Jungong Han

Xiaohan Ding, Xiangyu Zhang, Yizhuang Zhou et al.

We revisit large kernel design in modern convolutional neural networks (CNNs). Inspired by recent advances in vision transformers (ViTs), in this paper, we demonstrate that using a few large convolutional kernels instead of a stack of small kernels could be a more powerful paradigm. We suggested five guidelines, e.g., applying re-parameterized large depth-wise convolutions, to design efficient high-performance large-kernel CNNs. Following the guidelines, we propose RepLKNet, a pure CNN architecture whose kernel size is as large as 31x31, in contrast to commonly used 3x3. RepLKNet greatly closes the performance gap between CNNs and ViTs, e.g., achieving comparable or superior results than Swin Transformer on ImageNet and a few typical downstream tasks, with lower latency. RepLKNet also shows nice scalability to big data and large models, obtaining 87.8% top-1 accuracy on ImageNet and 56.0% mIoU on ADE20K, which is very competitive among the state-of-the-arts with similar model sizes. Our study further reveals that, in contrast to small-kernel CNNs, large-kernel CNNs have much larger effective receptive fields and higher shape bias rather than texture bias. Code & models at https://github.com/megvii-research/RepLKNet.

Ao Wang, Hui Chen, Zijia Lin et al.

Recently, lightweight Vision Transformers (ViTs) demonstrate superior performance and lower latency, compared with lightweight Convolutional Neural Networks (CNNs), on resource-constrained mobile devices. Researchers have discovered many structural connections between lightweight ViTs and lightweight CNNs. However, the notable architectural disparities in the block structure, macro, and micro designs between them have not been adequately examined. In this study, we revisit the efficient design of lightweight CNNs from ViT perspective and emphasize their promising prospect for mobile devices. Specifically, we incrementally enhance the mobile-friendliness of a standard lightweight CNN, \ie, MobileNetV3, by integrating the efficient architectural designs of lightweight ViTs. This ends up with a new family of pure lightweight CNNs, namely RepViT. Extensive experiments show that RepViT outperforms existing state-of-the-art lightweight ViTs and exhibits favorable latency in various vision tasks. Notably, on ImageNet, RepViT achieves over 80\% top-1 accuracy with 1.0 ms latency on an iPhone 12, which is the first time for a lightweight model, to the best of our knowledge. Besides, when RepViT meets SAM, our RepViT-SAM can achieve nearly 10$\times$ faster inference than the advanced MobileSAM. Codes and models are available at \url{https://github.com/THU-MIG/RepViT}.

Biao Qian, Yang Wang, Yong Wu et al.

Data-Free Quantization (DFQ) addresses data security concerns by synthesizing samples, without accessing real data. It has garnered increasing attention in the context of Vision Transformers (ViTs), owing to the superiority of the self-attention mechanism compared to classical convolutional operation. However, previous DFQ arts for ViTs often suffer from a distribution mismatch between synthetic samples and input distribution expected by quantized models Q, resulting in the suboptimal performance. In this paper, we propose a novel Masked Attention Alignment approach for Data-Free Quantization of ViTs, named MaskAQ, revealing that: 1) the semantics in the self-attention mechanism is predominantly localized to a sparse subset of patches, called informative regions; 2) the informative regions dominate the mutual information between synthetic samples and Q's outputs. To these ends, we incorporate differential entropy maximum over patch similarity of synthetic samples, to decouple informative regions from noisy background. To couple with varied Q, the informative regions are selected to align full-precision models with Q via a masked attention alignment objective, thus yielding high-quality synthetic samples. Furthermore, a periodic sample refreshing strategy comes up to endow MaskAQ with the capacity to continually adapt to the evolving state of Q throughout the training process, to preserve desirable mutual information with synthetic samples. Extensive experiments verify the merits of MaskAQ over state-of-the-art approaches across multiple backbones and downstream tasks. Our code is available at https://github.com/hfutqian/MaskAQ.

Xiaohan Ding, Honghao Chen, Xiangyu Zhang et al.

The well-designed structures in neural networks reflect the prior knowledge incorporated into the models. However, though different models have various priors, we are used to training them with model-agnostic optimizers such as SGD. In this paper, we propose to incorporate model-specific prior knowledge into optimizers by modifying the gradients according to a set of model-specific hyper-parameters. Such a methodology is referred to as Gradient Re-parameterization, and the optimizers are named RepOptimizers. For the extreme simplicity of model structure, we focus on a VGG-style plain model and showcase that such a simple model trained with a RepOptimizer, which is referred to as RepOpt-VGG, performs on par with or better than the recent well-designed models. From a practical perspective, RepOpt-VGG is a favorable base model because of its simple structure, high inference speed and training efficiency. Compared to Structural Re-parameterization, which adds priors into models via constructing extra training-time structures, RepOptimizers require no extra forward/backward computations and solve the problem of quantization. We hope to spark further research beyond the realms of model structure design. Code and models \url{https://github.com/DingXiaoH/RepOptimizers}.

Yutao Hu, Qixiong Wang, Wenqi Shao et al.

