavior and enhance interpretability has become paramount. To investigate the impact of changes in training data on a pre-trained model, a common approach is leave-one-out re-training. This entails systematically altering the training dataset by removing specific samples to observe resulting changes within the model. However, retraining the model for each altered dataset presents a significant computational challenge, given the need to perform this operation for every dataset variation. In this paper, we introduce an efficient framework for assessing data impact, comprising offline training and online evaluation stages. During the offline training phase, we approximate the influence of training data on the target model through a distilled synset, formulated as a reversed gradient matching problem. For online evaluation, we expedite the leave-one-out process using the synset, which is then utilized to compute the attribution matrix based on the evaluation objective. Experimental evaluations, including training data attribution and assessments of data quality, demonstrate that our proposed method achieves comparable model behavior evaluation while significantly speeding up the process compared to the direct retraining method.

Given a well-behaved neural network, is possible to identify its parent, based on which it was tuned? In this paper, we introduce a novel task known as neural lineage detection, aiming at discovering lineage relationships between parent and child models. Specifically, from a set of parent models, neural lineage detection predicts which parent model a child model has been fine-tuned from. We propose two approaches to address this task. (1) For practical convenience, we introduce a learning-free approach, which integrates an approximation of the finetuning process into the neural network representation similarity metrics, leading to a similarity-based lineage detection scheme. (2) For the pursuit of accuracy, we introduce a learning-based lineage detector comprising encoders and a transformer detector. Through experimentation, we have validated that our proposed learning-free and learning-based methods out-perform the baseline in various learning settings and are adaptable to a variety of visual models. Moreover, they also exhibit the ability to trace cross-generational lineage, identifying not only parent models but also their ancestors.

Inspired by the long-range modeling ability of ViTs, large-kernel convolutions are widely studied and adopted recently to enlarge the receptive field and improve model performance, like the remarkable work ConvNeXt which employs 7 × 7 depthwise convolution. Although such depth-wise operator only consumes a few FLOPs, it largely harms the model efficiency on powerful computing devices due to the high memory access costs. For example, ConvNeXt-T has similar FLOPs with ResNet-50 but only achieves ~60% throughputs when trained on A100 GPUs with full precision. Although reducing the kernel size of ConvNeXt can improve speed, it results in significant performance degradation, which poses a challenging problem: How to speed up large-kernel-based CNN models while preserving their performance. To tackle this issue, inspired by Inceptions, we propose to decompose large-kernel depth-wise convolution into four parallel branches along channel dimension, i.e., small square kernel, two orthogonal band kernels, and an identity mapping. With this new Inception depthwise convolution, we build a series of networks, namely IncepitonNeXt, which not only enjoy high throughputs but also maintain competitive performance. For instance, InceptionNeXt-T achieves 1.6 × higher training throughputs than ConvNeX-T, as well as attains 0.2% top-1 accuracy improvement on ImageNet-1K. We antici-pate InceptionNeXt can serve as an economical baseline for future architecture design to reduce carbon footprint.

With the development of low-altitude network, cross-domain collaboration between unmanned aerial vehicle (UAV) and ground station is becoming increasingly important. To ensure the legitimacy of identity, it is imperative to design a cross-domain authentication scheme. However, current cross-domain authentication schemes still face challenges such as low computational efficiency and high storage overhead, and still have some security vulnerabilities. This paper proposes a cross-domain authentication and key agreement scheme for UAVs, which primarily employs a combination of symmetric encryption and hash functions. By holding cross-domain tokens, UAVs can directly engage in cross-domain authentication, ultimately achieving key agreement with ground station. We conduct security and performance analyses, demonstrating the feasibility of our scheme in resource-constrained low-altitude network.

Self-supervised heterogeneous graph learning has achieved promising results in various real applications, but it still suffers from the following issues: (i) meta-paths can be employed to capture the homophily in the heterogeneous graph, but meta-paths are human-defined, requiring substantial expert knowledge and computational costs; and (ii) the heterogeneity in the heterogeneous graph is usually underutilized, leading to the loss of task-related information. To solve these issues, this paper proposes to capture both homophily and heterogeneity in the heterogeneous graph without pre-defined meta-paths. Specifically, we propose to learn a self-expressive matrix to capture the homophily from the subspace and nearby neighbors. Meanwhile, we propose to capture the heterogeneity by aggregating the information of nodes from different types. We further design a consistency loss and a specificity loss, respectively, to extract the consistent information between homophily and heterogeneity and to preserve their specific task-related information. We theoretically analyze that the learned homophilous representations exhibit the grouping effect to capture the homophily, and considering both homophily and heterogeneity introduces more task-related information. Extensive experimental results verify the superiority of the proposed method on different downstream tasks.

Unsupervised domain adaptation (UDA) has seen substantial efforts to improve model accuracy for an unlabeled target domain with the help of a labeled source domain. However, UDA models often exhibit poorly calibrated predictive uncertainty on target data, a problem that remains under-explored and poses risks in safety-critical UDA applications. The calibration problem in UDA is particularly challenging due to the absence of labeled target data and severe distribution shifts between domains. In this paper, we approach UDA calibration as a target-domain-specific unsupervised problem, different from mainstream solutions based on *covariate shift*. We introduce Pseudo-Calibration (PseudoCal), a novel post-hoc calibration framework. Our innovative use of inference-stage *mixup* synthesizes a labeled pseudo-target set capturing the structure of the real unlabeled target data. This turns the unsupervised calibration problem into a supervised one, easily solvable with *temperature scaling*. Extensive empirical evaluations across 5 diverse UDA scenarios involving 10 UDA methods consistently demonstrate the superior performance and versatility of PseudoCal over existing solutions.

The size of spatial receptive fields, from the early 3$\times$3 convolutions in VGGNet to the recent 7$\times$7 convolutions in ConvNeXt, has always played a critical role in architecture design. In this paper, we propose a Mixture of Receptive Fields (MoRF) instead of using a single receptive field. MoRF contains the combinations of multiple receptive fields with different sizes, e.g., convolutions with different kernel sizes, which can be regarded as experts. Such an approach serves two functions: one is to select the appropriate receptive field according to the input, and the other is to expand the network capacity. Furthermore, we also introduce two types of routing mechanisms, hard routing and soft routing to automatically select the appropriate receptive field experts. In the inference stage, the selected receptive field experts are merged via re-parameterization to maintain a similar inference speed compared to the single receptive field. To demonstrate the effectiveness of MoRF, we integrate the MoRF concept into multiple architectures, e.g., ResNet and ConvNeXt. Extensive experiments show that our approach outperforms the baselines in image classification, object detection, and segmentation tasks without significantly increasing the inference time.

Single domain generalization (single DG) aims at learning a robust model generalizable to unseen domains from only one training domain, making it a highly ambitious and challenging task. State-of-the-art approaches have mostly relied on data augmentations, such as adversarial perturbation and style enhancement, to synthesize new data and thus increase robustness. Nevertheless, they have largely overlooked the underlying coherence between the augmented domains, which in turn leads to inferior results in real-world scenarios. In this paper, we propose a simple yet effective scheme, termed as StyDeSty, to explicitly account for the alignment of the source and pseudo domains in the process of data augmentation, enabling them to interact with each other in a self-consistent manner and further giving rise to a latent domain with strong generalization power. The heart of StyDeSty lies in the interaction between a stylization module for generating novel stylized samples using the source domain, and a destylization module for transferring stylized and source samples to a latent domain to learn content-invariant features. The stylization and destylization modules work adversarially and reinforce each other. During inference, the destylization module transforms the input sample with an arbitrary style shift to the latent domain, in which the downstream tasks are carried out. Specifically, the location of the destylization layer within the backbone network is determined by a dedicated neural architecture search (NAS) strategy. We evaluate StyDeSty on multiple benchmarks and demonstrate that it yields encouraging results, outperforming the state of the art by up to 13.44% on classification accuracy. Codes are available https://github.com/Huage001/StyDeSty.

Fine-tuning pre-trained vision-language models (VLMs), e.g., CLIP, for the open-world generalization has gained increasing popularity due to its practical value. However, performance advancements are limited when relying solely on intricate algorithmic designs for a single model, even one exhibiting strong performance, e.g., CLIP-ViT-B/16. This paper, for the first time, explores the collaborative potential of leveraging much weaker VLMs to enhance the generalization of a robust single model. The affirmative findings motivate us to address the generalization problem from a novel perspective, i.e., ensemble of pre-trained VLMs. We introduce three customized ensemble strategies, each tailored to one specific scenario. Firstly, we introduce the zero-shot ensemble, automatically adjusting the logits of different models based on their confidence when only pre-trained VLMs are available. Furthermore, for scenarios with extra few-shot samples, we propose the training-free and tuning ensemble, offering flexibility based on the availability of computing resources. The code is available at https://github.com/zhiheLu/Ensemble_VLM.git.

We propose MM-Vet, an evaluation benchmark that examines large multimodal models (LMMs) on complicated multimodal tasks. Recent LMMs have shown various intriguing abilities, such as solving math problems written on the blackboard, reasoning about events and celebrities in news images, and explaining visual jokes. Rapid model advancements pose challenges to evaluation benchmark development. Problems include: (1) How to systematically structure and evaluate the complicated multimodal tasks; (2) How to design evaluation metrics that work well across question and answer types; and (3) How to give model insights beyond a simple performance ranking. To this end, we present MM-Vet, designed based on the insight that the intriguing ability to solve complicated tasks is often achieved by a generalist model being able to integrate different core vision-language (VL) capabilities. MM-Vet defines 6 core VL capabilities and examines the 16 integrations of interest derived from the capability combination. For evaluation metrics, we propose an LLM-based evaluator for open-ended outputs. The evaluator enables the evaluation across different question types and answer styles, resulting in a unified scoring metric. We evaluate representative LMMs on MM-Vet, providing insights into the capabilities of different LMM system paradigms and models.

Latent diffusion models have shown promising results in audio generation, making notable advancements over traditional methods.However, their performance, while impressive with short audio clips, faces challenges when extended to longer audio sequences.These challenges are due to model's selfattention mechanism and training predominantly on 10-second clips, which complicates the extension to longer audio without adaptation.In response to these issues, we introduce a novel approach, LiteFocus that enhances the inference of existing audio latent diffusion models in long audio synthesis.Observed the attention pattern in self-attention, we employ a dual sparse form for attention calculation, designated as same-frequency focus and cross-frequency compensation, which curtails the attention computation under same-frequency constraints, while enhancing audio quality through cross-frequency refillment.LiteFocus demonstrates substantial reduction on inference time with diffusion-based TTA model by 1.99× in synthesizing 80second audio clips while also obtaining improved audio quality.

In real-world photography, local motion blur often arises from the interplay between moving objects and stationary backgrounds during exposure. Existing deblurring methods face challenges in addressing local motion deblurring due to (i) the presence of arbitrary localized blurs and uncertain blur extents; (ii) the limited ability to accurately identify specific blurs resulting from ambiguous motion boundaries. These limitations often lead to suboptimal solutions when estimating blur maps and generating final deblurred images. To that end, we propose a novel method named Motion-Uncertainty-Guided Network (MUGNet), which harnesses a probabilistic representational model to explicitly address the intricacies stemming from motion uncertainties. Specifically, MUGNet consists of two key components, i.e., motion-uncertainty quantification (MUQ) module and motion-masked separable attention (M2SA) module, serving for complementary purposes. Concretely, MUQ aims to learn a conditional distribution for accurate and reliable blur map estimation, while the M2SA module is to enhance the representation of regions influenced by local motion blur and static background, which is achieved by promoting the establishment of extensive global interactions. We demonstrate the superiority of our MUGNet with extensive experiments. The code is publicly available at: https://github.com/zeyuxiao1997/MUGNet.

People nowadays use smartphones to capture photos from multimedia platforms. The presence of moire patterns resulting from spectral aliasing can significantly degrade the visual quality of images, particularly in ultra-high-definition (UHD) images. However, existing demoireing methods have mostly been designed for low-definition images, making them unsuitable for handling moire patterns in UHD images due to their substantial memory requirements. In this paper, we propose a novel patch bilateral compensation network (P-BiC) for the demoire pattern removal in UHD images, which is memory-efficient and prior-knowledge-based. Specifically, we divide the UHD images into small patches and perform patch-level demoireing to maintain the low memory cost even for ultra-large image sizes. Moreover, a pivotal insight, namely that the green channel of an image remains relatively less affected by moire patterns, while the tone information in moire images is still well-retained despite color shifts, is directly harnessed for the purpose of bilateral compensation. The bilateral compensation is achieved by two key components in our P-BiC, i.e., a green-guided detail transfer (G2DT) module that complements distorted features with the intact content, and a style-aware tone adjustment (STA) module for the color adjustment. We quantitatively and qualitatively evaluate the effectiveness of P-BiC with extensive experiments. The code is publicly available at: https://github.com/zeyuxiao1997/P-BiC.

Current approaches for compressing the Segment Anything Model (SAM) yield commendable results, yet necessitate extensive data to train a new network from scratch. Employing conventional pruning techniques can remarkably reduce data requirements but would suffer from a degradation in performance. To address this challenging trade-off, we introduce SlimSAM, a novel data-efficient SAM compression method that achieves superior performance with extremely less training data. The essence of SlimSAM is encapsulated in the alternate slimming framework which effectively enhances knowledge inheritance under severely limited training data availability and exceptional pruning ratio. Diverging from prior techniques, our framework progressively compresses the model by alternately pruning and distilling distinct, decoupled sub-structures. Disturbed Taylor pruning is also proposed to address the misalignment between the pruning objective and training target, thereby boosting the post-distillation after pruning. SlimSAM yields significant performance improvements while demanding over 10 times less training data than any other existing compression methods. Even when compared to the original SAM, SlimSAM achieves approaching performance while reducing parameter counts to merely 1.4% (9.1M), MACs to 0.8% (23G), and requiring only 0.1% (10k) of the SAM training data.

Diffusion models have garnered significant interest from the community for their great generative ability across various applications. However, their typical multi-step sequential-denoising nature gives rise to high cumulative latency, thereby precluding the possibilities of parallel computation. To address this, we introduce AsyncDiff, a universal and plug-and-play acceleration scheme that enables model parallelism across multiple devices. Our approach divides the cumbersome noise prediction model into multiple components, assigning each to a different device. To break the dependency chain between these components, it transforms the conventional sequential denoising into an asynchronous process by exploiting the high similarity between hidden states in consecutive diffusion steps. Consequently, each component is facilitated to compute in parallel on separate devices. The proposed strategy significantly reduces inference latency while minimally impacting the generative quality. Specifically, for the Stable Diffusion v2.1, AsyncDiff achieves a 2.7x speedup with negligible degradation and a 4.0x speedup with only a slight reduction of 0.38 in CLIP Score, on four NVIDIA A5000 GPUs. Our experiments also demonstrate AsyncDiff can be readily applied to video diffusion models with encouraging performances.

Transformer-based diffusion models have achieved significant advancements across a variety of generative tasks. However, producing high-quality outputs typically necessitates large transformer models, which result in substantial training and inference overhead. In this work, we investigate an alternative approach involving multiple experts for denoising, and introduce RemixDiT, a novel method designed to enhance output quality at a low cost. The goal of RemixDiT is to craft N diffusion experts for different denoising timesteps, yet without the need for expensive training of N independent models. To achieve this, RemixDiT employs K basis models (where K < N) and utilizes learnable mixing coefficients to adaptively craft expert models. This design offers two significant advantages: first, although the total model size is increased, the model produced by the mixing operation shares the same architecture as a plain model, making the overall model as efficient as a standard diffusion transformer. Second, the learnable mixing adaptively allocates model capacity across timesteps, thereby effectively improving generation quality. Experiments conducted on the ImageNet dataset demonstrate that RemixDiT achieves promising results compared to standard diffusion transformers and other multiple-expert methods.

Large Language Models (LLMs) are distinguished by their massive parameter counts, which typically result in significant redundancy. This work introduces MaskLLM, a learnable pruning method that establishes Semi-structured (or ``N:M'') Sparsity in LLMs, aimed at reducing computational overhead during inference. Instead of developing a new importance criterion, MaskLLM explicitly models N:M patterns as a learnable distribution through Gumbel Softmax sampling. This approach facilitates end-to-end training on large-scale datasets and offers two notable advantages: 1) High-quality Masks - our method effectively scales to large datasets and learns accurate masks; 2) Transferability - the probabilistic modeling of mask distribution enables the transfer learning of sparsity across domains or tasks. We assessed MaskLLM using 2:4 sparsity on various LLMs, including LLaMA-2, Nemotron-4, and GPT-3, with sizes ranging from 843M to 15B parameters, and our empirical results show substantial improvements over state-of-the-art methods. For instance, leading approaches achieve a perplexity (PPL) of 10 or greater on Wikitext compared to the dense model's 5.12 PPL, but MaskLLM achieves a significantly lower 6.72 PPL solely by learning the masks with frozen weights. Furthermore, MaskLLM's learnable nature allows customized masks for lossless application of 2:4 sparsity to downstream tasks or domains. Code is available at https://github.com/NVlabs/MaskLLM.

Diffusion Transformers have recently demonstrated unprecedented generative capabilities for various tasks. The encouraging results, however, come with the cost of slow inference, since each denoising step requires inference on a transformer model with a large scale of parameters. In this study, we make an interesting and somehow surprising observation: the computation of a large proportion of layers in the diffusion transformer, through introducing a caching mechanism, can be readily removed even without updating the model parameters. In the case of U-ViT-H/2, for example, we may remove up to 93.68% of the computation in the cache steps (46.84% for all steps), with less than 0.01 drop in FID. To achieve this, we introduce a novel scheme, named Learning-to-Cache (L2C), that learns to conduct caching in a dynamic manner for diffusion transformers. Specifically, by leveraging the identical structure of layers in transformers and the sequential nature of diffusion, we explore redundant computations between timesteps by treating each layer as the fundamental unit for caching. To address the challenge of the exponential search space in deep models for identifying layers to cache and remove, we propose a novel differentiable optimization objective. An input-invariant yet timestep-variant router is then optimized, which can finally produce a static computation graph. Experimental results show that L2C largely outperforms samplers such as DDIM and DPM-Solver, alongside prior cache-based methods at the same inference speed.