Referring image segmentation aims to segment the target object referred by a natural language expression. However, previous methods rely on the strong assumption that one sentence must describe one target in the image, which is often not the case in real-world applications. As a result, such methods fail when the expressions refer to either no objects or multiple objects. In this paper, we address this issue from two perspectives. First, we propose a Dual Multi-Modal Interaction (DMMI) Network, which contains two decoder branches and enables information flow in two directions. In the text-to-image decoder, text embedding is utilized to query the visual feature and localize the corresponding target. Meanwhile, the image-to-text decoder is implemented to reconstruct the erased entity-phrase conditioned on the visual feature. In this way, visual features are encouraged to contain the critical semantic information about target entity, which supports the accurate segmentation in the text-to-image decoder in turn. Secondly, we collect a new challenging but realistic dataset called Ref-ZOM, which includes image-text pairs under different settings. Extensive experiments demonstrate our method achieves state-of-the-art performance on different datasets, and the Ref-ZOM-trained model performs well on various types of text inputs. Codes and datasets are available at https://github.com/toggle1995/RIS-DMMI.

Zhuoxu Huang, Zhenkun Fan, Jungong Han et al.

Investigating spatial-temporal correlations, specifically how spatial points vary over time, is crucial for understanding point cloud videos. Traditional methods, particularly flow-based techniques, struggle with these correlations due to the unordered spatial arrangement of sequential point cloud data. To address this challenge, we propose a novel approach that regularizes spatial-temporal correlation learning by formulating the problem as a solvable Partial Differential Equation (PDE). While PDEs have long been effective in the physical domain, their application to novel sequential data like point cloud video remains underexplored. Inspired by fluid analysis, we construct a simplified PDE, and the process of solving PDE is guided and refined by a contrastive learning structure between the temporal embeddings and the spatial embeddings. With this extra supervision, our method, named MotionPDE, serves as an effective, plug-and-play enhancement module for existing backbone models, adding minimal computational overhead and parameters. Capitalizing on the contrastive learning process, we delve deeper into the self-supervised capabilities of MotionPDE, yielding promising results that underscore its utility and adaptability in point cloud video data interpretation. The code repo with trained checkpoints will be available at https://github.com/zhh6425/motionpde.git for facilitating future research.

Zixuan Ding, Ao Wang, Hui Chen et al.

Multi-label recognition (MLR) with incomplete labels is very challenging. Recent works strive to explore the image-to-label correspondence in the vision-language model, \ie, CLIP, to compensate for insufficient annotations. In spite of promising performance, they generally overlook the valuable prior about the label-to-label correspondence. In this paper, we advocate remedying the deficiency of label supervision for the MLR with incomplete labels by deriving a structured semantic prior about the label-to-label correspondence via a semantic prior prompter. We then present a novel Semantic Correspondence Prompt Network (SCPNet), which can thoroughly explore the structured semantic prior. A Prior-Enhanced Self-Supervised Learning method is further introduced to enhance the use of the prior. Comprehensive experiments and analyses on several widely used benchmark datasets show that our method significantly outperforms existing methods on all datasets, well demonstrating the effectiveness and the superiority of our method. Our code will be available at https://github.com/jameslahm/SCPNet.

Yunqi Miao, Alexandros Lattas, Jiankang Deng et al.

Near infrared (NIR) to Visible (VIS) face matching is challenging due to the significant domain gaps as well as a lack of sufficient data for cross-modality model training. To overcome this problem, we propose a novel method for paired NIR-VIS facial image generation. Specifically, we reconstruct 3D face shape and reflectance from a large 2D facial dataset and introduce a novel method of transforming the VIS reflectance to NIR reflectance. We then use a physically-based renderer to generate a vast, high-resolution and photorealistic dataset consisting of various poses and identities in the NIR and VIS spectra. Moreover, to facilitate the identity feature learning, we propose an IDentity-based Maximum Mean Discrepancy (ID-MMD) loss, which not only reduces the modality gap between NIR and VIS images at the domain level but encourages the network to focus on the identity features instead of facial details, such as poses and accessories. Extensive experiments conducted on four challenging NIR-VIS face recognition benchmarks demonstrate that the proposed method can achieve comparable performance with the state-of-the-art (SOTA) methods without requiring any existing NIR-VIS face recognition datasets. With slightly fine-tuning on the target NIR-VIS face recognition datasets, our method can significantly surpass the SOTA performance. Code and pretrained models are released under the insightface (https://github.com/deepinsight/insightface/tree/master/recognition).

Hao Chen, Yonghan Dong, Zheming Lu et al. · cmu

Few-Shot Segmentation (FSS) aims to segment the novel class images with a few annotated samples. In this paper, we propose a dense affinity matching (DAM) framework to exploit the support-query interaction by densely capturing both the pixel-to-pixel and pixel-to-patch relations in each support-query pair with the bidirectional 3D convolutions. Different from the existing methods that remove the support background, we design a hysteretic spatial filtering module (HSFM) to filter the background-related query features and retain the foreground-related query features with the assistance of the support background, which is beneficial for eliminating interference objects in the query background. We comprehensively evaluate our DAM on ten benchmarks under cross-category, cross-dataset, and cross-domain FSS tasks. Experimental results demonstrate that DAM performs very competitively under different settings with only 0.68M parameters, especially under cross-domain FSS tasks, showing its effectiveness and efficiency.

Shaohui Lin, Wenxuan Huang, Jiao Xie et al.