Self-supervised heterogeneous graph learning (SHGL) has shown promising potential in diverse scenarios. However, while existing SHGL methods share a similar essential with clustering approaches, they encounter two significant limitations: (i) noise in graph structures is often introduced during the message-passing process to weaken node representations, and (ii) cluster-level information may be inadequately captured and leveraged, diminishing the performance in downstream tasks. In this paper, we address these limitations by theoretically revisiting SHGL from the spectral clustering perspective and introducing a novel framework enhanced by rank and dual consistency constraints. Specifically, our framework incorporates a rank-constrained spectral clustering method that refines the affinity matrix to exclude noise effectively. Additionally, we integrate node-level and cluster-level consistency constraints that concurrently capture invariant and clustering information to facilitate learning in downstream tasks. We theoretically demonstrate that the learned representations are divided into distinct partitions based on the number of classes and exhibit enhanced generalization ability across tasks. Experimental results affirm the superiority of our method, showcasing remarkable improvements in several downstream tasks compared to existing methods.

Procedurally generated environments such as Procgen Benchmark provide a testbed for evaluating the agent's ability to robustly learn a relevant skill, by situating the agent in ever-changing levels. The diverse levels associated with varying contexts are naturally connected to curriculum learning. Existing works mainly focus on arranging the levels to explicitly form a curriculum. In this work, we take a close look at the learning process itself under the multi-level training in Procgen. Interestingly, the learning process exhibits a gradual shift from easy contexts to hard contexts, suggesting an implicit curriculum in multi-level training. Our analysis is made possible through C-Procgen, a benchmark we build upon Procgen that enables explicit control of the contexts. We believe our findings will foster a deeper understanding of learning in diverse contexts, and our benchmark will benefit future research in curriculum reinforcement learning.

In this paper, we present Language Model as Visual Explainer (\texttt{LVX}), a systematic approach for interpreting the internal workings of vision models using a tree-structured linguistic explanation, without the need for model training. Central to our strategy is the collaboration between vision models and LLM to craft explanations. On one hand, the LLM is harnessed to delineate hierarchical visual attributes, while concurrently, a text-to-image API retrieves images that are most aligned with these textual concepts. By mapping the collected texts and images to the vision model's embedding space, we construct a hierarchy-structured visual embedding tree. This tree is dynamically pruned and grown by querying the LLM using language templates, tailoring the explanation to the model. Such a scheme allows us to seamlessly incorporate new attributes while eliminating undesired concepts based on the model's representations. When applied to testing samples, our method provides human-understandable explanations in the form of attribute-laden trees. Beyond explanation, we retrained the vision model by calibrating it on the generated concept hierarchy, allowing the model to incorporate the refined knowledge of visual attributes. To access the effectiveness of our approach, we introduce new benchmarks and conduct rigorous evaluations, demonstrating its plausibility, faithfulness, and stability.

Recent 3D large reconstruction models (LRMs) can generate high-quality 3D content in sub-seconds by integrating multi-view diffusion models with scalable multi-view reconstructors. Current works further leverage 3D Gaussian Splatting as 3D representation for improved visual quality and rendering efficiency. However, we observe that existing Gaussian reconstruction models often suffer from multi-view inconsistency and blurred textures. We attribute this to the compromise of multi-view information propagation in favor of adopting powerful yet computationally intensive architectures (\eg, Transformers). To address this issue, we introduce MVGamba, a general and lightweight Gaussian reconstruction model featuring a multi-view Gaussian reconstructor based on the RNN-like State Space Model (SSM). Our Gaussian reconstructor propagates causal context containing multi-view information for cross-view self-refinement while generating a long sequence of Gaussians for fine-detail modeling with linear complexity. With off-the-shelf multi-view diffusion models integrated, MVGamba unifies 3D generation tasks from a single image, sparse images, or text prompts. Extensive experiments demonstrate that MVGamba outperforms state-of-the-art baselines in all 3D content generation scenarios with approximately only $0.1\times$ of the model size. The codes are available at \url{https://github.com/SkyworkAI/MVGamba}.

One-shot Federated learning (FL) is a powerful technology facilitating collaborative training of machine learning models in a single round of communication. While its superiority lies in communication efficiency and privacy preservation compared to iterative FL, one-shot FL often compromises model performance. Prior research has primarily focused on employing data-free knowledge distillation to optimize data generators and ensemble models for better aggregating local knowledge into the server model. However, these methods typically struggle with data heterogeneity, where inconsistent local data distributions can cause teachers to provide misleading knowledge. Additionally, they may encounter scalability issues with complex datasets due to inherent two-step information loss: first, during local training (from data to model), and second, when transferring knowledge to the server model (from model to inversed data). In this paper, we propose FedSD2C, a novel and practical one-shot FL framework designed to address these challenges. FedSD2C introduces a distiller to synthesize informative distillates directly from local data to reduce information loss and proposes sharing synthetic distillates instead of inconsistent local models to tackle data heterogeneity. Our empirical results demonstrate that FedSD2C consistently outperforms other one-shot FL methods with more complex and real datasets, achieving up to 2.6 $\times$ the performance of the best baseline. Code: https://github.com/Carkham/FedSD2C

Recent success of deep learning is largely attributed to the sheer amount of data used for training deep neural networks. Despite the unprecedented success, the massive data, unfortunately, significantly increases the burden on storage and transmission and further gives rise to a cumbersome model training process. Besides, relying on the raw data for training per se yields concerns about privacy and copyright. To alleviate these shortcomings, dataset distillation (DD), also known as dataset condensation (DC), was introduced and has recently attracted much research attention in the community. Given an original dataset, DD aims to derive a much smaller dataset containing synthetic samples, based on which the trained models yield performance comparable with those trained on the original dataset. In this paper, we give a comprehensive review and summary of recent advances in DD and its application. We first introduce the task formally and propose an overall algorithmic framework followed by all existing DD methods. Next, we provide a systematic taxonomy of current methodologies in this area, and discuss their theoretical interconnections. We also present current challenges in DD through extensive empirical studies and envision possible directions for future works.

MetaFormer, the abstracted architecture of Transformer, has been found to play a significant role in achieving competitive performance. In this paper, we further explore the capacity of MetaFormer, again, by migrating our focus away from the token mixer design: we introduce several baseline models under MetaFormer using the most basic or common mixers, and demonstrate their gratifying performance. We summarize our observations as follows: (1) MetaFormer ensures solid lower bound of performance. By merely adopting identity mapping as the token mixer, the MetaFormer model, termed IdentityFormer, achieves [Formula: see text]80% accuracy on ImageNet-1 K. (2) MetaFormer works well with arbitrary token mixers. When specifying the token mixer as even a random matrix to mix tokens, the resulting model RandFormer yields an accuracy of [Formula: see text]81%, outperforming IdentityFormer. Rest assured of MetaFormer's results when new token mixers are adopted. (3) MetaFormer effortlessly offers state-of-the-art results. With just conventional token mixers dated back five years ago, the models instantiated from MetaFormer already beat state of the art. (a) ConvFormer outperforms ConvNeXt. Taking the common depthwise separable convolutions as the token mixer, the model termed ConvFormer, which can be regarded as pure CNNs, outperforms the strong CNN model ConvNeXt. (b) CAFormer sets new record on ImageNet-1 K. By simply applying depthwise separable convolutions as token mixer in the bottom stages and vanilla self-attention in the top stages, the resulting model CAFormer sets a new record on ImageNet-1 K: it achieves an accuracy of 85.5% at 224 ×224 resolution, under normal supervised training without external data or distillation. In our expedition to probe MetaFormer, we also find that a new activation, StarReLU, reduces 71% FLOPs of activation compared with commonly-used GELU yet achieves better performance. Specifically, StarReLU is a variant of Squared ReLU dedicated to alleviating distribution shift. We expect StarReLU to find great potential in MetaFormer- like models alongside other neural networks. Code and models are available at https://github.com/sail-sg/metaformer.

Large-scale vision-language models (VLMs) pre-trained on billion-level data have learned general visual representations and broad visual concepts. In principle, the welllearned knowledge structure of the VLMs should be inherited appropriately when being transferred to downstream tasks with limited data. However, most existing efficient transfer learning (ETL) approaches for VLMs either damage or are excessively biased towards the prior knowledge, e.g., prompt tuning (PT) discards the pre-trained text-based classifier and builds a new one while adapter-style tuning (AT) fully relies on the pre-trained features. To address this, we propose a new efficient tuning approach for VLMs named Task Residual Tuning (TaskRes), which performs directly on the text-based classifier and explicitly decouples the prior knowledge of the pre-trained models and new knowledge regarding a target task. Specifically, TaskRes keeps the original classifier weights from the VLMs frozen and obtains a new classifier for the target task by tuning a set of prior-independent parameters as a residual to the original one, which enables reliable prior knowledge preservation and flexible task-specific knowledge exploration. The proposed TaskRes is simple yet effective, which significantly outperforms previous ETL methods (e.g., PT and AT) on 11 benchmark datasets while requiring minimal effort for the implementation. Our code is available at https://github.com/geekyutao/TaskRes.

Structural pruning enables model acceleration by removing structurally-grouped parameters from neural networks. However, the parameter-grouping patterns vary widely across different models, making architecture-specific pruners, which rely on manually-designed grouping schemes, non-generalizable to new architectures. In this work, we study a highly-challenging yet barely-explored task, any structural pruning, to tackle general structural pruning of arbitrary architecture like CNNs, RNNs, GNNs and Transformers. The most prominent obstacle towards this goal lies in the structural coupling, which not only forces different layers to be pruned simultaneously, but also expects all removed parameters to be consistently unimportant, thereby avoiding structural issues and significant performance degradation after pruning. To address this problem, we propose a general and fully automatic method, Dependency Graph (DepGraph), to explicitly model the dependency between layers and comprehensively group coupled parameters for pruning. In this work, we extensively evaluate our method on several architectures and tasks, including ResNe(X)t, DenseNet, MobileNet and Vision transformer for images, GAT for graph, DGCNN for 3D point cloud, alongside LSTM for language, and demonstrate that, even with a simple norm-based criterion, the proposed method consistently yields gratifying performances.

In this paper, we explore a novel model reusing task tailored for graph neural networks (GNNs), termed as “deep graph reprogramming”. We strive to reprogram a pretrained GNN, without amending raw node features nor model parameters, to handle a bunch of cross-level downstream tasks in various domains. To this end, we propose an innovative Data Reprogramming paradigm alongside a Model Reprogramming paradigm. The former one aims to address the challenge of diversified graph feature dimensions for various tasks on the input side, while the latter alleviates the dilemma of fixed per-task-per-model behavior on the model side. For data reprogramming, we specifically devise an elaborated Meta-FeatPadding method to deal with heterogeneous input dimensions, and also develop a transductive Edge-Slimming as well as an inductive Meta-GraPadding approach for diverse homogenous samples. Meanwhile, for model reprogramming, we propose a novel task-adaptive Reprogrammable-Aggregator, to endow the frozen model with larger expressive capacities in handling cross-domain tasks. Experiments on fourteen datasets across node/graph classification/regression, 3D object recognition, and distributed action recognition, demonstrate that the proposed methods yield gratifying results, on par with those by re-training from scratch.

Dataset distillation, also known as dataset condensation, aims to compress a large dataset into a compact synthetic one. Existing methods perform dataset condensation by assuming a fixed storage or transmission budget. When the budget changes, however, they have to repeat the synthesizing process with access to original datasets, which is highly cumbersome if not infeasible at all. In this paper, we explore the problem of slimmable dataset condensation, to extract a smaller synthetic dataset given only previous condensation results. We first study the limitations of existing dataset condensation algorithms on such a successive compression setting and identify two key factors: (1) the inconsistency of neural networks over different compression times and (2) the underdetermined solution space for synthetic data. Accordingly, we propose a novel training objective for slimmable dataset condensation to explicitly account for both factors. Moreover, synthetic datasets in our method adopt a significance-aware parameterization. Theoretical derivation indicates that an upper-bounded error can be achieved by discarding the minor components without training. Alternatively, if training is allowed, this strategy can serve as a strong initialization that enables a fast convergence. Extensive comparisons and ablations demonstrate the superiority of the proposed solution over existing methods on multiple benchmarks.

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.

Learning to predict agent motions with relationship reasoning is important for many applications. In motion prediction tasks, maintaining motion equivariance under Euclidean geometric transformations and invariance of agent interaction is a critical and fundamental principle. However, such equivariance and invariance properties are overlooked by most existing methods. To fill this gap, we propose Eq-Motion, an efficient equivariant motion prediction model with invariant interaction reasoning. To achieve motion equivariance, we propose an equivariant geometric feature learning module to learn a Euclidean transformable feature through dedicated designs of equivariant operations. To reason agent's interactions, we propose an invariant interaction reasoning module to achieve a more stable interaction modeling. To further promote more comprehensive motion features, we propose an invariant pattern feature learning module to learn an invariant pattern feature, which cooperates with the equivariant geometric feature to enhance network expressiveness. We conduct experiments for the proposed model on four distinct scenarios: particle dynamics, molecule dynamics, human skeleton motion prediction and pedestrian trajectory prediction. Experimental results show that our method is not only generally applicable, but also achieves state-of-the-art prediction performances on all the four tasks, improving by 24.0/30.1/8.6/9.2%. Code is available at https://github.com/MediaBrain-SJTU/EqMotion.

Despite the recent visually-pleasing results achieved, the massive computational cost has been a long-standing flaw for diffusion probabilistic models (DPMs), which, in turn, greatly limits their applications on resource-limited platforms. Prior methods towards efficient DPM, however, have largely focused on accelerating the testing yet overlooked their huge complexity and sizes. In this paper, we make a dedicated attempt to lighten DPM while striving to preserve its favourable performance. We start by training a small-sized latent diffusion model (LDM) from scratch, but observe a significant fidelity drop in the synthetic images. Through a thorough assessment, we find that DPM is intrinsically biased against high-frequency generation, and learns to recover different frequency components at different time-steps. These properties make compact networks unable to represent frequency dynamics with accurate high-frequency estimation. Towards this end, we introduce a customized design for slim DPM, which we term as Spectral Diffusion (SD), for light-weight image synthesis. SD incorporates wavelet gating in its architecture to enable frequency dynamic feature extraction at every reverse step, and conducts spectrum-aware distillation to promote high-frequency recovery by inverse weighting the objective based on spectrum magnitude. Experimental results demonstrate that, SD achieves 8–18 × computational complexity reduction as compared to the latent diffusion models on a series of conditional and unconditional image generation tasks while retaining competitive image fidelity.

In this paper, we study a novel task that enables partial knowledge transfer from pre-trained models, which we term as Partial Network Cloning (PNC). Unlike prior methods that update all or at least part of the parameters in the target network throughout the knowledge transfer process, PNC conducts partial parametric “cloning” from a source network and then injects the cloned module to the target, without modifying its parameters. Thanks to the transferred module, the target network is expected to gain additional functionality, such as inference on new classes; whenever needed, the cloned module can be readily removed from the target, with its original parameters and competence kept intact. Specifically, we introduce an innovative learning scheme that allows us to identify simultaneously the component to be cloned from the source and the position to be inserted within the target network, so as to ensure the optimal performance. Experimental results on several datasets demonstrate that, our method yields a significant improvement of 5% in accuracy and 50% in locality when compared with parameter-tuning based methods. Our code is available at https://github.com/JngwenYe/PNCloning.

Machine learning society has witnessed the emergence of a myriad of Out-of-Distribution (OoD) algorithms, which address the distribution shift between the training and the testing distribution by searching for a unified predictor or invariant feature representation. However, the task of directly mitigating the distribution shift in the unseen testing set is rarely investigated, due to the unavailability of the testing distribution during the training phase and thus the impossibility of training a distribution translator mapping between the training and testing distribution. In this paper, we explore how to bypass the requirement of testing distribution for distribution translator training and make the distribution translation useful for OoD prediction. We propose a portable Distribution Shift Inversion (DSI) algorithm, in which, before being fed into the prediction model, the OoD testing samples are first linearly combined with additional Gaussian noise and then transferred back towards the training distribution using a diffusion model trained only on the source distribution. Theoretical analysis reveals the feasibility of our method. Experimental results, on both multiple-domain generalization datasets and single-domain generalization datasets, show that our method provides a general performance gain when plugged into a wide range of commonly used OoD algorithms. Our code is available at https://github.com/yu-rp/Distribution-Shift-Iverson.

We propose a novel task for generating 3D dance movements that simultaneously incorporate both text and music modalities. Unlike existing works that generate dance movements using a single modality such as music, our goal is to produce richer dance movements guided by the instructive information provided by the text. However, the lack of paired motion data with both music and text modalities limits the ability to generate dance movements that integrate both. To alleviate this challenge, we propose to utilize a 3D human motion VQ-VAE to project the motions of the two datasets into a latent space consisting of quantized vectors, which effectively mix the motion tokens from the two datasets with different distributions for training. Additionally, we propose a cross-modal transformer to integrate text instructions into motion generation architecture for generating 3D dance movements without degrading the performance of music-conditioned dance generation. To better evaluate the quality of the generated motion, we introduce two novel metrics, namely Motion Prediction Distance (MPD) and Freezing Score (FS), to measure the coherence and freezing percentage of the generated motion. Extensive experiments show that our approach can generate realistic and coherent dance movements conditioned on both text and music while maintaining comparable performance with the two single modalities. Code is available at https://garfield-kh.github.io/TM2D/.

Text-to-motion generation is a formidable task, aiming to produce human motions that align with the input text while also adhering to human capabilities and physical laws. While there have been advancements in diffusion models, their application in discrete spaces remains underexplored. Current methods often overlook the varying significance of different motions, treating them uniformly. It is essential to recognize that not all motions hold the same relevance to a particular textual description. Some motions, being more salient and informative, should be given precedence during generation. In response, we introduce a Priority-Centric Motion Discrete Diffusion Model (M2DM), which utilizes a Transformer-based VQ-VAE to derive a concise, discrete motion representation, incorporating a global self-attention mechanism and a regularization term to counteract code collapse. We also present a motion discrete diffusion model that employs an innovative noise schedule, determined by the significance of each motion token within the entire motion sequence. This approach retains the most salient motions during the reverse diffusion process, leading to more semantically rich and varied motions. Additionally, we formulate two strategies to gauge the importance of motion tokens, drawing from both textual and visual indicators. Comprehensive experiments on the HumanML3D and KIT-ML datasets confirm that our model surpasses existing techniques in fidelity and diversity, particularly for intricate textual descriptions.

Dataset distillation aims at a small synthetic dataset to mimic the training performance on neural networks of a given large dataset. Existing approaches heavily rely on an iterative optimization to update synthetic data and multiple forward-backward passes over thousands of neural network spaces, which introduce significant overhead for computation and are inconvenient in scenarios requiring high efficiency. In this paper, we focus on few-shot dataset distillation, where a distilled dataset is synthesized with only a few or even a single network. To this end, we introduce the notion of distillation space, such that synthetic data optimized only in this specific space can achieve the effect of those optimized through numerous neural networks, with dramatically accelerated training and reduced computational cost. To learn such a distillation space, we first formulate the problem as a quad-level optimization framework and propose a bi-level algorithm. Nevertheless, the algorithm in its original form has a large memory footprint in practice due to the back-propagation through an unrolled computational graph. We then convert the problem of learning the distillation space to a first-order one based on image translation. Specifically, the synthetic images are optimized in an arbitrary but fixed neural space and then translated to those in the targeted distillation space. We pre-train the translator on some large datasets like ImageNet so that it requires only a limited number of adaptation steps on the target dataset. Extensive experiments demonstrate that the translator after pre-training and a limited number of adaptation steps achieves comparable distillation performance with state of the arts, with ∼ 15× acceleration. It also exerts satisfactory generalization performance across different datasets, storage budgets, and numbers of classes.

Recently, polar-based representation has shown promising properties in perceptual tasks. In addition to Cartesian-based approaches, which separate point clouds unevenly, representing point clouds as polar grids has been recognized as an alternative due to (1) its advantage in robust performance under different resolutions and (2) its superiority in streaming-based approaches. However, state-of-the-art polar-based detection methods inevitably suffer from the feature distortion problem because of the non-uniform division of polar representation, resulting in a non-negligible performance gap compared to Cartesian-based approaches. To tackle this issue, we present PARTNER, a novel 3D object detector in the polar coordinate. PARTNER alleviates the dilemma of feature distortion with global representation re-alignment and facilitates the regression by introducing instance-level geometric information into the detection head. Extensive experiments show overwhelming advantages in streaming-based detection and different resolutions. Furthermore, our method outperforms the previous polar-based works with remarkable margins of 3.68% and 9.15% on Waymo and ONCE validation set, thus achieving competitive results over the state-of-the-art methods.

Vision Transformer has demonstrated impressive success across various vision tasks. However, its heavy computation cost, which grows quadratically with respect to the token sequence length, largely limits its power in handling large feature maps. To alleviate the computation cost, previous works rely on either fine-grained self-attentions restricted to local small regions, or global self-attentions but to shorten the sequence length resulting in coarse granularity. In this paper, we propose a novel model, termed as Self-guided Transformer (SG-Former), towards effective global self-attention with adaptive fine granularity. At the heart of our approach is to utilize a significance map, which is estimated through hybrid-scale self-attention and evolves itself during training, to reallocate tokens based on the significance of each region. Intuitively, we assign more tokens to the salient regions for achieving fine-grained attention, while allocating fewer tokens to the minor regions in exchange for efficiency and global receptive fields. The proposed SG-Former achieves performance superior to state of the art: our base size model achieves 84.7% Top-1 accuracy on ImageNet-1K, 51.2mAP bbAP on CoCo, 52.7mIoU on ADE20K surpassing the Swin Transformer by +1.3% / +2.7 mAP/ +3 mIoU, with lower computation costs and fewer parameters. The code is available at https://github.com/OliverRensu/SG-Former

Exploring spatial-temporal dependencies from observed motions is one of the core challenges of human motion prediction. Previous methods mainly focus on dedicated network structures to model the spatial and temporal dependencies. This paper considers a new direction by introducing a model learning framework with auxiliary tasks. In our auxiliary tasks, partial body joints’ coordinates are corrupted by either masking or adding noise and the goal is to recover corrupted coordinates depending on the rest coordinates. To work with auxiliary tasks, we propose a novel auxiliary-adapted transformer, which can handle incomplete, corrupted motion data and achieve coordinate recovery via capturing spatial-temporal dependencies. Through auxiliary tasks, the auxiliary-adapted transformer is promoted to capture more comprehensive spatial-temporal dependencies among body joints’ coordinates, leading to better feature learning. Extensive experimental results have shown that our method outperforms state-of-the-art methods by remarkable margins of 7.2%, 3.7%, and 9.4% in terms of 3D mean per joint position error (MPJPE) on the Human3.6M, CMU Mocap, and 3DPW datasets, respectively. We also demonstrate that our method is more robust under data missing cases and noisy data cases. Code is available at https://github.com/MediaBrain-SJTU/AuxFormer.

Diffusion Probabilistic Models (DPMs) have recently demonstrated impressive results on various generative tasks. Despite its promises, the learned representations of pre-trained DPMs, however, have not been fully understood. In this paper, we conduct an in-depth investigation of the representation power of DPMs, and propose a novel knowledge transfer method that leverages the knowledge acquired by generative DPMs for recognition tasks. Our study begins by examining the feature space of DPMs, revealing that DPMs are inherently denoising autoencoders that balance the representation learning with regularizing model capacity. To this end, we introduce a novel knowledge transfer paradigm named RepFusion. Our paradigm extracts representations at different time steps from off-the-shelf DPMs and dynamically employs them as supervision for student networks, in which the optimal time is determined through reinforcement learning. We evaluate our approach on several image classification, semantic segmentation, and landmark detection benchmarks, and demonstrate that it outperforms state-of-the-art methods. Our results uncover the potential of DPMs as a powerful tool for representation learning and provide insights into the usefulness of generative models beyond sample generation. The code is available at https://github.com/Adamdad/Repfusion.

In Reinforcement Learning (RL), Laplacian Representation (LapRep) is a task-agnostic state representation that encodes the geometry of the environment. A desirable property of LapRep stated in prior works is that the Euclidean distance in the LapRep space roughly reflects the reachability between states, which motivates the usage of this distance for reward shaping. However, we find that LapRep does not necessarily have this property in general: two states having a small distance under LapRep can actually be far away in the environment. Such a mismatch would impede the learning process in reward shaping. To fix this issue, we introduce a Reachability-Aware Laplacian Representation (RA-LapRep), by properly scaling each dimension of LapRep. Despite the simplicity, we demonstrate that RA-LapRep can better capture the inter-state reachability as compared to LapRep, through both theoretical explanations and experimental results. Additionally, we show that this improvement yields a significant boost in reward shaping performance and benefits bottleneck state discovery.

Adapter-style efficient transfer learning (ETL) has shown excellent performance in the tuning of vision-language models (VLMs) under the low-data regime, where only a few additional parameters are introduced to excavate the task-specific knowledge based on the general and powerful representation of VLMs. However, most adapter-style works face two limitations: (i) modeling task-specific knowledge with a single modality only; and (ii) overlooking the exploitation of the inter-class relationships in downstream tasks, thereby leading to sub-optimal solutions. To mitigate that, we propose an effective adapter-style tuning strategy, dubbed GraphAdapter, which performs the textual adapter by explicitly modeling the dual-modality structure knowledge (i.e., the correlation of different semantics/classes in textual and visual modalities) with a dual knowledge graph. In particular, the dual knowledge graph is established with two sub-graphs, i.e., a textual knowledge sub-graph, and a visual knowledge sub-graph, where the nodes and edges represent the semantics/classes and their correlations in two modalities, respectively. This enables the textual feature of each prompt to leverage the task-specific structure knowledge from both textual and visual modalities, yielding a more effective classifier for downstream tasks. Extensive experimental results on 11 benchmark datasets reveal that our GraphAdapter significantly outperforms the previous adapter-based methods.

Generative modeling has recently undergone remarkable advancements, primarily propelled by the transformative implications of Diffusion Probabilistic Models (DPMs). The impressive capability of these models, however, often entails significant computational overhead during both training and inference. To tackle this challenge, we present Diff-Pruning, an efficient compression method tailored for learning lightweight diffusion models from pre-existing ones, without the need for extensive re-training. The essence of Diff-Pruning is encapsulated in a Taylor expansion over pruned timesteps, a process that disregards non-contributory diffusion steps and ensembles informative gradients to identify important weights. Our empirical assessment, undertaken across several datasets highlights two primary benefits of our proposed method: 1) Efficiency: it enables approximately a 50\% reduction in FLOPs at a mere 10% to 20% of the original training expenditure; 2) Consistency: the pruned diffusion models inherently preserve generative behavior congruent with their pre-trained models.

Vision-and-Language Navigation (VLN) is a challenging task that requires an agent to navigate through complex environments based on natural language instructions. In contrast to conventional approaches, which primarily focus on the spatial domain exploration, we propose a paradigm shift toward the Fourier domain. This alternative perspective aims to enhance visual-textual matching, ultimately improving the agent's ability to understand and execute navigation tasks based on the given instructions. In this study, we first explore the significance of high-frequency information in VLN and provide evidence that it is instrumental in bolstering visual-textual matching processes. Building upon this insight, we further propose a sophisticated and versatile Frequency-enhanced Data Augmentation (FDA) technique to improve the VLN model's capability of capturing critical high-frequency information. Specifically, this approach requires the agent to navigate in environments where only a subset of high-frequency visual information corresponds with the provided textual instructions, ultimately fostering the agent's ability to selectively discern and capture pertinent high-frequency features according to the given instructions. Promising results on R2R, RxR, CVDN and REVERIE demonstrate that our FDA can be readily integrated with existing VLN approaches, improving performance without adding extra parameters, and keeping models simple and efficient. The code is available at https://github.com/hekj/FDA.

Unsupervised domain adaptation (UDA) has been widely applied in improving model generalization on unlabeled target data. However, accurately selecting the best UDA model for the target domain is challenging due to the absence of labeled target data and domain distribution shifts. Traditional model selection approaches involve training extra models with source data to estimate the target validation risk. Recent studies propose practical methods that are based on measuring various properties of model predictions on target data. Although effective for some UDA models, these methods often lack stability and may lead to poor selections for other UDA models. In this paper, we present MixVal, an innovative model selection method that operates solely with unlabeled target data during inference. MixVal leverages mixed target samples with pseudo labels to directly probe the learned target structure by each UDA model. Specifically, MixVal employs two distinct types of probes: the intra-cluster mixed samples for evaluating neighborhood density and the inter-cluster mixed samples for investigating the classification boundary. With this comprehensive probing strategy, MixVal elegantly combines the strengths of two state-of-the-art model selection methods, Entropy and SND. We extensively evaluate MixVal on 11 UDA methods across 4 adaptation settings, including classification and segmentation tasks. Experimental results consistently demonstrate that MixVal achieves state-of-the-art performance and maintains exceptional stability in model selection. Code is available at \url{https://github.com/LHXXHB/MixVal}.

Existing dataset distillation (DD) techniques typically rely on iterative strategies to synthesize condensed datasets, where datasets before and after distillation are forward and backward through neural networks a massive number of times. Despite the promising results achieved, the time efficiency of prior approaches is still far from satisfactory. Moreover, when different sizes of synthetic datasets are required, they have to repeat the iterative training procedures, which is highly cumbersome and lacks flexibility. In this paper, different from the time-consuming forward-backward passes, we introduce a generative fashion for dataset distillation with significantly improved efficiency. Specifically, synthetic samples are produced by a generator network conditioned on the initialization of DD, while synthetic labels are obtained by solving a least-squares problem in a feature space. Our theoretical analysis reveals that the errors of synthetic datasets solved in the original space and then processed by any conditional generators are upper-bounded. To find a satisfactory generator efficiently, we propose a meta-learning algorithm, where a meta generator is trained on a large dataset so that only a few steps are required to adapt to a target dataset. The meta generator is termed as MGDD in our approach. Once adapted, it can handle arbitrary sizes of synthetic datasets, even for those unseen during adaptation. Experiments demonstrate that the generator adapted with only a limited number of steps performs on par with those state-of-the-art DD methods and yields $22\times$ acceleration.

Large language models (LLMs) have shown remarkable capabilities in language understanding and generation. However, such impressive capability typically comes with a substantial model size, which presents significant challenges in both the deployment, inference, and training stages. With LLM being a general-purpose task solver, we explore its compression in a task-agnostic manner, which aims to preserve the multi-task solving and language generation ability of the original LLM. One challenge to achieving this is the enormous size of the training corpus of LLM, which makes both data transfer and model post-training over-burdensome. Thus, we tackle the compression of LLMs within the bound of two constraints: being task-agnostic and minimizing the reliance on the original training dataset. Our method, named LLM-pruner, adopts structural pruning that selectively removes non-critical coupled structures based on gradient information, maximally preserving the majority of the LLM's functionality. To this end, the performance of pruned models can be efficiently recovered through tuning techniques, LoRA, in merely 3 hours, requiring only 50K data. We validate the LLM-Pruner on three LLMs, including LLaMA, Vicuna, and ChatGLM, and demonstrate that the compressed models still exhibit satisfactory capabilities in zero-shot classification and generation. The code will be made public.

In this paper, we make a bold attempt toward an ambitious task: given a pre-trained classifier, we aim to reconstruct an image generator, without relying on any data samples. From a black-box perspective, this challenge seems intractable, since it inevitably involves identifying the inverse function for a classifier, which is, by nature, an information extraction process. As such, we resort to leveraging the knowledge encapsulated within the parameters of the neural network. Grounded on the theory of Maximum-Margin Bias of gradient descent, we propose a novel learning paradigm, in which the generator is trained to ensure that the convergence conditions of the network parameters are satisfied over the generated distribution of the samples. Empirical validation from various image generation tasks substantiates the efficacy of our strategy.

MOTIVATION: Biological experimental approaches to protein-protein interaction (PPI) site prediction are critical for understanding the mechanisms of biochemical processes but are time-consuming and laborious. With the development of Deep Learning (DL) techniques, the most popular Convolutional Neural Networks (CNN)-based methods have been proposed to address these problems. Although significant progress has been made, these methods still have limitations in encoding the characteristics of each amino acid in protein sequences. Current methods cannot efficiently explore the nature of Position Specific Scoring Matrix (PSSM), secondary structure and raw protein sequences by processing them all together. For PPI site prediction, how to effectively model the PPI context with attention to prediction remains an open problem. In addition, the long-distance dependencies of PPI features are important, which is very challenging for many CNN-based methods because the innate ability of CNN is difficult to outperform auto-regressive models like Transformers. RESULTS: To effectively mine the properties of PPI features, a novel hybrid neural network named HN-PPISP is proposed, which integrates a Multi-layer Perceptron Mixer (MLP-Mixer) module for local feature extraction and a two-stage multi-branch module for global feature capture. The model merits Transformer, TextCNN and Bi-LSTM as a powerful alternative for PPI site prediction. On the one hand, this is the first application of an advanced Transformer (i.e. MLP-Mixer) with a hybrid network for sequence-based PPI prediction. On the other hand, unlike existing methods that treat global features altogether, the proposed two-stage multi-branch hybrid module firstly assigns different attention scores to the input features and then encodes the feature through different branch modules. In the first stage, different improved attention modules are hybridized to extract features from the raw protein sequences, secondary structure and PSSM, respectively. In the second stage, a multi-branch network is designed to aggregate information from both branches in parallel. The two branches encode the features and extract dependencies through several operations such as TextCNN, Bi-LSTM and different activation functions. Experimental results on real-world public datasets show that our model consistently achieves state-of-the-art performance over seven remarkable baselines. AVAILABILITY: The source code of HN-PPISP model is available at https://github.com/ylxu05/HN-PPISP.

Abstract Recently, existing FCNs‐based methods have shown their advantages in processing object boundaries. However, these methods still suffer from false object interference, which appears in saliency predictions. To solve this problem, an edge‐interior feature fusion (EIFF) framework is proposed, which consists of an internal‐boundary decoupled generation structure with receptive field enlargement and attention mechanism enhancement, and a salient feature refinement module. Specifically, the framework first learns edge features and interior features through an internal‐boundary decoupling generation network, which is supervised by labels obtained by decoupling ground‐truth through an image erosion algorithm. Then, feature refinement module (FRM) is designed to purify the coarse prediction by focusing on the ambiguous regions through a mining strategy to generate the final saliency map. To compensate for shortcomings of the BCE and IU loss, we also introduce a weighted loss to guide our model to focus more on the error‐prone parts. Experimental results on five benchmark datasets demonstrate that the proposed method performs favorably against 19 state‐of‐the‐art approaches under four standard metrics.