Filter pruning simultaneously accelerates the computation and reduces the memory overhead of CNNs, which can be effectively applied to edge devices and cloud services. In this paper, we propose a novel Knowledge-driven Differential Filter Sampler~(KDFS) with Masked Filter Modeling~(MFM) framework for filter pruning, which globally prunes the redundant filters based on the prior knowledge of a pre-trained model in a differential and non-alternative optimization. Specifically, we design a differential sampler with learnable sampling parameters to build a binary mask vector for each layer, determining whether the corresponding filters are redundant. To learn the mask, we introduce masked filter modeling to construct PCA-like knowledge by aligning the intermediate features from the pre-trained teacher model and the outputs of the student decoder taking sampling features as the input. The mask and sampler are directly optimized by the Gumbel-Softmax Straight-Through Gradient Estimator in an end-to-end manner in combination with global pruning constraint, MFM reconstruction error, and dark knowledge. Extensive experiments demonstrate the proposed KDFS's effectiveness in compressing the base models on various datasets. For instance, the pruned ResNet-50 on ImageNet achieves $55.36\%$ computation reduction, and $42.86\%$ parameter reduction, while only dropping $0.35\%$ Top-1 accuracy, significantly outperforming the state-of-the-art methods. The code is available at \url{https://github.com/Osilly/KDFS}.

Boyang Xia, Zhihao Wang, Wenhao Wu et al. · amazon-science

Efficient video recognition is a hot-spot research topic with the explosive growth of multimedia data on the Internet and mobile devices. Most existing methods select the salient frames without awareness of the class-specific saliency scores, which neglect the implicit association between the saliency of frames and its belonging category. To alleviate this issue, we devise a novel Temporal Saliency Query (TSQ) mechanism, which introduces class-specific information to provide fine-grained cues for saliency measurement. Specifically, we model the class-specific saliency measuring process as a query-response task. For each category, the common pattern of it is employed as a query and the most salient frames are responded to it. Then, the calculated similarities are adopted as the frame saliency scores. To achieve it, we propose a Temporal Saliency Query Network (TSQNet) that includes two instantiations of the TSQ mechanism based on visual appearance similarities and textual event-object relations. Afterward, cross-modality interactions are imposed to promote the information exchange between them. Finally, we use the class-specific saliencies of the most confident categories generated by two modalities to perform the selection of salient frames. Extensive experiments demonstrate the effectiveness of our method by achieving state-of-the-art results on ActivityNet, FCVID and Mini-Kinetics datasets. Our project page is at https://lawrencexia2008.github.io/projects/tsqnet .

Zhenliang Li, Yutao Hu, Qixiong Wang et al.

Recent advances in generative image editing have improved the realism and controllability of localized image manipulation, raising new challenges for image manipulation detection and localization (IMDL). However, existing IMDL benchmarks still have limitations in visual realism, manipulation diversity, and generator coverage, making it difficult to reflect recent trends in image manipulation. To address these limitations, we introduce Impostor, a high-quality AI-edited image manipulation localization dataset containing 100K manipulated images. Impostor is constructed by CraftAgent, a closed-loop agent framework that integrates scene perception, editing planning, manipulation execution, quality validation, and iterative reflection to automatically generate diverse and visually realistic manipulated images. Moreover, Impostor contains images generated by seven recent AIGC models across three manipulation types and includes multiple manipulated regions, providing a more comprehensive benchmark for AIGC-based IMDL. Furthermore, we propose PhaseAware-Net (PANet), a semantic-forensic framework that introduces local phase modeling and semantic-forensic consistency learning to better localize semantically plausible yet forensically disrupted manipulated regions. Extensive experiments show that Impostor poses significant challenges to existing large vision-language models (LVLMs) and specialized IMDL methods, while PANet achieves superior performance on Impostor and multiple public benchmarks.

Mengyao Lyu, Tianxiang Hao, Xinhao Xu et al.

Domain Adaptation (DA) facilitates knowledge transfer from a source domain to a related target domain. This paper investigates a practical DA paradigm, namely Source data-Free Active Domain Adaptation (SFADA), where source data becomes inaccessible during adaptation, and a minimum amount of annotation budget is available in the target domain. Without referencing the source data, new challenges emerge in identifying the most informative target samples for labeling, establishing cross-domain alignment during adaptation, and ensuring continuous performance improvements through the iterative query-and-adaptation process. In response, we present learn from the learnt (LFTL), a novel paradigm for SFADA to leverage the learnt knowledge from the source pretrained model and actively iterated models without extra overhead. We propose Contrastive Active Sampling to learn from the hypotheses of the preceding model, thereby querying target samples that are both informative to the current model and persistently challenging throughout active learning. During adaptation, we learn from features of actively selected anchors obtained from previous intermediate models, so that the Visual Persistence-guided Adaptation can facilitate feature distribution alignment and active sample exploitation. Extensive experiments on three widely-used benchmarks show that our LFTL achieves state-of-the-art performance, superior computational efficiency and continuous improvements as the annotation budget increases. Our code is available at https://github.com/lyumengyao/lftl.

De Cheng, Gerong Wang, Bo Wang et al.