Existing 3D human pose estimation methods often suffer inferior generalization performance to new datasets, largely due to the limited diversity of 2D-3D pose pairs in the training data. To address this problem, we present PoseAug, a novel auto-augmentation framework that learns to augment the available training poses towards greater diversity and thus enhances the generalization power of the trained 2D-to-3D pose estimator. Specifically, PoseAug introduces a novel pose augmentor that learns to adjust various geometry factors of a pose through differentiable operations. With such differentiable capacity, the augmentor can be jointly optimized with the 3D pose estimator and take the estimation error as feedback to generate more diverse and harder poses in an online manner. PoseAug is generic and handy to be applied to various 3D pose estimation models. It is also extendable to aid pose estimation from video frames. To demonstrate this, we introduce PoseAug-V, a simple yet effective method that decomposes video pose augmentation into end pose augmentation and conditioned intermediate pose generation. Extensive experiments demonstrate that PoseAug and its extension PoseAug-V bring clear improvements for frame-based and video-based 3D pose estimation on several out-of-domain 3D human pose benchmarks.

Temporal graphs exhibit dynamic interactions between nodes over continuous time, whose topologies evolve with time elapsing. The whole temporal neighborhood of nodes reveals the varying preferences of nodes. However, previous works usually generate dynamic representation with limited neighbors for simplicity, which results in both inferior performance and high latency of online inference. Therefore, in this paper, we propose a novel method of temporal graph convolution with the whole neighborhood, namely Temporal Aggregation and Propagation Graph Neural Networks (TAP-GNN). Specifically, we first analyze the computational complexity of the dynamic representation problem by unfolding the temporal graph in a message-passing paradigm. The expensive complexity motivates us to design the AP (aggregation and propagation) block, which significantly reduces the repeated computation of historical neighbors. The final TAP-GNN supports online inference in the graph stream scenario, which incorporates the temporal information into node embeddings with a temporal activation function and a projection layer besides several AP blocks. Experimental results on various real-life temporal networks show that our proposed TAP-GNN outperforms existing temporal graph methods by a large margin in terms of both predictive performance and online inference latency.

Data-free knowledge distillation (DFKD) has recently been attracting increasing attention from research communities, attributed to its capability to compress a model only using synthetic data. Despite the encouraging results achieved, state-of-the-art DFKD methods still suffer from the inefficiency of data synthesis, making the data-free training process extremely time-consuming and thus inapplicable for large-scale tasks. In this work, we introduce an efficacious scheme, termed as FastDFKD, that allows us to accelerate DFKD by a factor of orders of magnitude. At the heart of our approach is a novel strategy to reuse the shared common features in training data so as to synthesize different data instances. Unlike prior methods that optimize a set of data independently, we propose to learn a meta-synthesizer that seeks common features as the initialization for the fast data synthesis. As a result, FastDFKD achieves data synthesis within only a few steps, significantly enhancing the efficiency of data-free training. Experiments over CIFAR, NYUv2, and ImageNet demonstrate that the proposed FastDFKD achieves 10x and even 100x acceleration while preserving performances on par with state of the art. Code is available at https://github.com/zju-vipa/Fast-Datafree.

Knowledge distillation (KD) has recently emerged as a powerful strategy to transfer knowledge from a pre-trained teacher model to a lightweight student, and has demonstrated its unprecedented success over a wide spectrum of applications. In spite of the encouraging results, the KD process \emph{per se} poses a potential threat to network ownership protection, since the knowledge contained in network can be effortlessly distilled and hence exposed to a malicious user. In this paper, we propose a novel framework, termed as Safe Distillation Box~(SDB), that allows us to wrap a pre-trained model in a virtual box for intellectual property protection. Specifically, SDB preserves the inference capability of the wrapped model to all users, but precludes KD from unauthorized users. For authorized users, on the other hand, SDB carries out a knowledge augmentation scheme to strengthen the KD performances and the results of the student model. In other words, all users may employ a model in SDB for inference, but only authorized users get access to KD from the model. The proposed SDB imposes no constraints over the model architecture, and may readily serve as a plug-and-play solution to protect the ownership of a pre-trained network. Experiments across various datasets and architectures demonstrate that, with SDB, the performance of an unauthorized KD drops significantly while that of an authorized gets enhanced, demonstrating the effectiveness of SDB.

Jinming Zhao, Tenggan Zhang, Jingwen Hu, Yuchen Liu, Qin Jin, Xinchao Wang, Haizhou Li. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Preserving geometric structures in the scene plays a vital role in image stitching. However, most of the existing methods ignore the large-scale layouts reflected by straight lines or curves, decreasing overall stitching quality. To address this issue, this work presents a structure-preserving stitching approach that produces images with natural visual effects and less distortion. Our method first employs deep learning-based edge detection to extract various types of large-scale edges. Then, the extracted edges are sampled to construct multiple groups of triangles to represent geometric structures. Meanwhile, a GEometric Structure preserving (GES) energy term is introduced to make these triangles undergo similarity transformation. Further, an optimized GES energy term is presented to reasonably determine the weights of the sampling points on the geometric structure, and the term is added into the Global Similarity Prior (GSP) stitching model called GES-GSP to achieve a smooth transition between local alignment and geometric structure preservation. The effectiveness of GES-GSP is validated through comprehensive experiments on a stitching dataset. The experimental results show that the proposed method outperforms several state-of-the-art methods in geometric structure preservation and obtains more natural stitching results. The code and dataset are available at https://github.com/flowerDuo/GES-GSP-Stitching.

Existing self-supervised 3D human pose estimation schemes have largely relied on weak supervisions like consistency loss to guide the learning, which, inevitably, leads to inferior results in real-world scenarios with unseen poses. In this paper, we propose a novel self-supervised approach that allows us to explicitly generate 2D-3D pose pairs for augmenting supervision, through a self-enhancing dual-loop learning framework. This is made possible via introducing a reinforcement-learning-based imitator, which is learned jointly with a pose estimator alongside a pose hallucinator; the three components form two loops during the training process, complementing and strengthening one another. Specifically, the pose estimator transforms an input 2D pose sequence to a low-fidelity 3D output, which is then enhanced by the imitator that enforces physical constraints. The refined 3D poses are subsequently fed to the hallucinator for producing even more diverse data, which are, in turn, strengthened by the imitator and further utilized to train the pose estimator. Such a co-evolution scheme, in practice, enables training a pose estimator on self-generated motion data without relying on any given 3D data. Extensive experiments across various benchmarks demonstrate that our approach yields encouraging results significantly outperforming the state of the art and, in some cases, even on par with results of fully-supervised methods. Notably, it achieves 89.1% 3D PCK on MPI-INF-3DHP under self-supervised cross-dataset evaluation setup, improving upon the previous best self-supervised method [16], [26] by 8.6%.

Vanilla unsupervised domain adaptation methods tend to optimize the model with fixed neural architecture, which is not very practical in real-world scenarios since the target data is usually processed by different resource-limited devices. It is therefore of great necessity to facilitate architecture adaptation across various devices. In this paper, we introduce a simple framework, Slimmable Domain Adaptation, to improve cross-domain generalization with a weight-sharing model bank, from which models of different capacities can be sampled to accommodate different accuracy-efficiency trade-offs. The main challenge in this frame-work lies in simultaneously boosting the adaptation performance of numerous models in the model bank. To tackle this problem, we develop a Stochastic EnsEmble Distillation method to fully exploit the complementary knowledge in the model bank for inter-model interaction. Nevertheless, considering the optimization conflict between inter-model interaction and intra-model adaptation, we augment the existing bi-classifier domain confusion architecture into an Optimization-Separated Tri-Classifier counterpart. After optimizing the model bank, architecture adaptation is leveraged via our proposed Unsupervised Performance Evaluation Metric. Under various resource constraints, our framework surpasses other competing approaches by a very large margin on multiple benchmarks. It is also worth emphasizing that our framework can preserve the performance improvement against the source-only model even when the computing complexity is reduced to 1/64. Code will be available at https://github.com/HIK-LAB/SlimDA.

Recent Vision Transformer (ViT) models have demonstrated encouraging results across various computer vision tasks, thanks to its competence in modeling long-range de-pendencies of image patches or tokens via self-attention. These models, however, usually designate the similar receptive fields of each token feature within each layer. Such a constraint inevitably limits the ability of each self-attention layer in capturing multi-scale features, thereby leading to performance degradation in handling images with multiple objects of different scales. To address this issue, we propose a novel and generic strategy, termed shunted self-attention (SSA), that allows ViTs to model the attentions at hybrid scales per attention layer. The key idea of SSA is to inject heterogeneous receptive field sizes into tokens: before computing the self-attention matrix, it selectively merges tokens to represent larger object features while keeping certain tokens to preserve fine-grained features. This novel merging scheme enables the self-attention to learn relationships between objects with different sizes, and simultaneously reduces the token numbers and the computational cost. Extensive experiments across various tasks demonstrate the superiority of SSA. Specifically, the SSA-based transformer achieve 84.0% Top-1 accuracy and out-performs the state-of-the-art Focal Transformer on Ima-geNet with only half of the model size and computation cost, and surpasses Focal Transformer by 1.3 mAP on COCO and 2.9 mIOU on ADE20K under similar parameter and computation cost. Code has been released at https://github.com/OliverRensulShunted-Transformer.

Dataset condensation aims at reducing the network training effort through condensing a cumbersome training set into a compact synthetic one. State-of-the-art approaches largely rely on learning the synthetic data by matching the gradients between the real and synthetic data batches. Despite the intuitive motivation and promising results, such gradient-based methods, by nature, easily overfit to a biased set of samples that produce dominant gradients, and thus lack a global supervision of data distribution. In this paper, we propose a novel scheme to Condense dataset by Aligning FEatures (CAFE), which explicitly attempts to preserve the real-feature distribution as well as the discriminant power of the resulting synthetic set, lending itself to strong generalization capability to various architectures. At the heart of our approach is an effective strategy to align features from the real and synthetic data across various scales, while accounting for the classification of real samples. Our scheme is further backed up by a novel dynamic bi-level optimization, which adaptively adjusts parameter updates to prevent over-/under-fitting. We validate the proposed CAFE across various datasets, and demonstrate that it generally outperforms the state of the art: on the SVHN dataset, for example, the performance gain is up to 11%. Extensive experiments and analysis verify the effectiveness and necessity of proposed designs.

We present a simple and effective framework, named Point2Seq, for 3D object detection from point clouds. In contrast to previous methods that normally predict attributes of 3D objects all at once, we expressively model the interdependencies between attributes of 3D objects, which in turn enables a better detection accuracy. Specifically, we view each 3D object as a sequence of words and reformulate the 3D object detection task as decoding words from 3D scenes in an auto-regressive manner. We further propose a lightweight scene-to-sequence decoder that can auto-regressively generate words conditioned on features from a 3D scene as well as cues from the preceding words. The predicted words eventually constitute a set of sequences that completely describe the 3D objects in the scene, and all the predicted sequences are then automatically assigned to the respective ground truths through similarity-based sequence matching. Our approach is conceptually intuitive and can be readily plugged upon most existing 3D-detection backbones without adding too much computational over-head; the sequential decoding paradigm we proposed, on the other hand, can better exploit information from complex 3D scenes with the aid of preceding predicted words. Without bells and whistles, our method significantly out-performs previous anchor- and center-based 3D object detection frameworks, yielding the new state of the art on the challenging ONCE dataset as well as the Waymo Open Dataset. Code is available at https://github.com/ocNflag/point2seq.

Transformers have shown great potential in computer vision tasks. A common belief is their attention-based token mixer module contributes most to their competence. However, recent works show the attention-based module in transformers can be replaced by spatial MLPs and the resulted models still perform quite well. Based on this observation, we hypothesize that the general architecture of the transformers, instead of the specific token mixer module, is more essential to the model's performance. To verify this, we deliberately replace the attention module in transformers with an embarrassingly simple spatial pooling operator to conduct only basic token mixing. Surprisingly, we observe that the derived model, termed as PoolFormer, achieves competitive performance on multiple computer vision tasks. For example, on ImageNet-1K, PoolFormer achieves 82.1 % top-1 accuracy, surpassing well-tuned vision transformer/MLP-like baselines DeiT-B/ResMLP-B24 by 0.3%/1.1% accuracy with 35%/52% fewer parameters and 49%/61% fewer MACs. The effectiveness of Pool-Former verifies our hypothesis and urges us to initiate the concept of “MetaFormer”, a general architecture abstracted from transformers without specifying the token mixer. Based on the extensive experiments, we argue that MetaFormer is the key player in achieving superior results for recent transformer and MLP-like models on vision tasks. This work calls for more future research dedicated to improving MetaFormer instead of focusing on the token mixer modules. Additionally, our proposed PoolFormer could serve as a starting baseline for future MetaFormer architecture design.

Text-based video segmentation aims to segment the target object in a video based on a describing sentence. Incorporating motion information from optical flow maps with appearance and linguistic modalities is crucial yet has been largely ignored by previous work. In this paper, we design a method to fuse and align appearance, motion, and linguistic features to achieve accurate segmentation. Specifically, we propose a multi-modal video transformer, which can fuse and aggregate multi-modal and temporal features between frames. Furthermore, we design a language-guided feature fusion module to progressively fuse appearance and motion features in each feature level with guidance from linguistic features. Finally, a multi-modal alignment loss is proposed to alleviate the semantic gap between features from different modalities. Extensive experiments on A2D Sentences and J-HMDB Sentences verify the performance and the generalization ability of our method compared to the state-of-the-art methods.

Multimodal emotion recognition study is hindered by the lack of labelled corpora in terms of scale and diversity, due to the high annotation cost and label ambiguity. In this paper, we propose a multimodal pre-training model MEmoBERT for multimodal emotion recognition, which learns multimodal joint representations through self-supervised learning from a self-collected large-scale unlabeled video data that come in sheer volume. Furthermore, unlike the conventional "pre-train, finetune" paradigm, we propose a prompt-based method that reformulates the downstream emotion classification task as a masked text prediction one, bringing the downstream task closer to the pre-training. Extensive experiments on two benchmark datasets, IEMOCAP and MSP-IMPROV, show that our proposed MEmoBERT significantly enhances emotion recognition performance.

Self-training for unsupervised domain adaptive object detection is a challenging task, of which the performance depends heavily on the quality of pseudo boxes. Despite the promising results, prior works have largely overlooked the uncertainty of pseudo boxes during self-training. In this paper, we present a simple yet effective framework, termed as Probabilistic Teacher (PT), which aims to capture the uncertainty of unlabeled target data from a gradually evolving teacher and guides the learning of a student in a mutually beneficial manner. Specifically, we propose to leverage the uncertainty-guided consistency training to promote classification adaptation and localization adaptation, rather than filtering pseudo boxes via an elaborate confidence threshold. In addition, we conduct anchor adaptation in parallel with localization adaptation, since anchor can be regarded as a learnable parameter. Together with this framework, we also present a novel Entropy Focal Loss (EFL) to further facilitate the uncertainty-guided self-training. Equipped with EFL, PT outperforms all previous baselines by a large margin and achieve new state-of-the-arts.

Data-free knowledge distillation (DFKD) conducts knowledge distillation via eliminating the dependence of original training data, and has recently achieved impressive results in accelerating pre-trained language models. At the heart of DFKD is to reconstruct a synthetic dataset by inverting the parameters of the uncompressed model. Prior DFKD approaches, however, have largely relied on hand-crafted priors of the target data distribution for the reconstruction, which can be inevitably biased and often incompetent to capture the intrinsic distributions. To address this problem, we propose a prompt-based method, termed as PromptDFD, that allows us to take advantage of learned language priors, which effectively harmonizes the synthetic sentences to be semantically and grammatically correct. Specifically, PromptDFD leverages a pre-trained generative model to provide language priors and introduces a reinforced topic prompter to control data synthesis, making the generated samples thematically relevant and semantically plausible, and thus friendly to downstream tasks. As shown in our experiments, the proposed method substantially improves the synthesis quality and achieves considerable improvements on distillation performance. In some cases, PromptDFD even gives rise to results on par with those from the data-driven knowledge distillation with access to the original training data.

Unsupervised Person Re-identification (U-ReID) with pseudo labeling recently reaches a competitive performance compared to fully-supervised ReID methods based on modern clustering algorithms. However, such clustering-based scheme becomes computationally prohibitive for large-scale datasets, making it infeasible to be applied in real-world application. How to efficiently leverage endless unlabeled data with limited computing resources for better U-ReID is under-explored. In this paper, we make the first attempt to the large-scale U-ReID and propose a "small data for big task" paradigm dubbed Meta Clustering Learning (MCL). MCL only pseudo-labels a subset of the entire unlabeled data via clustering to save computing for the first-phase training. After that, the learned cluster centroids, termed as meta-prototypes in our MCL, are regarded as a proxy annotator to softly annotate the rest unlabeled data for further polishing the model. To alleviate the potential noisy labeling issue in the polishment phase, we enforce two well-designed loss constraints to promise intra-identity consistency and inter-identity strong correlation. For multiple widely-used U-ReID benchmarks, our method significantly saves computational cost while achieving a comparable or even better performance compared to prior works.

Existing fine-tuning methods either tune all parameters of the pre-trained model (full fine-tuning), which is not efficient, or only tune the last linear layer (linear probing), which suffers a significant accuracy drop compared to the full fine-tuning. In this paper, we propose a new parameter-efficient fine-tuning method termed as SSF, representing that researchers only need to Scale and Shift the deep Features extracted by a pre-trained model to catch up with the performance of full fine-tuning. In this way, SSF also surprisingly outperforms other parameter-efficient fine-tuning approaches even with a smaller number of tunable parameters. Furthermore, different from some existing parameter-efficient fine-tuning methods (e.g., Adapter or VPT) that introduce the extra parameters and computational cost in the training and inference stages, SSF only adds learnable parameters during the training stage, and these additional parameters can be merged into the original pre-trained model weights via re-parameterization in the inference phase. With the proposed SSF, our model obtains 2.46% (90.72% vs. 88.54%) and 11.48% (73.10% vs. 65.57%) performance improvement on FGVC and VTAB-1k in terms of Top-1 accuracy compared to the full fine-tuning but only fine-tuning about 0.3M parameters. We also conduct amounts of experiments in various model families (CNNs, Transformers, and MLPs) and datasets. Results on 26 image classification datasets in total and 3 robustness & out-of-distribution datasets show the effectiveness of SSF. Code is available at https://github.com/dongzelian/SSF.

In this paper, we study dataset distillation (DD), from a novel perspective and introduce a \emph{dataset factorization} approach, termed \emph{HaBa}, which is a plug-and-play strategy portable to any existing DD baseline. Unlike conventional DD approaches that aim to produce distilled and representative samples, \emph{HaBa} explores decomposing a dataset into two components: data \emph{Ha}llucination networks and \emph{Ba}ses, where the latter is fed into the former to reconstruct image samples. The flexible combinations between bases and hallucination networks, therefore, equip the distilled data with exponential informativeness gain, which largely increase the representation capability of distilled datasets. To furthermore increase the data efficiency of compression results, we further introduce a pair of adversarial contrastive \xw{constraints} on the resultant hallucination networks and bases, which increase the diversity of generated images and inject more discriminant information into the factorization. Extensive comparisons and experiments demonstrate that our method can yield significant improvement on downstream classification tasks compared with previous state of the arts, while reducing the total number of compressed parameters by up to 65\%. Moreover, distilled datasets by our approach also achieve \textasciitilde10\% higher accuracy than baseline methods in cross-architecture generalization. Our code is available \href{https://github.com/Huage001/DatasetFactorization}{here}.

Recent studies show that transformer has strong capability of building long-range dependencies, yet is incompetent in capturing high frequencies that predominantly convey local information. To tackle this issue, we present a novel and general-purpose $\textit{Inception Transformer}$, or $\textit{iFormer}$ for short, that effectively learns comprehensive features with both high- and low-frequency information in visual data. Specifically, we design an Inception mixer to explicitly graft the advantages of convolution and max-pooling for capturing the high-frequency information to transformers. Different from recent hybrid frameworks, the Inception mixer brings greater efficiency through a channel splitting mechanism to adopt parallel convolution/max-pooling path and self-attention path as high- and low-frequency mixers, while having the flexibility to model discriminative information scattered within a wide frequency range. Considering that bottom layers play more roles in capturing high-frequency details while top layers more in modeling low-frequency global information, we further introduce a frequency ramp structure, i.e., gradually decreasing the dimensions fed to the high-frequency mixer and increasing those to the low-frequency mixer, which can effectively trade-off high- and low-frequency components across different layers. We benchmark the iFormer on a series of vision tasks, and showcase that it achieves impressive performance on image classification, COCO detection and ADE20K segmentation. For example, our iFormer-S hits the top-1 accuracy of 83.4% on ImageNet-1K, much higher than DeiT-S by 3.6%, and even slightly better than much bigger model Swin-B (83.3%) with only 1/4 parameters and 1/3 FLOPs. Code and models are released at https://github.com/sail-sg/iFormer.

Spiking neural networks (SNNs) are shown to be more biologically plausible and energy efficient over their predecessors. However, there is a lack of an efficient and generalized training method for deep SNNs, especially for deployment on analog computing substrates. In this paper, we put forward a generalized learning rule, termed Local Tandem Learning (LTL). The LTL rule follows the teacher-student learning approach by mimicking the intermediate feature representations of a pre-trained ANN. By decoupling the learning of network layers and leveraging highly informative supervisor signals, we demonstrate rapid network convergence within five training epochs on the CIFAR-10 dataset while having low computational complexity. Our experimental results have also shown that the SNNs thus trained can achieve comparable accuracies to their teacher ANNs on CIFAR-10, CIFAR-100, and Tiny ImageNet datasets. Moreover, the proposed LTL rule is hardware friendly. It can be easily implemented on-chip to perform fast parameter calibration and provide robustness against the notorious device non-ideality issues. It, therefore, opens up a myriad of opportunities for training and deployment of SNN on ultra-low-power mixed-signal neuromorphic computing chips.

In this paper, we explore a novel knowledge-transfer task, termed as Deep Model Reassembly (DeRy), for general-purpose model reuse.Given a collection of heterogeneous models pre-trained from distinct sources and with diverse architectures, the goal of DeRy, as its name implies, is to first dissect each model into distinctive building blocks, and then selectively reassemble the derived blocks to produce customized networks under both the hardware resource and performance constraints. Such ambitious nature of DeRy inevitably imposes significant challenges, including, in the first place, the feasibility of its solution. We strive to showcase that, through a dedicated paradigm proposed in this paper, DeRy can be made not only possibly but practically efficiently. Specifically, we conduct the partitions of all pre-trained networks jointly via a cover set optimization, and derive a number of equivalence set, within each of which the network blocks are treated as functionally equivalent and hence interchangeable. The equivalence sets learned in this way, in turn, enable picking and assembling blocks to customize networks subject to certain constraints, which is achieved via solving an integer program backed up with a training-free proxy to estimate the task performance. The reassembled models give rise to gratifying performances with the user-specified constraints satisfied. We demonstrate that on ImageNet, the best reassemble model achieves 78.6% top-1 accuracy without fine-tuning, which could be further elevated to 83.2% with end-to-end fine-tuning. Our code is available at https://github.com/Adamdad/DeRy.

Recent Generative Adversarial Networks (GAN) models deliver visually authentic images and find their applications in various applications. Due to their heavy size, however, it remains cumbersome to deploy state-of-the-art GAN on the edge side, such as mobile terminals. In this paper, we propose a practical and very effective approach for distilling GAN models, so as to produce competent student models that come with smaller sizes. Unlike other convolutional neural network models whose final output spans a small or reasonably-sized space, based on which knowledge distillation is carried out, GAN produces images as output that span an enormous space, making the distillation much more challenging. To this end, we propose an effective dual-step distillation approach tailored for GAN. Instead of directly enforcing the student output distribution to imitate that of the teacher, we partition the generated images into two sets, the “realer” one and the “fake” one, based on the image quality, and sequentially minimize the discrepancy first between fake images of the student and the teacher, and then between realer ones. This seemingly simple strategy turns out highly effective in terms of exploring output space structure and further the knowledge transfer. Experimental results have shown that the proposed method can be easily generalized and demonstrate the learned portable generator produces competitive or even better results both in quality and quantity.

The goal of image steganography is to hide a full-sized image, termed secret, into another, termed cover. Prior image steganography algorithms can conceal only one secret within one cover. In this paper, we propose an adaptive local image steganography (AdaSteg) system that allows for scale- and location-adaptive image steganography. By adaptively hiding the secret on a local scale, the proposed system makes the steganography more secured, and further enables multi-secret steganography within one single cover. Specifically, this is achieved via two stages, namely the adaptive patch selection stage and secret encryption stage. Given a pair of secret and cover, first, the optimal local patch for concealment is determined adaptively by exploiting deep reinforcement learning with the proposed steganography quality function and policy network. The secret image is then converted into a patch of encrypted noises, resembling the process of generating adversarial examples, which are further encoded to a local region of the cover to realize a more secured steganography. Furthermore, we propose a novel criterion for the assessment of local steganography, and also collect a challenging dataset that is specialized for the task of image steganography, thus contributing to a standardized benchmark for the area. Experimental results demonstrate that the proposed model yields results superior to the state of the art in both security and capacity.

Video highlight detection (VHD) has emerged as a pressing task due to the unprecedentedly increasing amount of video data, such as those from e-commerce live-broadcasting platforms. Many approaches focus on exploiting text data, in the form of video description or time-sync comments, to facilitate the VHD task. Despite the promising results, they have largely overlooked the noises inherent in the text data and have mostly relied on isolating the feature of text and video. In this paper, we introduce a novel model to handle VHD, termed Coevolution Network (CoEvo-Net), that allows us to account for joint learning of the language and video features explicitly via a coevolution paradigm, in which features from the two data modalities progressively refine each other. This is achieved by a dedicated CoEvo-Cell that takes language and video together as inputs, extracts cross-modality, and filters the undesired parts of the input, such as words in a sentence. Furthermore, we release a large-scale dataset of e-commerce for VHD, in which each video is coupled with a sentence for description, to benchmark the sentence-based VHD approaches. Extensive experiments on the released dataset demonstrate that CoEvo-Net achieves state-of-the-art performance. Our dataset and code will be made publicly available.

Recent image-generation methods have demonstrated that realistic images can be produced from captions. Despite the promising results achieved, existing caption-based generation methods confront a dilemma. On the one hand, the image generator should be provided with sufficient details for realistic hallucination, meaning that longer sentences with rich content are preferred, but on the other hand, the generator is meanwhile fragile to long sentences due to their complex semantics and syntax like long-range dependencies and the combinatorial explosion of object visual features. Toward alleviating this dilemma, a novel approach is proposed in this article to hallucinate images from attribute pairs, which can be extracted from natural language processing (NLP) toolsets in the presence of complex semantics and syntax. Attribute pairs, therefore, enable our image generator to tackle long sentences handily and alleviate the combinatorial explosion, and at the same time, allow us to enlarge the training dataset and to produce hallucinations from randomly combined attribute pairs at ease. Experiments on widely used datasets demonstrate that the proposed approach yields results superior to the state of the art.

Heterogeneous graph embedding aims at learning low-dimensional representations from a graph featuring nodes and edges of diverse natures, and meanwhile preserving the underlying topology. Existing approaches along this line have largely relied onmeta-paths, which are by nature hand-crafted and pre-defined transition rules, so as to explore the semantics of a graph. Despite the promising results, defining meta-paths requires domain knowledge, and thus when the test distribution deviates from the priors, such methods are prone to errors. In this paper, we propose a self-learning scheme for heterogeneous graph embedding, termed as self-guided walk (SILK), that bypasses meta-paths and learns adaptive attentions for node walking. SILK assumes no prior knowledge or annotation is provided, and conducts a customized random walk to encode the contexts of the heterogeneous graph of interest. Specifically, this is achieved via maintaining a dynamically-updatedguidance matrixthat records the node-conditioned transition potentials. Experimental results on four real-world datasets demonstrate that SILK significantly outperforms state-of-the-art methods.

When confronted with objects of unknown types in an image, humans can effortlessly and precisely tell their visual boundaries. This recognition mechanism and underlying generalization capability seem to contrast to state-of-the-art image segmentation networks that rely on large-scale category-aware annotated training samples. In this paper, we make an attempt towards building models that explicitly account for visual boundary knowledge, in hope to reduce the training effort on segmenting unseen categories. Specifically, we investigate a new task termed as Boundary Knowledge Translation (BKT). Given a set of fully labeled categories, BKT aims to translate the visual boundary knowledge learned from the labeled categories, to a set of novel categories, each of which is provided only a few labeled samples. To this end, we propose a Translation Segmentation Network (Trans-Net), which comprises a segmentation network and two boundary discriminators. The segmentation network, combined with a boundary-aware self-supervised mechanism, is devised to conduct foreground segmentation, while the two discriminators work together in an adversarial manner to ensure an accurate segmentation of the novel categories under light supervision. Exhaustive experiments demonstrate that, with only tens of labeled samples as guidance, Trans-Net achieves close results on par with fully supervised methods.

The microvascular invasion (MVI) is a major prognostic factor in hepatocellular carcinoma, which is one of the malignant tumors with the highest mortality rate. The diagnosis of MVI needs discovering the vessels that contain hepatocellular carcinoma cells and counting their number in each vessel, which depends heavily on experiences of the doctor, is largely subjective and time-consuming. However, there is no algorithm as yet tailored for the MVI detection from pathological images. This paper collects the first pathological liver image dataset containing $522$ whole slide images with labels of vessels, MVI, and hepatocellular carcinoma grades. The first and essential step for the automatic diagnosis of MVI is the accurate segmentation of vessels. The unique characteristics of pathological liver images, such as super-large size, multi-scale vessel, and blurred vessel edges, make the accurate vessel segmentation challenging. Based on the collected dataset, we propose an Edge-competing Vessel Segmentation Network (EVS-Net), which contains a segmentation network and two edge segmentation discriminators. The segmentation network, combined with an edge-aware self-supervision mechanism, is devised to conduct vessel segmentation with limited labeled patches. Meanwhile, two discriminators are introduced to distinguish whether the segmented vessel and background contain residual features in an adversarial manner. In the training stage, two discriminators are devised to compete for the predicted position of edges. Exhaustive experiments demonstrate that, with only limited labeled patches, EVS-Net achieves a close performance of fully supervised methods, which provides a convenient tool for the pathological liver vessel segmentation. Code is publicly available at https://github.com/wang97zh/EVS-Net.

Data mixing augmentation has proved effective in training deep models. Recent methods mix labels mainly according to the mixture proportion of image pixels. Due to the major discriminative information of a fine-grained image usually resides in subtle regions, these methods tend to introduce heavy label noise in fine-grained recognition. We propose Semantically Proportional Mixing (SnapMix) that exploits class activation map (CAM) to lessen the label noise in augmenting fine-grained data. SnapMix generates the target label for a mixed image by estimating its intrinsic semantic composition. This strategy can adapt to asymmetric mixing operations and ensure semantic correspondence between synthetic images and target labels. Experiments show that our method consistently outperforms existing mixed-based approaches regardless of different datasets or network depths. Further, by incorporating the mid-level features, the proposed SnapMix achieves top-level performance, demonstrating its potential to serve as a strong baseline for fine-grained recognition.

Catastrophic forgetting refers to the tendency that a neural network ``forgets'' the previous learned knowledge upon learning new tasks. Prior methods have been focused on overcoming this problem on convolutional neural networks (CNNs), where the input samples like images lie in a grid domain, but have largely overlooked graph neural networks (GNNs) that handle non-grid data. In this paper, we propose a novel scheme dedicated to overcoming catastrophic forgetting problem and hence strengthen continual learning in GNNs. At the heart of our approach is a generic module, termed as topology-aware weight preserving (TWP), applicable to arbitrary form of GNNs in a plug-and-play fashion. Unlike the main stream of CNN-based continual learning methods that rely on solely slowing down the updates of parameters important to the downstream task, TWP explicitly explores the local structures of the input graph, and attempts to stabilize the parameters playing pivotal roles in the topological aggregation. We evaluate TWP on different GNN backbones over several datasets, and demonstrate that it yields performances superior to the state of the art. Code is publicly available at https://github.com/hhliu79/TWP.

Knowledge distillation has demonstrated encouraging performances in deep model compression. Most existing approaches, however, require massive labeled data to accomplish the knowledge transfer, making the model compression a cumbersome and costly process. In this paper, we investigate the practical few-shot knowledge distillation scenario, where we assume only a few samples without human annotations are available for each category. To this end, we introduce a principled dual-stage distillation scheme tailored for few-shot data. In the first step, we graft the student blocks one by one onto the teacher, and learn the parameters of the grafted block intertwined with those of the other teacher blocks. In the second step, the trained student blocks are progressively connected and then together grafted onto the teacher network, allowing the learned student blocks to adapt themselves to each other and eventually replace the teacher network. Experiments demonstrate that our approach, with only a few unlabeled samples, achieves gratifying results on CIFAR10,CIFAR100, and ILSVRC-2012. On CIFAR10 and CIFAR100, our performances are even on par with those of knowledge distillation schemes that utilize the full datasets. The source code is available at https://github.com/zju-vipa/NetGraft.

Modern online multiple object tracking (MOT) methods usually focus on two directions to improve tracking performance. One is to predict new positions in an incoming frame based on tracking information from previous frames, and the other is to enhance data association by generating more discriminative identity embeddings. Some works combined both directions within one framework but handled them as two individual tasks, thus gaining little mutual benefits. In this paper, we propose a novel unified model with synergy between position prediction and embedding association. The two tasks are linked by temporal-aware target attention and distractor attention, as well as identity-aware memory aggregation model. Specifically, the attention modules can make the prediction focus more on targets and less on distractors, therefore more reliable embeddings can be extracted accordingly for association. On the other hand, such reliable embeddings can boost identity-awareness through memory aggregation, hence strengthen attention modules and suppress drifts. In this way, the synergy between position prediction and embedding association is achieved, which leads to strong robustness to occlusions. Extensive experiments demonstrate the superiority of our proposed model over a wide range of existing methods on MOTChallenge benchmarks. Our code and models are publicly available at https://github.com/songguocode/TADAM.

State-of-the-art video super-resolution (VSR) methods focus on exploiting inter- and intra-frame correlations to estimate high-resolution (HR) video frames from low-resolution (LR) ones. In this paper, we study VSR from an exotic perspective, by explicitly looking into the role of temporal frequency of video frames. Through experiments, we observe that a higher frequency, and hence a smaller pixel displacement between consecutive frames, tends to de-liver favorable super-resolved results. This discovery motivates us to introduce Event Cameras, a novel sensing de-vice that responds instantly to pixel intensity changes and produces up to millions of asynchronous events per second, to facilitate VSR. To this end, we propose an Event-based VSR framework (E-VSR), of which the key component is an asynchronous interpolation (EAI) module that reconstructs a high-frequency (HF) video stream with uniform and tiny pixel displacements between neighboring frames from an event stream. The derived HF video stream is then encoded into a VSR module to recover the desired HR videos. Furthermore, an LR bi-directional interpolation loss and an HR self-supervision loss are also introduced to respectively regulate the EAI and VSR modules. Experiments on both real-world and synthetic datasets demonstrate that the proposed approach yields results superior to the state of the art.