Zero-shot learning (ZSL) aims to learn models that can recognize unseen image semantics based on the training of data with seen semantics. Recent studies either leverage the global image features or mine discriminative local patch features to associate the extracted visual features to the semantic attributes. However, due to the lack of the necessary top-down guidance and semantic alignment for ensuring the model attending to the real attribute-correlation regions, these methods still encounter a significant semantic gap between the visual modality and the attribute modality, which makes their prediction on unseen semantics unreliable. To solve this problem, this paper establishes a novel transformer encoder-decoder model, called hybrid routing transformer (HRT). In HRT encoder, we embed an active attention, which is constructed by both the bottom-up and the top-down dynamic routing pathways to generate the attribute-aligned visual feature. While in HRT decoder, we use static routing to calculate the correlation among the attribute-aligned visual features, the corresponding attribute semantics, and the class attribute vectors to generate the final class label predictions. This design makes the presented transformer model a hybrid of 1) top-down and bottom-up attention pathways and 2) dynamic and static routing pathways. Comprehensive experiments on three widely-used benchmark datasets, namely CUB, SUN, and AWA2, are conducted. The obtained experimental results demonstrate the effectiveness of the proposed method.

Zhuoxu Huang, Zhiyou Zhao, Banghuai Li et al.

Transformer with its underlying attention mechanism and the ability to capture long-range dependencies makes it become a natural choice for unordered point cloud data. However, separated local regions from the general sampling architecture corrupt the structural information of the instances, and the inherent relationships between adjacent local regions lack exploration, while local structural information is crucial in a transformer-based 3D point cloud model. Therefore, in this paper, we propose a novel module named Local Context Propagation (LCP) to exploit the message passing between neighboring local regions and make their representations more informative and discriminative. More specifically, we use the overlap points of adjacent local regions (which statistically show to be prevalent) as intermediaries, then re-weight the features of these shared points from different local regions before passing them to the next layers. Inserting the LCP module between two transformer layers results in a significant improvement in network expressiveness. Finally, we design a flexible LCPFormer architecture equipped with the LCP module. The proposed method is applicable to different tasks and outperforms various transformer-based methods in benchmarks including 3D shape classification and dense prediction tasks such as 3D object detection and semantic segmentation. Code will be released for reproduction.

Hui-Yue Yang, Hui Chen, Lihao Liu et al.

Unsupervised anomaly detection (AD) aims to train robust detection models using only normal samples, while can generalize well to unseen anomalies. Recent research focuses on a unified unsupervised AD setting in which only one model is trained for all classes, i.e., n-class-one-model paradigm. Feature-reconstruction-based methods achieve state-of-the-art performance in this scenario. However, existing methods often suffer from a lack of sufficient contextual awareness, thereby compromising the quality of the reconstruction. To address this issue, we introduce a novel Reconstruction as Sequence (RAS) method, which enhances the contextual correspondence during feature reconstruction from a sequence modeling perspective. In particular, based on the transformer technique, we integrate a specialized RASFormer block into RAS. This block enables the capture of spatial relationships among different image regions and enhances sequential dependencies throughout the reconstruction process. By incorporating the RASFormer block, our RAS method achieves superior contextual awareness capabilities, leading to remarkable performance. Experimental results show that our RAS significantly outperforms competing methods, well demonstrating the effectiveness and superiority of our method. Our code is available at https://github.com/Nothingtolose9979/RAS.

Haoran Wang, Di Xu, Dongliang He et al.

Video-text retrieval (VTR) is an attractive yet challenging task for multi-modal understanding, which aims to search for relevant video (text) given a query (video). Existing methods typically employ completely heterogeneous visual-textual information to align video and text, whilst lacking the awareness of homogeneous high-level semantic information residing in both modalities. To fill this gap, in this work, we propose a novel visual-linguistic aligning model named HiSE for VTR, which improves the cross-modal representation by incorporating explicit high-level semantics. First, we explore the hierarchical property of explicit high-level semantics, and further decompose it into two levels, i.e. discrete semantics and holistic semantics. Specifically, for visual branch, we exploit an off-the-shelf semantic entity predictor to generate discrete high-level semantics. In parallel, a trained video captioning model is employed to output holistic high-level semantics. As for the textual modality, we parse the text into three parts including occurrence, action and entity. In particular, the occurrence corresponds to the holistic high-level semantics, meanwhile both action and entity represent the discrete ones. Then, different graph reasoning techniques are utilized to promote the interaction between holistic and discrete high-level semantics. Extensive experiments demonstrate that, with the aid of explicit high-level semantics, our method achieves the superior performance over state-of-the-art methods on three benchmark datasets, including MSR-VTT, MSVD and DiDeMo.

Xiaodong Chen, Wu Liu, Xinchen Liu et al.

Recognizing human actions from point cloud videos has attracted tremendous attention from both academia and industry due to its wide applications like automatic driving, robotics, and so on. However, current methods for point cloud action recognition usually require a huge amount of data with manual annotations and a complex backbone network with high computation costs, which makes it impractical for real-world applications. Therefore, this paper considers the task of semi-supervised point cloud action recognition. We propose a Masked Pseudo-Labeling autoEncoder (\textbf{MAPLE}) framework to learn effective representations with much fewer annotations for point cloud action recognition. In particular, we design a novel and efficient \textbf{De}coupled \textbf{s}patial-\textbf{t}emporal Trans\textbf{Former} (\textbf{DestFormer}) as the backbone of MAPLE. In DestFormer, the spatial and temporal dimensions of the 4D point cloud videos are decoupled to achieve efficient self-attention for learning both long-term and short-term features. Moreover, to learn discriminative features from fewer annotations, we design a masked pseudo-labeling autoencoder structure to guide the DestFormer to reconstruct features of masked frames from the available frames. More importantly, for unlabeled data, we exploit the pseudo-labels from the classification head as the supervision signal for the reconstruction of features from the masked frames. Finally, comprehensive experiments demonstrate that MAPLE achieves superior results on three public benchmarks and outperforms the state-of-the-art method by 8.08\% accuracy on the MSR-Action3D dataset.

Ao Wang, Hui Chen, Zijia Lin et al.

Vision Transformers (ViTs) have emerged as state-of-the-art models for various vision tasks recently. However, their heavy computation costs remain daunting for resource-limited devices. To address this, researchers have dedicated themselves to compressing redundant information in ViTs for acceleration. However, existing approaches generally sparsely drop redundant image tokens by token pruning or brutally remove channels by channel pruning, leading to a sub-optimal balance between model performance and inference speed. Moreover, they struggle when transferring compressed models to downstream vision tasks that require the spatial structure of images, such as semantic segmentation. To tackle these issues, we propose CAIT, a joint \underline{c}ompression method for ViTs that achieves a harmonious blend of high \underline{a}ccuracy, fast \underline{i}nference speed, and favorable \underline{t}ransferability to downstream tasks. Specifically, we introduce an asymmetric token merging (ATME) strategy to effectively integrate neighboring tokens. It can successfully compress redundant token information while preserving the spatial structure of images. On top of it, we further design a consistent dynamic channel pruning (CDCP) strategy to dynamically prune unimportant channels in ViTs. Thanks to CDCP, insignificant channels in multi-head self-attention modules of ViTs can be pruned uniformly, significantly enhancing the model compression. Extensive experiments on multiple benchmark datasets show that our proposed method can achieve state-of-the-art performance across various ViTs.

Changrui Chen, Kurt Debattista, Jungong Han

Due to the costliness of labelled data in real-world applications, semi-supervised object detectors, underpinned by pseudo labelling, are appealing. However, handling confusing samples is nontrivial: discarding valuable confusing samples would compromise the model generalisation while using them for training would exacerbate the confirmation bias issue caused by inevitable mislabelling. To solve this problem, this paper proposes to use confusing samples proactively without label correction. Specifically, a virtual category (VC) is assigned to each confusing sample such that they can safely contribute to the model optimisation even without a concrete label. It is attributed to specifying the embedding distance between the training sample and the virtual category as the lower bound of the inter-class distance. Moreover, we also modify the localisation loss to allow high-quality boundaries for location regression. Extensive experiments demonstrate that the proposed VC learning significantly surpasses the state-of-the-art, especially with small amounts of available labels.

Fan Yang, Xinhao Xu, Hui Chen et al.

The ground plane prior is a very informative geometry clue in monocular 3D object detection (M3OD). However, it has been neglected by most mainstream methods. In this paper, we identify two key factors that limit the applicability of ground plane prior: the projection point localization issue and the ground plane tilt issue. To pick up the ground plane prior for M3OD, we propose a Ground Plane Enhanced Network (GPENet) which resolves both issues at one go. For the projection point localization issue, instead of using the bottom vertices or bottom center of the 3D bounding box (BBox), we leverage the object's ground contact points, which are explicit pixels in the image and easy for the neural network to detect. For the ground plane tilt problem, our GPENet estimates the horizon line in the image and derives a novel mathematical expression to accurately estimate the ground plane equation. An unsupervised vertical edge mining algorithm is also proposed to address the occlusion of the horizon line. Furthermore, we design a novel 3D bounding box deduction method based on a dynamic back projection algorithm, which could take advantage of the accurate contact points and the ground plane equation. Additionally, using only M3OD labels, contact point and horizon line pseudo labels can be easily generated with NO extra data collection and label annotation cost. Extensive experiments on the popular KITTI benchmark show that our GPENet can outperform other methods and achieve state-of-the-art performance, well demonstrating the effectiveness and the superiority of the proposed approach. Moreover, our GPENet works better than other methods in cross-dataset evaluation on the nuScenes dataset. Our code and models will be published.

Yunqi Miao, Jiankang Deng, Guiguang Ding et al.

Unsupervised person re-identification (ReID) aims to train a feature extractor for identity retrieval without exploiting identity labels. Due to the blind trust in imperfect clustering results, the learning is inevitably misled by unreliable pseudo labels. Albeit the pseudo label refinement has been investigated by previous works, they generally leverage auxiliary information such as camera IDs and body part predictions. This work explores the internal characteristics of clusters to refine pseudo labels. To this end, Confidence-Guided Centroids (CGC) are proposed to provide reliable cluster-wise prototypes for feature learning. Since samples with high confidence are exclusively involved in the formation of centroids, the identity information of low-confidence samples, i.e., boundary samples, are NOT likely to contribute to the corresponding centroid. Given the new centroids, current learning scheme, where samples are enforced to learn from their assigned centroids solely, is unwise. To remedy the situation, we propose to use Confidence-Guided pseudo Label (CGL), which enables samples to approach not only the originally assigned centroid but other centroids that are potentially embedded with their identity information. Empowered by confidence-guided centroids and labels, our method yields comparable performance with, or even outperforms, state-of-the-art pseudo label refinement works that largely leverage auxiliary information.