In this paper, we study a novel knowledge transfer task in the domain of graph neural networks (GNNs). We strive to train a multi-talented student GNN, without accessing human annotations, that “amalgamates” knowledge from a couple of teacher GNNs with heterogeneous architectures and handling distinct tasks. The student derived in this way is expected to integrate the expertise from both teachers while maintaining a compact architecture. To this end, we propose an innovative approach to train a slimmable GNN that enables learning from teachers with varying feature dimensions. Meanwhile, to explicitly align topological semantics between the student and teachers, we introduce a topological attribution map (TAM) to highlight the structural saliency in a graph, based on which the student imitates the teachers’ ways of aggregating information from neighbors. Experiments on seven datasets across various tasks, including multi-label classification and joint segmentation-classification, demonstrate that the learned student, with a lightweight architecture, achieves gratifying results on par with and sometimes even superior to those of the teachers in their specializations. Our code is publicly available at https://github.com/ycjing/AmalgamateGNN.PyTorch.

What scene elements, if any, are indispensable for recognizing a scene? We strive to answer this question through the lens of an exotic learning scheme. Our goal is to identify a collection of such pivotal elements, which we term as Scene Essence, to be those that would alter scene recognition if taken out from the scene. To this end, we devise a novel approach that learns to partition the scene objects into two groups, essential ones and minor ones, under the supervision that if only the essential ones are kept while the minor ones are erased in the input image, a scene recognizer would preserve its original prediction. Specifically, we introduce a learnable graph neural network (GNN) for labelling scene objects, based on which the minor ones are wiped off by an off-the-shelf image inpainter. The features of the inpainted image derived in this way, together with those learned from the GNN with the minor-object nodes pruned, are expected to fool the scene discriminator. Both subjective and objective evaluations on Places365, SUN397, and MIT67 datasets demonstrate that, the learned Scene Essence yields a visually plausible image that convincingly retains the original scene category.

Generative Adversarial Networks (GANs) have demonstrated unprecedented success in various image generation tasks. The encouraging results, however, come at the price of a cumbersome training process, during which the generator and discriminator are alternately updated in two stages. In this paper, we investigate a general training scheme that enables training GANs efficiently in only one stage. Based on the adversarial losses of the generator and discriminator, we categorize GANs into two classes, Symmetric GANs and Asymmetric GANs, and introduce a novel gradient decomposition method to unify the two, allowing us to train both classes in one stage and hence alleviate the training effort. We also computationally analyze the efficiency of the proposed method, and empirically demonstrate that, the proposed method yields a solid 1.5× acceleration across various datasets and network architectures. Furthermore, we show that the proposed method is readily applicable to other adversarial-training scenarios, such as data-free knowledge distillation. The code is available at https://github.com/zju-vipa/OSGAN.

Knowledge distillation pursues a diminutive yet well-behaved student network by harnessing the knowledge learned by a cumbersome teacher model. Prior methods achieve this by making the student imitate shallow behaviors, such as soft targets, features, or attention, of the teacher. In this paper, we argue that what really matters for distillation is the intrinsic problem-solving process captured by the teacher. By dissecting the decision process in a layer-wise manner, we found that the decision-making procedure in the teacher model is conducted in a coarse-to-fine manner, where coarse-grained discrimination (e.g., animal vs vehicle) is attained in early layers, and fine-grained dis-crimination (e.g., dog vs cat, car vs truck) in latter layers. Motivated by this observation, we propose a new distillation method, dubbed as Tree-like Decision Distillation (TDD), to endow the student with the same problem-solving mechanism as that of the teacher. Extensive experiments demonstrated that TDD yields competitive performance compared to state of the arts. More importantly, it enjoys better interpretability due to its interpretable decision distillation instead of dark knowledge distillation.

Multi-person pose estimation and tracking serve as crucial steps for video understanding. Most state-of-the-art approaches rely on first estimating poses in each frame and only then implementing data association and refinement. Despite the promising results achieved, such a strategy is inevitably prone to missed detections especially in heavily-cluttered scenes, since this tracking-by-detection paradigm is, by nature, largely dependent on visual evidences that are absent in the case of occlusion. In this paper, we propose a novel online approach to learning the pose dynamics, which are independent of pose detections in current fame, and hence may serve as a robust estimation even in challenging scenarios including occlusion. Specifically, we derive this prediction of dynamics through a graph neural network (GNN) that explicitly accounts for both spatial-temporal and visual information. It takes as input the historical pose tracklets and directly predicts the corresponding poses in the following frame for each tracklet. The predicted poses will then be aggregated with the detected poses, if any, at the same frame so as to produce the final pose, potentially recovering the occluded joints missed by the estimator. Experiments on PoseTrack 2017 and Pose-Track 2018 datasets demonstrate that the proposed method achieves results superior to the state of the art on both human pose estimation and tracking tasks.

Neural trees aim at integrating deep neural networks and decision trees so as to bring the best of the two worlds, including representation learning from the former and faster inference from the latter. In this paper, we introduce a novel approach, termed as Self-born Wiring (SeBoW), to learn neural trees from a mother deep neural network. In contrast to prior neural-tree approaches that either adopt a pre-defined structure or grow hierarchical layers in a progressive manner, task-adaptive neural trees in SeBoW evolve from a deep neural network through a construction-by-destruction process, enabling a global-level parameter optimization that further yields favorable results. Specifically, given a designated network configuration like VGG, SeBoW disconnects all the layers and derives isolated filter groups, based on which a global-level wiring process is conducted to attach a subset of filter groups, eventually bearing a lightweight neural tree. Extensive experiments demonstrate that, with a lower computational cost, SeBoW outperforms all prior neural trees by a significant margin and even achieves results on par with predominant non-tree networks like ResNets. Moreover, SeBoW proves its scalability to large-scale datasets like ImageNet, which has been barely explored by prior tree networks.

Detection and segmentation of nuclei are fundamental analysis operations in pathology images, the assessments derived from which serve as the gold standard for cancer diagnosis. Manual segmenting nuclei is expensive and time-consuming. What’s more, accurate segmentation detection of nuclei can be challenging due to the large appearance variation, conjoined and overlapping nuclei, and serious degeneration of histological structures. Supervised methods highly rely on massive annotated samples. The existing two unsupervised methods are prone to failure on degenerated samples. This paper proposes a Mutual-Complementing Framework (MCF) for nuclei detection and segmentation in pathology images. Two branches of MCF are trained in the mutual-complementing manner, where the detection branch complements the pseudo mask of the segmentation branch, while the progressive trained segmentation branch complements the missing nucleus templates through calculating the mask residual between the predicted mask and detected result. In the detection branch, two response map fusion strategies and gradient direction based postprocessing are devised to obtain the optimal detection response. Furthermore, the confidence loss combined with the synthetic samples and self-finetuning is adopted to train the segmentation network with only high confidence areas. Extensive experiments demonstrate that MCF achieves comparable performance with only a few nucleus patches as supervision. Especially, MCF possesses good robustness (only dropping by about 6%) on degenerated samples, which are critical and common cases in clinical diagnosis.

Learning mid-level representation for fine-grained recognition is easily dominated by a limited number of highly discriminative patterns, degrading its robustness and generalization capability. To this end, we propose a novel Stochastic Partial Swap (SPS)1scheme to address this issue. Our method performs element-wise swapping for partial features between samples to inject noise during training. It equips a regularization effect similar to Dropout, which promotes more neurons to represent the concepts. Furthermore, it also exhibits other advantages: 1) suppressing over-activation to some part patterns to improve feature representativeness, and 2) enriching pattern combination and simulating noisy cases to enhance classifier generalization. We verify the effectiveness of our approach through comprehensive experiments across four network backbones and three fine-grained datasets. Moreover, we demonstrate its ability to complement high-level representations, allowing a simple model to achieve performance comparable to the top-performing technologies in fine-grained recognition, indoor scene recognition, and material recognition while improving model interpretability.

In this paper, we study a novel meta aggregation scheme towards binarizing graph neural networks (GNNs). We begin by developing a vanilla 1-bit GNN framework that binarizes both the GNN parameters and the graph features. Despite the lightweight architecture, we observed that this vanilla framework suffered from insufficient discriminative power in distinguishing graph topologies, leading to a dramatic drop in performance. This discovery motivates us to devise meta aggregators to improve the expressive power of vanilla binarized GNNs, of which the aggregation schemes can be adaptively changed in a learnable manner based on the binarized features. Towards this end, we propose two dedicated forms of meta neighborhood aggregators, an exclusive meta aggregator termed as Greedy Gumbel Neighborhood Aggregator (GNA), and a diffused meta aggregator termed as Adaptable Hybrid Neighborhood Aggregator (ANA). GNA learns to exclusively pick one single optimal aggregator from a pool of candidates, while ANA learns a hybrid aggregation behavior to simultaneously retain the benefits of several individual aggregators. Furthermore, the proposed meta aggregators may readily serve as a generic plugin module into existing full-precision GNNs. Experiments across various domains demonstrate that the proposed method yields results superior to the state of the art.

Given a reference object of an unknown type in an image, human observers can effortlessly find the objects of the same category in another image and precisely tell their visual boundaries. Such visual cognition capability of humans seems absent from the current research spectrum of computer vision. Existing segmentation networks, for example, rely on a humongous amount of labeled data, which is laborious and costly to collect and annotate; besides, the performance of segmentation networks tend to downgrade as the number of the category increases. In this paper, we introduce a novel Reference semantic segmentation Network (Ref-Net) to conduct visual boundary knowledge translation. Ref-Net contains a Reference Segmentation Module (RSM) and a Boundary Knowledge Translation Module (BKTM). Inspired by the human recognition mechanism, RSM is devised only to segment the same category objects based on the features of the reference objects. BKTM, on the other hand, introduces two boundary discriminator branches to conduct inner and outer boundary segmentation of the target object in an adversarial manner, and translate the annotated boundary knowledge of open-source datasets into the segmentation network. Exhaustive experiments demonstrate that, with tens of finely-grained annotated samples as guidance, Ref-Net achieves results on par with fully supervised methods on six datasets. Our code can be found in the supplementary material.

Model inversion, whose goal is to recover training data from a pre-trained model, has been recently proved feasible. However, existing inversion methods usually suffer from the mode collapse problem, where the synthesized instances are highly similar to each other and thus show limited effectiveness for downstream tasks, such as knowledge distillation. In this paper, we propose Contrastive Model Inversion (CMI), where the data diversity is explicitly modeled as an optimizable objective, to alleviate the mode collapse issue. Our main observation is that, under the constraint of the same amount of data, higher data diversity usually indicates stronger instance discrimination. To this end, we introduce in CMI a contrastive learning objective that encourages the synthesizing instances to be distinguishable from the already synthesized ones in previous batches. Experiments of pre-trained models on CIFAR-10, CIFAR-100, and Tiny-ImageNet demonstrate that CMI not only generates more visually plausible instances than the state of the arts, but also achieves significantly superior performance when the generated data are used for knowledge distillation. Code is available at https://github.com/zju-vipa/DataFree.

Knowledge distillation (KD) has recently emerged as an efficacious scheme for learning compact deep neural networks (DNNs). Despite the promising results achieved, the rationale that interprets the behavior of KD has yet remained largely understudied. In this paper, we introduce a novel task-oriented attention model, termed as KDExplainer, to shed light on the working mechanism underlying the vanilla KD. At the heart of KDExplainer is a Hierarchical Mixture of Experts (HME), in which a multi-class classification is reformulated as a multi-task binary one. Through distilling knowledge from a free-form pre-trained DNN to KDExplainer, we observe that KD implicitly modulates the knowledge conflicts between different subtasks, and in reality has much more to offer than label smoothing. Based on such findings, we further introduce a portable tool, dubbed as virtual attention module (VAM), that can be seamlessly integrated with various DNNs to enhance their performance under KD. Experimental results demonstrate that with a negligible additional cost, student models equipped with VAM consistently outperform their non-VAM counterparts across different benchmarks. Furthermore, when combined with other KD methods, VAM remains competent in promoting results, even though it is only motivated by vanilla KD. The code is available at https:// github.com/zju-vipa/KDExplainer.

Knowledge distillation~(KD) aims to craft a compact student model that imitates the behavior of a pre-trained teacher in a target domain. Prior KD approaches, despite their gratifying results, have largely relied on the premise that \emph{in-domain} data is available to carry out the knowledge transfer. Such an assumption, unfortunately, in many cases violates the practical setting, since the original training data or even the data domain is often unreachable due to privacy or copyright reasons. In this paper, we attempt to tackle an ambitious task, termed as \emph{out-of-domain} knowledge distillation~(OOD-KD), which allows us to conduct KD using only OOD data that can be readily obtained at a very low cost. Admittedly, OOD-KD is by nature a highly challenging task due to the agnostic domain gap. To this end, we introduce a handy yet surprisingly efficacious approach, dubbed as~\textit{MosaicKD}. The key insight behind MosaicKD lies in that, samples from various domains share common local patterns, even though their global semantic may vary significantly; these shared local patterns, in turn, can be re-assembled analogous to mosaic tiling, to approximate the in-domain data and to further alleviating the domain discrepancy. In MosaicKD, this is achieved through a four-player min-max game, in which a generator, a discriminator, a student network, are collectively trained in an adversarial manner, partially under the guidance of a pre-trained teacher. We validate MosaicKD over {classification and semantic segmentation tasks} across various benchmarks, and demonstrate that it yields results much superior to the state-of-the-art counterparts on OOD data. Our code is available at \url{https://github.com/zju-vipa/MosaicKD}.

In this paper, we propose a novel unsupervised approach for sequence matching by explicitly accounting for the locality properties in the sequences. In contrast to conventional approaches that rely on frame-to-frame matching, we conduct matching using sequencelet or seqlet, a sub-sequence wherein the frames share strong similarities and are thus grouped together. The optimal seqlets and matching between them are learned jointly, without any supervision from users. The learned seqlets preserve the locality information at the scale of interest and resolve the ambiguities during matching, which are omitted by frame-based matching methods. We show that our proposed approach outperforms the state-of-the-art ones on datasets of different domains including human actions, facial expressions, speech, and character strokes.

State-of-the-art multi-object tracking (MOT) methods follow the tracking-by-detection paradigm, where object trajectories are obtained by associating per-frame outputs of object detectors. In crowded scenes, however, detectors often fail to obtain accurate detections due to heavy occlusions and high crowd density. In this paper, we propose a new MOT paradigm, tracking-by-counting, tailored for crowded scenes. Using crowd density maps, we jointly model detection, counting, and tracking of multiple targets as a network flow program, which simultaneously finds the global optimal detections and trajectories of multiple targets over the whole video. This is in contrast to prior MOT methods that either ignore the crowd density and thus are prone to errors in crowded scenes, or rely on a suboptimal two-step process using heuristic density-aware point-tracks for matching targets. Our approach yields promising results on public benchmarks of various domains including people tracking, cell tracking, and fish tracking.

Prior normalization methods rely on affine transformations to produce arbitrary image style transfers, of which the parameters are computed in a pre-defined way. Such manually-defined nature eventually results in the high-cost and shared encoders for both style and content encoding, making style transfer systems cumbersome to be deployed in resource-constrained environments like on the mobile-terminal side. In this paper, we propose a new and generalized normalization module, termed as Dynamic Instance Normalization (DIN), that allows for flexible and more efficient arbitrary style transfers. Comprising an instance normalization and a dynamic convolution, DIN encodes a style image into learnable convolution parameters, upon which the content image is stylized. Unlike conventional methods that use shared complex encoders to encode content and style, the proposed DIN introduces a sophisticated style encoder, yet comes with a compact and lightweight content encoder for fast inference. Experimental results demonstrate that the proposed approach yields very encouraging results on challenging style patterns and, to our best knowledge, for the first time enables an arbitrary style transfer using MobileNet-based lightweight architecture, leading to a reduction factor of more than twenty in computational cost as compared to existing approaches. Furthermore, the proposed DIN provides flexible support for state-of-the-art convolutional operations, and thus triggers novel functionalities, such as uniform-stroke placement for non-natural images and automatic spatial-stroke control.

Lip reading has witnessed unparalleled development in recent years thanks to deep learning and the availability of large-scale datasets. Despite the encouraging results achieved, the performance of lip reading, unfortunately, remains inferior to the one of its counterpart speech recognition, due to the ambiguous nature of its actuations that makes it challenging to extract discriminant features from the lip movement videos. In this paper, we propose a new method, termed as Lip by Speech (LIBS), of which the goal is to strengthen lip reading by learning from speech recognizers. The rationale behind our approach is that the features extracted from speech recognizers may provide complementary and discriminant clues, which are formidable to be obtained from the subtle movements of the lips, and consequently facilitate the training of lip readers. This is achieved, specifically, by distilling multi-granularity knowledge from speech recognizers to lip readers. To conduct this cross-modal knowledge distillation, we utilize an efficacious alignment scheme to handle the inconsistent lengths of the audios and videos, as well as an innovative filtering strategy to refine the speech recognizer's prediction. The proposed method achieves the new state-of-the-art performance on the CMLR and LRS2 datasets, outperforming the baseline by a margin of 7.66% and 2.75% in character error rate, respectively.

Exploring the intrinsic interconnections between the knowledge encoded in PRe-trained Deep Neural Networks (PR-DNNs) of heterogeneous tasks sheds light on their mutual transferability, and consequently enables knowledge transfer from one task to another so as to reduce the training effort of the latter. In this paper, we propose the DEeP Attribution gRAph (DEPARA) to investigate the transferability of knowledge learned from PR-DNNs. In DEPARA, nodes correspond to the inputs and are represented by their vectorized attribution maps with regards to the outputs of the PR-DNN. Edges denote the relatedness between inputs and are measured by the similarity of their features extracted from the PR-DNN. The knowledge transferability of two PR-DNNs is measured by the similarity of their corresponding DEPARAs. We apply DEPARA to two important yet under-studied problems in transfer learning: pre-trained model selection and layer selection. Extensive experiments are conducted to demonstrate the effectiveness and superiority of the proposed method in solving both these problems. Code, data and models reproducing the results in this paper are available at https://github.com/zju-vipa/DEPARA.