Mingqi Gao, Yunqi Miao, Jungong Han

Human Mesh Recovery (HMR) aims to reconstruct 3D human pose and shape from 2D observations and is fundamental to human-centric understanding in real-world scenarios. While recent image-based HMR methods such as SAM 3D Body achieve strong robustness on in-the-wild images, they rely on per-frame inference when applied to videos, leading to temporal inconsistency and degraded performance under occlusions. We address these issues without extra training by leveraging the inherent human continuity in videos. We propose SAM-Body4D, a training-free framework for temporally consistent and occlusion-robust HMR from videos. We first generate identity-consistent masklets using a promptable video segmentation model, then refine them with an Occlusion-Aware module to recover missing regions. The refined masklets guide SAM 3D Body to produce consistent full-body mesh trajectories, while a padding-based parallel strategy enables efficient multi-human inference. Experimental results demonstrate that SAM-Body4D achieves improved temporal stability and robustness in challenging in-the-wild videos, without any retraining. Our code and demo are available at: https://github.com/gaomingqi/sam-body4d.

Xuan Wang, Yinan Wu, Haoran Duan et al.

Vision-Language-Action (VLA) models commonly adapt pretrained Vision-Language Models (VLMs) to robot control by mapping visual observations and language instructions to continuous actions. Existing approaches typically take an action-insufficiency view, assuming that pretrained VLM latents either lack directly usable action information or should be shielded from action-learning signals. Against this view, our \textit{Quotient Theory for VLA} shows that pretrained VLM latents are not action-insufficient but action-sufficient: they already contain the information needed for control, yet remain overcomplete by distinguishing prompt-level variations that induce the same optimal action behavior. To operationalize this theory, we propose QuoVLA, a quotient-space framework for VLA that compresses pretrained VLM latents into action-sufficient representations. Specifically, QuoVLA instantiates this principle with a quantization module and a dual-branch design with relative temporal-complexity regularization, preserving action-relevant information while removing prompt-level redundancy. Extensive experiments across multiple benchmarks demonstrate that QuoVLA achieves strong performance, with particularly notable improvements in generalization under visual, linguistic, and environmental distribution shifts. Our code will be made publicly available.

Mengyao Lyu, Yuhong Yang, Haiwen Hong et al.

The prevalent use of commercial and open-source diffusion models (DMs) for text-to-image generation prompts risk mitigation to prevent undesired behaviors. Existing concept erasing methods in academia are all based on full parameter or specification-based fine-tuning, from which we observe the following issues: 1) Generation alternation towards erosion: Parameter drift during target elimination causes alternations and potential deformations across all generations, even eroding other concepts at varying degrees, which is more evident with multi-concept erased; 2) Transfer inability & deployment inefficiency: Previous model-specific erasure impedes the flexible combination of concepts and the training-free transfer towards other models, resulting in linear cost growth as the deployment scenarios increase. To achieve non-invasive, precise, customizable, and transferable elimination, we ground our erasing framework on one-dimensional adapters to erase multiple concepts from most DMs at once across versatile erasing applications. The concept-SemiPermeable structure is injected as a Membrane (SPM) into any DM to learn targeted erasing, and meantime the alteration and erosion phenomenon is effectively mitigated via a novel Latent Anchoring fine-tuning strategy. Once obtained, SPMs can be flexibly combined and plug-and-play for other DMs without specific re-tuning, enabling timely and efficient adaptation to diverse scenarios. During generation, our Facilitated Transport mechanism dynamically regulates the permeability of each SPM to respond to different input prompts, further minimizing the impact on other concepts. Quantitative and qualitative results across ~40 concepts, 7 DMs and 4 erasing applications have demonstrated the superior erasing of SPM. Our code and pre-tuned SPMs are available on the project page https://lyumengyao.github.io/projects/spm.

Hao Cao, Wenqi Guo, Zhijin Qin et al.

Entropy coding is widely used in typical learned image compression (LIC) that converts latents into a compact bitstream. However, entropy coding is typically sequential and becomes the coding latency bottleneck. To overcome it, we present Entropy-Coding Free Learned Image Compression (EF-LIC), a multi-rate framework that generates compact representation by removing statistical and correlation redundancy with low coding latency. First, we introduce unconstrained vector quantization and prove that its index distribution approaches the maximum-entropy bound, yielding minimal statistical redundancy. Second, we propose a context-conditioned autoregressive transform that directly reparameterizes the latents to reduce inter-dependency. Theoretical analysis shows that EF-LIC can remove correlation redundancy as effectively as typical LIC with entropy coding, leading to comparable compression performance. Experiments show EF-LIC achieves up to 67.86% bitrate reduction over MS-ILLM on Kodak with LPIPS. Ablation studies further show EF-LIC matches the compression performance of its entropy-coding based variant while achieving over $3\times$ faster encoding and $5\times$ faster decoding.

Fan Yang, Sicheng Zhao, Yanhao Zhang et al.

Recent advancements in autonomous driving, augmented reality, robotics, and embodied intelligence have necessitated 3D perception algorithms. However, current 3D perception methods, especially specialized small models, exhibit poor generalization in open scenarios. On the other hand, multimodal large language models (MLLMs) excel in general capacity but underperform in 3D tasks, due to weak 3D local spatial object perception, poor text-based geometric numerical output, and inability to handle camera focal variations. To address these challenges, we propose the following solutions: Spatial-Enhanced Local Feature Mining for better spatial feature extraction, 3D Query Token-Derived Info Decoding for precise geometric regression, and Geometry Projection-Based 3D Reasoning for handling camera focal length variations. We employ parameter-efficient fine-tuning for a pre-trained MLLM and develop LLMI3D, a powerful 3D perception MLLM. Additionally, we have constructed the IG3D dataset, which provides fine-grained descriptions and question-answer annotations. Extensive experiments demonstrate that our LLMI3D achieves state-of-the-art performance, outperforming other methods by a large margin.

Ao Wang, Lihao Liu, Hui Chen et al.

Object detection and segmentation are widely employed in computer vision applications, yet conventional models like YOLO series, while efficient and accurate, are limited by predefined categories, hindering adaptability in open scenarios. Recent open-set methods leverage text prompts, visual cues, or prompt-free paradigm to overcome this, but often compromise between performance and efficiency due to high computational demands or deployment complexity. In this work, we introduce YOLOE, which integrates detection and segmentation across diverse open prompt mechanisms within a single highly efficient model, achieving real-time seeing anything. For text prompts, we propose Re-parameterizable Region-Text Alignment (RepRTA) strategy. It refines pretrained textual embeddings via a re-parameterizable lightweight auxiliary network and enhances visual-textual alignment with zero inference and transferring overhead. For visual prompts, we present Semantic-Activated Visual Prompt Encoder (SAVPE). It employs decoupled semantic and activation branches to bring improved visual embedding and accuracy with minimal complexity. For prompt-free scenario, we introduce Lazy Region-Prompt Contrast (LRPC) strategy. It utilizes a built-in large vocabulary and specialized embedding to identify all objects, avoiding costly language model dependency. Extensive experiments show YOLOE's exceptional zero-shot performance and transferability with high inference efficiency and low training cost. Notably, on LVIS, with 3$\times$ less training cost and 1.4$\times$ inference speedup, YOLOE-v8-S surpasses YOLO-Worldv2-S by 3.5 AP. When transferring to COCO, YOLOE-v8-L achieves 0.6 AP$^b$ and 0.4 AP$^m$ gains over closed-set YOLOv8-L with nearly 4$\times$ less training time. Code and models are available at https://github.com/THU-MIG/yoloe.

Zhe Qian, Yanbiao Ma, Zhuohan Ouyang et al.

Multimodal Large Reasoning Models (MLRMs) have achieved remarkable strides in visual reasoning through test time compute scaling, yet long chain reasoning remains prone to hallucinations. We identify a concerning phenomenon termed the Reasoning Vision Truth Disconnect (RVTD): hallucinations are strongly correlated with cognitive bifurcation points that often exhibit high entropy states. We attribute this vulnerability to a breakdown in visual semantic anchoring, localized within the network's intermediate layers; specifically, during these high uncertainty transitions, the model fails to query visual evidence, reverting instead to language priors. Consequently, we advocate a shift from solely outcome level supervision to augmenting it with fine grained internal attention guidance. To this end, we propose V-STAR (Visual Structural Training with Attention Reinforcement), a lightweight, holistic training paradigm designed to internalize visually aware reasoning capabilities. Central to our approach is the Hierarchical Visual Attention Reward (HVAR), integrated within the GRPO framework. Upon detecting high entropy states, this mechanism dynamically incentivizes visual attention across critical intermediate layers, thereby anchoring the reasoning process back to the visual input. Furthermore, we introduce the Forced Reflection Mechanism (FRM), a trajectory editing strategy that disrupts cognitive inertia by triggering reflection around high entropy cognitive bifurcation points and encouraging verification of subsequent steps against the visual input, thereby translating external debiasing interventions into an intrinsic capability for hallucination mitigation.

Jingren Liu, Zhong Ji, YunLong Yu et al.

Parameter-efficient fine-tuning for continual learning (PEFT-CL) has shown promise in adapting pre-trained models to sequential tasks while mitigating catastrophic forgetting problem. However, understanding the mechanisms that dictate continual performance in this paradigm remains elusive. To unravel this mystery, we undertake a rigorous analysis of PEFT-CL dynamics to derive relevant metrics for continual scenarios using Neural Tangent Kernel (NTK) theory. With the aid of NTK as a mathematical analysis tool, we recast the challenge of test-time forgetting into the quantifiable generalization gaps during training, identifying three key factors that influence these gaps and the performance of PEFT-CL: training sample size, task-level feature orthogonality, and regularization. To address these challenges, we introduce NTK-CL, a novel framework that eliminates task-specific parameter storage while adaptively generating task-relevant features. Aligning with theoretical guidance, NTK-CL triples the feature representation of each sample, theoretically and empirically reducing the magnitude of both task-interplay and task-specific generalization gaps. Grounded in NTK analysis, our framework imposes an adaptive exponential moving average mechanism and constraints on task-level feature orthogonality, maintaining intra-task NTK forms while attenuating inter-task NTK forms. Ultimately, by fine-tuning optimizable parameters with appropriate regularization, NTK-CL achieves state-of-the-art performance on established PEFT-CL benchmarks. This work provides a theoretical foundation for understanding and improving PEFT-CL models, offering insights into the interplay between feature representation, task orthogonality, and generalization, contributing to the development of more efficient continual learning systems.