Existing knowledge distillation methods focus on convolutional neural networks (CNNs), where the input samples like images lie in a grid domain, and have largely overlooked graph convolutional networks (GCN) that handle non-grid data. In this paper, we propose to our best knowledge the first dedicated approach to distilling knowledge from a pre-trained GCN model. To enable the knowledge transfer from the teacher GCN to the student, we propose a local structure preserving module that explicitly accounts for the topological semantics of the teacher. In this module, the local structure information from both the teacher and the student are extracted as distributions, and hence minimizing the distance between these distributions enables topology-aware knowledge transfer from the teacher, yielding a compact yet high-performance student model. Moreover, the proposed approach is readily extendable to dynamic graph models, where the input graphs for the teacher and the student may differ. We evaluate the proposed method on two different datasets using GCN models of different architectures, and demonstrate that our method achieves the state-of-the-art knowledge distillation performance for GCN models.

Recent advances in deep learning have provided procedures for learning one network to amalgamate multiple streams of knowledge from the pre-trained Convolutional Neural Network (CNN) models, thus reduce the annotation cost. However, almost all existing methods demand massive training data, which may be unavailable due to privacy or transmission issues. In this paper, we propose a data-free knowledge amalgamate strategy to craft a well-behaved multi-task student network from multiple single/multi-task teachers. The main idea is to construct the group-stack generative adversarial networks (GANs) which have two dual generators. First one generator is trained to collect the knowledge by reconstructing the images approximating the original dataset utilized for pre-training the teachers. Then a dual generator is trained by taking the output from the former generator as input. Finally we treat the dual part generator as the target network and regroup it. As demonstrated on several benchmarks of multi-label classification, the proposed method without any training data achieves the surprisingly competitive results, even compared with some full-supervised methods.

High-fidelity face completion is a challenging task due to the rich and subtle facial textures involved. What makes it more complicated is the correlations between different facial components, for example, the symmetry in texture and structure between both eyes. While recent works adopted the attention mechanism to learn the contextual relations among elements of the face, they have largely overlooked the disastrous impacts of inaccurate attention scores; in addition, they fail to pay sufficient attention to key facial components, the completion results of which largely determine the authenticity of a face image. Accordingly, in this paper, we design a comprehensive framework for face completion based on the U-Net structure. Specifically, we propose a dual spatial attention module to efficiently learn the correlations between facial textures at multiple scales; moreover, we provide an oracle supervision signal to the attention module to ensure that the obtained attention scores are reasonable. Furthermore, we take the location of the facial components as prior knowledge and impose a multi-discriminator on these regions, with which the fidelity of facial components is significantly promoted. Extensive experiments on two high-resolution face datasets including CelebA-HQ and Flickr-Faces-HQ demonstrate that the proposed approach outperforms state-of-the-art methods by large margins.

Event camera is a kind of bio-inspired sensor which is able to capture the motion in asynchronous events stream. An event is triggered when the pixel has a brightness change. In spatio-temporal space, those events will form an event cloud, which has specific 3D geometry to capture the dynamic scene. To analyze the event cloud, previous works usually convert event streams into frame-based images which did not fully utilize its 3D geometry in the spatio-temporal event space. In this work, we propose to recognize the spatio-temporal 3D event clouds for gesture recognition using Dynamic Graph CNN (DGCNN) which directly takes 3D points as input and is successfully used for 3D object recognition. We adapt DGCNN to perform action recognition by recognizing 3D geometry features in spatio-temporal space of the event data. We achieve state-of-the-art accuracy of 98.56% on the IBM DVS128 Gesture dataset and 95.94% on the DHP19 dataset.

The ability to disassemble the features of objects and background is crucial for many machine learning tasks, including image classification, image editing, visual concepts learning, and so on. However, existing (semi-)supervised methods all need a large amount of annotated samples, while unsupervised methods can't handle real-world images with complicated backgrounds. In this paper, we introduce the One-sample Guided Object Representation Disassembling (One-GORD) method, which only requires one annotated sample for each object category to learn disassembled object representation from unannotated images. For the annotated one-sample, we first adopt some data augmentation strategies to generate some synthetic samples, which can guide the disassembling of the object features and background features. For the unannotated images, two self-supervised mechanisms: dual-swapping and fuzzy classification are introduced to disassemble object features from the background with the guidance of annotated one-sample. What's more, we devise two metrics to evaluate the disassembling performance from the perspective of representation and image, respectively. Experiments demonstrate that the One-GORD achieves competitive dissembling performance and can handle natural scenes with complicated backgrounds.

Graphs have been widely adopted to denote structural connections between entities. The relations are in many cases heterogeneous, but entangled together and denoted merely as a single edge between a pair of nodes. For example, in a social network graph, users in different latent relationships like friends and colleagues, are usually connected via a bare edge that conceals such intrinsic connections. In this paper, we introduce a novel graph convolutional network (GCN), termed as factorizable graph convolutional network (FactorGCN), that explicitly disentangles such intertwined relations encoded in a graph. FactorGCN takes a simple graph as input, and disentangles it into several factorized graphs, each of which represents a latent and disentangled relation among nodes. The features of the nodes are then aggregated separately in each factorized latent space to produce disentangled features, which further leads to better performances for downstream tasks. We evaluate the proposed FactorGCN both qualitatively and quantitatively on the synthetic and real-world datasets, and demonstrate that it yields truly encouraging results in terms of both disentangling and feature aggregation. Code is publicly available at https://github.com/ihollywhy/FactorGCN.PyTorch.

Human parsing and matting play important roles in various applications, such as dress collocation, clothing recommendation, and image editing. In this paper, we propose a lightweight hybrid model that unifies the fully-supervised hierarchical-granularity parsing task and the unsupervised matting one. Our model comprises two parts, the extensible hierarchical semantic segmentation block using CNN and the matting module composed of guided filters. Given a human image, the segmentation block stage-1 first obtains a primitive segmentation map to separate the human and the background. The primitive segmentation is then fed into stage-2 together with the original image to give a rough segmentation of human body. This procedure is repeated in the stage-3 to acquire a refined segmentation. The matting module takes as input the above estimated segmentation maps and produces the matting map, in a fully unsupervised manner. The obtained matting map is then in turn fed back to the CNN in the first block for refining the semantic segmentation results.

Despite having achieved excellent performance on various tasks, deep neural networks have been shown to be susceptible to adversarial examples, i.e., visual inputs crafted with structural imperceptible noise. To explain this phenomenon, previous works implicate the weak capability of the classification models and the difficulty of the classification tasks. These explanations appear to account for some of the empirical observations but lack deep insight into the intrinsic nature of adversarial examples, such as the generation method and transferability. Furthermore, previous works generate adversarial examples completely rely on a specific classifier (model). Consequently, the attack ability of adversarial examples is strongly dependent on the specific classifier. More importantly, adversarial examples cannot be generated without a trained classifier. In this paper, we raise a question: what is the real cause of the generation of adversarial examples? To answer this question, we propose a new concept, called the adversarial region, which explains the existence of adversarial examples as perturbations perpendicular to the tangent plane of the data manifold. This view yields a clear explanation of the transfer property across different models of adversarial examples. Moreover, with the notion of the adversarial region, we propose a novel target-free method to generate adversarial examples via principal component analysis. We verify our adversarial region hypothesis on a synthetic dataset and demonstrate through extensive experiments on real datasets that the adversarial examples generated by our method have competitive or even strong transferability compared with model-dependent adversarial example generating methods. Moreover, our experiment shows that the proposed method is more robust to defensive methods than previous methods.

Class imbalance has emerged as one of the major challenges for medical image segmentation. The model cascade (MC) strategy, a popular scheme, significantly alleviates the class imbalance issue via running a set of individual deep models for coarse-to-fine segmentation. Despite its outstanding performance, however, this method leads to undesired system complexity and also ignores the correlation among the models. To handle these flaws in the MC approach, we propose in this paper a light-weight deep model, i.e., the One-pass Multi-task Network (OM-Net) to solve class imbalance better than MC does, while requiring only one-pass computation for brain tumor segmentation. First, OM-Net integrates the separate segmentation tasks into one deep model, which consists of shared parameters to learn joint features, as well as task-specific parameters to learn discriminative features. Second, to more effectively optimize OM-Net, we take advantage of the correlation among tasks to design both an online training data transfer strategy and a curriculum learning-based training strategy. Third, we further propose sharing prediction results between tasks, which enables us to design a cross-task guided attention (CGA) module. By following the guidance of the prediction results provided by the previous task, CGA can adaptively recalibrate channel-wise feature responses based on the category-specific statistics. Finally, a simple yet effective post-processing method is introduced to refine the segmentation results of the proposed attention network. Extensive experiments are conducted to demonstrate the effectiveness of the proposed techniques. Most impressively, we achieve state-of-the-art performance on the BraTS 2015 testing set and BraTS 2017 online validation set. Using these proposed approaches, we also won joint third place in the BraTS 2018 challenge among 64 participating teams.The code will be made publicly available at https://github.com/chenhong-zhou/OM-Net.

Despite the remarkable progress in person re-identification (Re-ID), such approaches still suffer from the failure cases where the discriminative body parts are missing. To mitigate this type of failure, we propose a simple yet effective Horizontal Pyramid Matching (HPM) approach to fully exploit various partial information of a given person, so that correct person candidates can be identified even if some key parts are missing. With HPM, we make the following contributions to produce more robust feature representations for the Re-ID task: 1) we learn to classify using partial feature representations at different horizontal pyramid scales, which successfully enhance the discriminative capabilities of various person parts; 2) we exploit average and max pooling strategies to account for person-specific discriminative information in a global-local manner. To validate the effectiveness of our proposed HPM method, extensive experiments are conducted on three popular datasets including Market-1501, DukeMTMCReID and CUHK03. Respectively, we achieve mAP scores of 83.1%, 74.5% and 59.7% on these challenging benchmarks, which are the new state-of-the-arts.

With the rapid development of deep learning, there have been an unprecedentedly large number of trained deep network models available online. Reusing such trained models can significantly reduce the cost of training the new models from scratch, if not infeasible at all as the annotations used for the training original networks are often unavailable to public. We propose in this paper to study a new model-reusing task, which we term as knowledge amalgamation. Given multiple trained teacher networks, each of which specializes in a different classification problem, the goal of knowledge amalgamation is to learn a lightweight student model capable of handling the comprehensive classification. We assume no other annotations except the outputs from the teacher models are available, and thus focus on extracting and amalgamating knowledge from the multiple teachers. To this end, we propose a pilot two-step strategy to tackle the knowledge amalgamation task, by learning first the compact feature representations from teachers and then the network parameters in a layer-wise manner so as to build the student model. We apply this approach to four public datasets and obtain very encouraging results: even without any human annotation, the obtained student model is competent to handle the comprehensive classification task and in most cases outperforms the teachers in individual sub-tasks.

What can we tell from a single motion-blurred image? We show in this paper that a 3D scene can be revealed. Unlike prior methods that focus on producing a deblurred image, we propose to estimate and take advantage of the hidden message of a blurred image, the relative motion trajectory, to restore the 3D scene collapsed during the exposure process. To this end, we train a deep network that jointly predicts the motion trajectory, the deblurred image, and the depth one, all of which in turn form a collaborative and self-supervised cycle that supervise one another to reproduce the input blurred image, enabling plausible 3D scene reconstruction from a single blurred image. We test the proposed model on several large-scale datasets we constructed based on benchmarks, as well as real-world blurred images, and show that it yields very encouraging quantitative and qualitative results.

In this paper, we investigate a novel deep-model reusing task. Our goal is to train a lightweight and versatile student model, without human-labelled annotations, that amalgamates the knowledge and masters the expertise of two pre-trained teacher models working on heterogeneous problems, one on scene parsing and the other on depth estimation. To this end, we propose an innovative training strategy that learns the parameters of the student intertwined with the teachers, achieved by ``projecting'' its amalgamated features onto each teacher's domain and computing the loss. We also introduce two options to generalize the proposed training strategy to handle three or more tasks simultaneously. The proposed scheme yields very encouraging results. As demonstrated on several benchmarks, the trained student model achieves results even superior to those of the teachers in their own expertise domains and on par with the state-of-the-art fully supervised models relying on human-labelled annotations.

A massive number of well-trained deep networks have been released by developers online. These networks may focus on different tasks and in many cases are optimized for different datasets. In this paper, we study how to exploit such heterogeneous pre-trained networks, known as teachers, so as to train a customized student network that tackles a set of selective tasks defined by the user. We assume no human annotations are available, and each teacher may be either single- or multi-task. To this end, we introduce a dual-step strategy that first extracts the task-specific knowledge from the heterogeneous teachers sharing the same sub-task, and then amalgamates the extracted knowledge to build the student network. To facilitate the training, we employ a selective learning scheme where, for each unlabelled sample, the student learns adaptively from only the teacher with the least prediction ambiguity. We evaluate the proposed approach on several datasets and the experimental results demonstrate that the student, learned by such adaptive knowledge amalgamation, achieves performances even better than those of the teachers.

Not all the human body parts have the same degree of freedom (DOF) due to the physiological structure. For example, the limbs may move more flexibly and freely than the torso does. Most of the existing 3D pose estimation methods, despite the very promising results achieved, treat the body joints equally and consequently often lead to larger reconstruction errors on the limbs. In this paper, we propose a progressive approach that explicitly accounts for the distinct DOFs among the body parts. We model parts with higher DOFs like the elbows, as dependent components of the corresponding parts with lower DOFs like the torso, of which the 3D locations can be more reliably estimated. Meanwhile, the high-DOF parts may, in turn, impose a constraint on where the low-DOF ones lie. As a result, parts with different DOFs supervise one another, yielding physically constrained and plausible pose-estimation results. To further facilitate the prediction of the high-DOF parts, we introduce a pose-attribution estimation, where the relative location of a limb joint with respect to the torso, which has the least DOF of a human body, is explicitly estimated and further fed to the joint-estimation module. The proposed approach achieves very promising results, outperforming the state of the art on several benchmarks.

An increasing number of well-trained deep networks have been released online by researchers and developers, enabling the community to reuse them in a plug-and-play way without accessing the training annotations. However, due to the large number of network variants, such public-available trained models are often of different architectures, each of which being tailored for a specific task or dataset. In this paper, we study a deep-model reusing task, where we are given as input pre-trained networks of heterogeneous architectures specializing in distinct tasks, as teacher models. We aim to learn a multitalented and light-weight student model that is able to grasp the integrated knowledge from all such heterogeneous-structure teachers, again without accessing any human annotation. To this end, we propose a common feature learning scheme, in which the features of all teachers are transformed into a common space and the student is enforced to imitate them all so as to amalgamate the intact knowledge. We test the proposed approach on a list of benchmarks and demonstrate that the learned student is able to achieve very promising performance, superior to those of the teachers in their specialized tasks.

Graph convolutional networks (GCN) have recently demonstrated their potential in analyzing non-grid structure data that can be represented as graphs. The core idea is to encode the local topology of a graph, via convolutions, into the feature of a center node. In this paper, we propose a novel GCN model, which we term as Shortest Path Graph Attention Network (SPAGAN). Unlike conventional GCN models that carry out node-based attentions, on either first-order neighbors or random higher-order ones, the proposed SPAGAN conducts path-based attention that explicitly accounts for the influence of a sequence of nodes yielding the minimum cost, or shortest path, between the center node and its higher-order neighbors. SPAGAN therefore allows for a more informative and intact exploration of the graph structure and further the more effective aggregation of information from distant neighbors, as compared to node-based GCN methods. We test SPAGAN for the downstream classification task on several standard datasets, and achieve performances superior to the state of the art.

Many well-trained Convolutional Neural Network~(CNN) models have now been released online by developers for the sake of effortless reproducing. In this paper, we treat such pre-trained networks as teachers and explore how to learn a target student network for customized tasks, using multiple teachers that handle different tasks. We assume no human-labelled annotations are available, and each teacher model can be either single- or multi-task network, where the former is a degenerated case of the latter. The student model, depending on the customized tasks, learns the related knowledge filtered from the multiple teachers, and eventually masters the complete or a subset of expertise from all teachers. To this end, we adopt a layer-wise training strategy, which entangles the student's network block to be learned with the corresponding teachers. As demonstrated on several benchmarks, the learned student network achieves very promising results, even outperforming the teachers on the customized tasks.

Exploring the transferability between heterogeneous tasks sheds light on their intrinsic interconnections, and consequently enables knowledge transfer from one task to another so as to reduce the training effort of the latter. In this paper, we propose an embarrassingly simple yet very efficacious approach to estimating the transferability of deep networks, especially those handling vision tasks. Unlike the seminal work of \emph{taskonomy} that relies on a large number of annotations as supervision and is thus computationally cumbersome, the proposed approach requires no human annotations and imposes no constraints on the architectures of the networks. This is achieved, specifically, via projecting deep networks into a \emph{model space}, wherein each network is treated as a point and the distances between two points are measured by deviations of their produced attribution maps. The proposed approach is several-magnitude times faster than taskonomy, and meanwhile preserves a task-wise topological structure highly similar to the one obtained by taskonomy. Code is available at \url{https://github.com/zju-vipa/TransferbilityFromAttributionMaps}.

3D human pose estimation from monocular images has become a heated area in computer vision recently. For years, most deep neural network based practices have adopted either an end-to-end approach, or a two-stage approach. An end-to-end network typically estimates 3D human poses directly from 2D input images, but it suffers from the shortage of 3D human pose data. It is also obscure to know if the inaccuracy stems from limited visual under-standing or 2D-to-3D mapping. Whereas a two-stage directly lifts those 2D keypoint outputs to the 3D space, after utilizing an existing network for 2D keypoint detections. However, they tend to ignore some useful contextual hints from the 2D raw image pixels. In this paper, we introduce a two-stage architecture that can eliminate the main disadvantages of both these approaches. During the first stage we use an existing state-of-the-art detector to estimate 2D poses. To add more con-textual information to help lifting 2D poses to 3D poses, we propose 3D Part Affinity Fields (3D-PAFs). We use 3D-PAFs to infer 3D limb vectors, and combine them with 2D poses to regress the 3D coordinates. We trained and tested our proposed framework on Human3.6M, the most popular 3D human pose benchmark dataset. Our approach achieves the state-of-the-art performance, which proves that with right selections of contextual information, a simple regression model can be very powerful in estimating 3D poses.

This paper performs a comprehensive and comparative evaluation of the state-of-the-art local features for the task of image-based 3D reconstruction. The evaluated local features cover the recently developed ones by using powerful machine learning techniques and the elaborately designed handcrafted features. To obtain a comprehensive evaluation, we choose to include both float type features and binary ones. Meanwhile, two kinds of datasets have been used in this evaluation. One is a dataset of many different scene types with groundtruth 3D points, containing images of different scenes captured at fixed positions, for quantitative performance evaluation of different local features in the controlled image capturing situation. The other dataset contains Internet scale image sets of several landmarks with a lot of unrelated images, which is used for qualitative performance evaluation of different local features in the free image collection situation. Our experimental results show that binary features are competent to reconstruct scenes from controlled image sequences with only a fraction of processing time compared to using float type features. However, for the case of a large scale image set with many distracting images, float type features show a clear advantage over binary ones. Currently, the most traditional SIFT is very stable with regard to scene types in this specific task and produces very competitive reconstruction results among all the evaluated local features. Meanwhile, although the learned binary features are not as competitive as the handcrafted ones, learning float type features with CNN is promising but still requires much effort in the future.

Multitask learning aims to improve the performance on related tasks by exploring the interdependence among them. Existing multitask learning methods explore the relatedness among tasks on the basis of the input features and the model parameters. In this paper, we focus on nonparametric multitask learning and propose to measure task relatedness from a novel perspective in a reproducing kernel Hilbert space (RKHS). Past works have shown that the objective function for a given task can be approximated using the top eigenvalues and corresponding eigenfunctions of a predefined integral operator on an RKHS. In our method, we formulate our objective for multitask learning as a linear combination of two sets of eigenfunctions, common eigenfunctions shared by different tasks and unique eigenfunctions in individual tasks, such that the eigenfunctions for one task can provide additional information on another and help to improve its performance. We present both theoretical and empirical validations of our proposed approach. The theoretical analysis demonstrates that our learning algorithm is uniformly argument stable and that the convergence rate of the generalization upper bound can be improved by learning multiple tasks. Experiments on several benchmark multitask learning data sets show that our method yields promising results.

Localizing text in the wild is challenging in the situations of complicated geometric layout of the targets like random orientation and large aspect ratio. In this paper, we propose a geometry-aware modeling approach tailored for scene text representation with an end-to-end learning scheme. In our approach, a novel Instance Transformation Network (ITN) is presented to learn the geometry-aware representation encoding the unique geometric configurations of scene text instances with in-network transformation embedding, resulting in a robust and elegant framework to detect words or text lines at one pass. An end-to-end multi-task learning strategy with transformation regression, text/non-text classification and coordinates regression is adopted in the ITN. Experiments on the benchmark datasets demonstrate the effectiveness of the proposed approach in detecting scene text in various geometric configurations.

This paper presents Botzone, a competitive platform for game AI education and research. It aims to simplify the teaching process of game AI courses, inspire learners to self-study, and acting as a dataset for game AI research. This platform is a universal online multi-agent game AI platform, designed to evaluate different implementations of game AI by applying them to agents in a variety of games and compete with each other, featuring an ELO ranking system and a contest system for users to evaluate their AI programs. It has been successfully used in various AI competitions and courses in practice, and has the extensibility to support more games and languages, as well as further usages such as studying machine learning on game AI. In this paper, we firstly describe the structure and features of Botzone, then focus on our experience in utilizing Botzone for a programming course.

Learning interpretable disentangled representations is a crucial yet challenging task. In this paper, we propose a weakly semi-supervised method, termed as Dual Swap Disentangling (DSD), for disentangling using both labeled and unlabeled data. Unlike conventional weakly supervised methods that rely on full annotations on the group of samples, we require only limited annotations on paired samples that indicate their shared attribute like the color. Our model takes the form of a dual autoencoder structure. To achieve disentangling using the labeled pairs, we follow a encoding-swap-decoding'' process, where we first swap the parts of their encodings corresponding to the shared attribute, and then decode the obtained hybrid codes to reconstruct the original input pairs. For unlabeled pairs, we follow theencoding-swap-decoding'' process twice on designated encoding parts and enforce the final outputs to approximate the input pairs. By isolating parts of the encoding and swapping them back and forth, we impose the dimension-wise modularity and portability of the encodings of the unlabeled samples, which implicitly encourages disentangling under the guidance of labeled pairs. This dual swap mechanism, tailored for semi-supervised setting, turns out to be very effective. Experiments on image datasets from a wide domain show that our model yields state-of-the-art disentangling performances.

Single-image dehazing is a challenging problem due to its ill-posed nature. Existing methods rely on a suboptimal two-step approach, where an intermediate product like a depth map is estimated, based on which the haze-free image is subsequently generated using an artificial prior formula. In this paper, we propose a light dual-task Neural Network called LDTNet that restores the haze-free image in one shot. We use transmission map estimation as an auxiliary task to assist the main task, haze removal, in feature extraction and to enhance the generalization of the network. In LDTNet, the haze-free image and the transmission map are produced simultaneously. As a result, the artificial prior is reduced to the smallest extent. Extensive experiments demonstrate that our algorithm achieves superior performance against the state-of-the-art methods on both synthetic and real-world images.

In this paper, we propose to exploit the interactions between non-associable tracklets to facilitate multi-object tracking. We introduce two types of tracklet interactions, close interaction and distant interaction. The close interaction imposes physical constraints between two temporally overlapping tracklets and more importantly, allows us to learn local classifiers to distinguish targets that are close to each other in the spatiotemporal domain. The distant interaction, on the other hand, accounts for the higher-order motion and appearance consistency between two temporally isolated tracklets. Our approach is modeled as a binary labeling problem and solved using the efficient Quadratic Pseudo-Boolean Optimization (QPBO). It yields promising tracking performance on the challenging PETS09 and MOT16 dataset. Our code will be made publicly available upon the acceptance of the manuscript.

Video super-resolution (SR) aims at estimating a high-resolution (HR) video sequence from a low-resolution (LR) one. Given that deep learning has been successfully applied to the task of single image SR, which demonstrates the strong capability of neural networks for modeling spatial relation within one single image, the key challenge to conduct video SR is how to efficiently and effectively exploit the temporal dependency among consecutive LR frames other than the spatial relation. However, this remains challenging because complex motion is difficult to model and can bring detrimental effects if not handled properly. We tackle the problem of learning temporal dynamics from two aspects. First, we propose a temporal adaptive neural network that can adaptively determine the optimal scale of temporal dependency. Inspired by the Inception module in GoogLeNet [1], filters of various temporal scales are applied to the input LR sequence before their responses are adaptively aggregated, in order to fully exploit the temporal relation among consecutive LR frames. Second, we decrease the complexity of motion among neighboring frames using a spatial alignment network that can be end-to-end trained with the temporal adaptive network and has the merit of increasing the robustness to complex motion and the efficiency compared to competing image alignment methods. We provide a comprehensive evaluation of the temporal adaptation and the spatial alignment modules. We show the temporal adaptive design considerably improve SR quality over its plain counterparts, and the spatial alignment network is able to attain comparable SR performance with the sophisticated optical flow based approach, but requires much less running time. Overall our proposed model with learned temporal dynamics is shown to achieve state-of-the-art SR results in terms of not only spatial consistency but also temporal coherence on public video datasets. More information can be found in.

Deep compression refers to removing the redundancy of parameters and feature maps for deep learning models. Low-rank approximation and pruning for sparse structures play a vital role in many compression works. However, weight filters tend to be both low-rank and sparse. Neglecting either part of these structure information in previous methods results in iteratively retraining, compromising accuracy, and low compression rates. Here we propose a unified framework integrating the low-rank and sparse decomposition of weight matrices with the feature map reconstructions. Our model includes methods like pruning connections as special cases, and is optimized by a fast SVD-free algorithm. It has been theoretically proven that, with a small sample, due to its generalizability, our model can well reconstruct the feature maps on both training and test data, which results in less compromising accuracy prior to the subsequent retraining. With such a warm start to retrain, the compression method always possesses several merits: (a) higher compression rates, (b) little loss of accuracy, and (c) fewer rounds to compress deep models. The experimental results on several popular models such as AlexNet, VGG-16, and GoogLeNet show that our model can significantly reduce the parameters for both convolutional and fully-connected layers. As a result, our model reduces the size of VGG-16 by 15×, better than other recent compression methods that use a single strategy.

Many state-of-the-art approaches to multi-object tracking rely on detecting them in each frame independently, grouping detections into short but reliable trajectory segments, and then further grouping them into full trajectories. This grouping typically relies on imposing local smoothness constraints but almost never on enforcing more global ones on the trajectories. In this paper, we propose a non-Markovian approach to imposing global consistency by using behavioral patterns to guide the tracking algorithm. When used in conjunction with state-of-the-art tracking algorithms, this further increases their already good performance on multiple challenging datasets. We show significant improvements both in supervised settings where ground truth is available and behavioral patterns can be learned from it, and in completely unsupervised settings.

The practical application of Computed Tomography (CT) faces the dilemma between higher image resolution and less X-ray exposure for patients, motivating the research on CT super-resolution (SR). In this paper, we apply state-of-the-art SR techniques to reconstruct CT images using two proposed advanced CT SR models based on Convolutional Neural Networks (CNNs) and residual learning: a single-slice CT SR network (S-CTSRN), and a multi-slice CT SR network (M-CTSRN). S-CTSRN improves the high-frequency feature extraction by incorporating the residual learning strategy, while M-CTSRN further utilizes the coherence between neighboring CT slices for better SR reconstruction. We evaluate both models on a large-scale CT dataset1, and obtain competitive results both quantitatively and qualitatively.

This letter proposes a novel rotation-invariant feature for object detection in optical remote sensing images. Different from previous rotation-invariant features, the proposed rotation-invariant matrix (RIM) can incorporate partial angular spatial information in addition to radial spatial information. Moreover, it can be further calculated between different rings for a redundant representation of the spatial layout. Based on the RIM, we further propose an RIM_FV_RPP feature for object detection. For an image region, we first densely extract RIM features from overlapping blocks; then, these RIM features are encoded into Fisher vectors; finally, a pyramid pooling strategy that hierarchically accumulates Fisher vectors in ring subregions is used to encode richer spatial information while maintaining rotation invariance. Both of the RIM and RIM_FV_RPP are rotation invariant. Experiments on airplane and car detection in optical remote sensing images demonstrate the superiority of our feature to the state of the art.

Minimum-cost flow algorithms have recently achieved state-of-the-art results in multi-object tracking. However, they rely on the whole image sequence as input. When deployed in real-time applications or in distributed settings, these algorithms first operate on short batches of frames and then stitch the results into full trajectories. This decoupled strategy is prone to errors because the batch-based tracking errors may propagate to the final trajectories and cannot be corrected by other batches. In this paper, we propose a greedy batch-based minimum-cost flow approach for tracking multiple objects. Unlike existing approaches that conduct batch-based tracking and stitching sequentially, we optimize consecutive batches jointly so that the tracking results on one batch may benefit the results on the other. Specifically, we apply a generalized minimum-cost flows (MCF) algorithm on each batch and generate a set of conflicting trajectories. These trajectories comprise the ones with high probabilities, but also those with low probabilities potentially missed by detectors and trackers. We then apply the generalized MCF again to obtain the optimal matching between trajectories from consecutive batches. Our proposed approach is simple, effective, and does not require training. We demonstrate the power of our approach on data sets of different scenarios.

We propose a novel approach to automatically tracking elliptical cell populations in time-lapse image sequences. Given an initial segmentation, we account for partial occlusions and overlaps by generating an over-complete set of competing detection hypotheses. To this end, we fit ellipses to portions of the initial regions and build a hierarchy of ellipses, which are then treated as cell candidates. We then select temporally consistent ones by solving to optimality an integer program with only one type of flow variables. This eliminates the need for heuristics to handle missed detections due to partial occlusions and complex morphology. We demonstrate the effectiveness of our approach on a range of challenging sequences consisting of clumped cells and show that it outperforms state-of-the-art techniques.

Tracking the ball is critical for video-based analysis of team sports. However, it is difficult, especially in low-resolution images, due to the small size of the ball, its speed that creates motion blur, and its often being occluded by players. In this paper, we propose a generic and principled approach to modeling the interaction between the ball and the players while also imposing appropriate physical constraints on the ball's trajectory. We show that our approach, formulated in terms of a Mixed Integer Program, is more robust and more accurate than several state-of-the-art approaches on real-life volleyball, basketball, and soccer sequences.

In this paper, we show that tracking different kinds of interacting objects can be formulated as a network-flow mixed integer program. This is made possible by tracking all objects simultaneously using intertwined flow variables and expressing the fact that one object can appear or disappear at locations where another is in terms of linear flow constraints. Our proposed method is able to track invisible objects whose only evidence is the presence of other objects that contain them. Furthermore, our tracklet-based implementation yields real-time tracking performance. We demonstrate the power of our approach on scenes involving cars and pedestrians, bags being carried and dropped by people, and balls being passed from one player to the next in team sports. In particular, we show that by estimating jointly and globally the trajectories of different types of objects, the presence of the ones which were not initially detected based solely on image evidence can be inferred from the detections of the others.

Images are usually represented by features from multiple views, e.g., color and texture. In image classification, the goal is to fuse all the multi-view features in a reasonable manner and achieve satisfactory classification performance. However, the features are often different in nature and it is nontrivial to fuse them. Particularly, some extracted features are redundant or noisy and are consequently not discriminative for classification. To alleviate these problems in an image classification context, we propose in this paper a novel multi-view embedding framework, termed as Grassmannian regularized structured multi-view embedding, or GrassReg for short. GrassReg transfers the graph Laplacian obtained from each view to a point on the Grassmann manifold and penalizes the disagreement between different views according to Grassmannian distance. Therefore, a view that is consistent with others is more important than a view that disagrees with others for learning a unified subspace for multi-view data representation. In addition, we impose the group sparsity penalty onto the low-dimensional embeddings obtained hence they can better explore the group structure of the intrinsic data distribution. Empirically, we compare GrassReg with representative multi-view algorithms and show the effectiveness of GrassReg on a number of multi-view image data sets.

Locality-Sensitive Hashing (LSH) approximates nearest neighbors in high dimensions by projecting original data into low-dimensional subspaces. The basic idea is to hash data samples to ensure that the probability of collision is much higher for samples that are close to each other than for those that are far apart. However, by applying k random hashing functions on original data, LSH fails to find the most discriminant hashing-subspaces, so the nearest neighbor approximation is inefficient. To alleviate this problem, we propose the Grassmann Hashing (GRASH) for approximate nearest neighbor search in high dimensions. GRASH first introduces a set of subspace candidates from Linear Discriminant Analysis (LDA). Then it applies Grassmann metric to select the optimal subspaces for hashing. Finally, it generates hashing codes based on non-uniform bucket size design motivated by Lloyd-Max quantization. The proposed GRASH model enjoys a number of merits: 1) GRASH introduces the Grassmann metric to measure the similarity between different hashing subspaces, so the hashing function can better capture the data diversity; 2) GRASH obtains the subspace candidates from LDA, so it incorporates the discriminant information into the hashing functions; 3) GRASH extends LSH's 1-d hashing subspaces to m-d, i.e. it is a multidimensional extension of hashing approximation; 4) motivated by Lloyd-Max quantization, GRASH applies non-uniform size bucket to generate hashing codes, so the distortion can be minimized. Experimental results on a number of datasets confirm the validity of our proposed model.

Conventional linear subspace learning methods like principal component analysis (PCA), linear discriminant analysis (LDA) derive subspaces from the whole data set. These approaches have limitations in the sense that they are linear while the data distribution we are trying to model is typically nonlinear. Moreover, these algorithms fail to incorporate local variations of the intrinsic sample distribution manifold. Therefore, these algorithms are ineffective when applied on large scale datasets. Kernel versions of these approaches can alleviate the problem to certain degree but face a serious computational challenge when data set is large, where the computing involves Eigen/QP problems of size N × N. When N is large, kernel versions are not computationally practical. To tackle the aforementioned problems and improve recognition/searching performance, especially on large scale image datasets, we propose a novel local subspace indexing model for image search termed Subspace Indexing Model on Grassmann Manifold (SIM-GM). SIM-GM partitions the global space into local patches with a hierarchical structure; the global model is, therefore, approximated by piece-wise linear local subspace models. By further applying the Grassmann manifold distance, SIM-GM is able to organize localized models into a hierarchy of indexed structure, and allow fast query selection of the optimal ones for classification. Our proposed SIM-GM enjoys a number of merits: 1) it is able to deal with a large number of training samples efficiently; 2) it is a query-driven approach, i.e., it is able to return an effective local space model, so the recognition performance could be significantly improved; 3) it is a common framework, which can incorporate many learning algorithms. Theoretical analysis and extensive experimental results confirm the validity of this model.

In the original paper named above (ibid., vol. 20, no. 9, pp. 2627-2635, Sep 2011), Table II incorrectly appeared as a duplicate of Table III. Instead, Table II should have correctly appeared as shown here.