Ao Wang, Fengyuan Sun, Hui Chen et al.

Multimodal Large Language Models (MLLMs) have recently demonstrated strong performance across a wide range of vision-language tasks, garnering significant attention in the computer vision. However, their efficient deployment remains a substantial challenge due to high computational costs and memory requirements. Recognizing the redundancy of information within the vision modality, recent studies have explored methods for compressing visual tokens in MLLMs to enhance efficiency in a training-free manner. Despite their effectiveness, existing methods like Fast rely on the attention between visual tokens and prompt text tokens as the importance indicator, overlooking the relevance to response text and thus introducing perception bias. In this paper, we demonstrate that in MLLMs, the [CLS] token in the visual encoder inherently knows which visual tokens are important for MLLMs. Building on this prior, we introduce a simple yet effective method for train-free visual token compression, called VTC-CLS. Firstly, it leverages the attention score of the [CLS] token on visual tokens as an importance indicator for pruning visual tokens. Besides, we also explore ensembling the importance scores derived by the [CLS] token from different layers to capture the key visual information more comprehensively. Extensive experiments demonstrate that our VTC-CLS achieves the state-of-the-art performance across various tasks compared with baseline methods. It also brings notably less computational costs in a training-free manner, highlighting its effectiveness and superiority. Code and models are available at \url{https://github.com/THU-MIG/VTC-CLS}.

Yi Liu, Chengxin Li, Shoukun Xu et al.

Multi-modal fusion has played a vital role in multi-modal scene understanding. Most existing methods focus on cross-modal fusion involving two modalities, often overlooking more complex multi-modal fusion, which is essential for real-world applications like autonomous driving, where visible, depth, event, LiDAR, etc., are used. Besides, few attempts for multi-modal fusion, \emph{e.g.}, simple concatenation, cross-modal attention, and token selection, cannot well dig into the intrinsic shared and specific details of multiple modalities. To tackle the challenge, in this paper, we propose a Part-Whole Relational Fusion (PWRF) framework. For the first time, this framework treats multi-modal fusion as part-whole relational fusion. It routes multiple individual part-level modalities to a fused whole-level modality using the part-whole relational routing ability of Capsule Networks (CapsNets). Through this part-whole routing, our PWRF generates modal-shared and modal-specific semantics from the whole-level modal capsules and the routing coefficients, respectively. On top of that, modal-shared and modal-specific details can be employed to solve the issue of multi-modal scene understanding, including synthetic multi-modal segmentation and visible-depth-thermal salient object detection in this paper. Experiments on several datasets demonstrate the superiority of the proposed PWRF framework for multi-modal scene understanding. The source code has been released on https://github.com/liuyi1989/PWRF.

Ao Wang, Hui Chen, Lihao Liu et al.

Over the past years, YOLOs have emerged as the predominant paradigm in the field of real-time object detection owing to their effective balance between computational cost and detection performance. Researchers have explored the architectural designs, optimization objectives, data augmentation strategies, and others for YOLOs, achieving notable progress. However, the reliance on the non-maximum suppression (NMS) for post-processing hampers the end-to-end deployment of YOLOs and adversely impacts the inference latency. Besides, the design of various components in YOLOs lacks the comprehensive and thorough inspection, resulting in noticeable computational redundancy and limiting the model's capability. It renders the suboptimal efficiency, along with considerable potential for performance improvements. In this work, we aim to further advance the performance-efficiency boundary of YOLOs from both the post-processing and model architecture. To this end, we first present the consistent dual assignments for NMS-free training of YOLOs, which brings competitive performance and low inference latency simultaneously. Moreover, we introduce the holistic efficiency-accuracy driven model design strategy for YOLOs. We comprehensively optimize various components of YOLOs from both efficiency and accuracy perspectives, which greatly reduces the computational overhead and enhances the capability. The outcome of our effort is a new generation of YOLO series for real-time end-to-end object detection, dubbed YOLOv10. Extensive experiments show that YOLOv10 achieves state-of-the-art performance and efficiency across various model scales. For example, our YOLOv10-S is 1.8$\times$ faster than RT-DETR-R18 under the similar AP on COCO, meanwhile enjoying 2.8$\times$ smaller number of parameters and FLOPs. Compared with YOLOv9-C, YOLOv10-B has 46\% less latency and 25\% fewer parameters for the same performance.

Yufei Guo, Jing Ma, Tianlu Zhang et al.

Multimodal item embeddings are crucial for e-commerce item-to-item (I2I) retrieval, yet real-world product images often contain promotional overlays and background clutter that inject spurious visual cues and degrade retrieval robustness. This issue is particularly pronounced in MLRM-style pipelines, where a frozen vision encoder is connected to an LLM through a lightweight connector that must selectively aggregate visual tokens. We propose Text-Guided Q-Former (TGQ-Former), a text-guided visual representation learning framework that leverages structured metadata as semantic guidance for visual token extraction while preserving complementary visual evidence. Concretely, TGQ-Former employs a hybrid-query connector to disentangle metadata-anchored and exploratory visual streams, and introduces a lightweight reliability-aware dual-gated vector modulation module to adaptively calibrate their contributions under noisy inputs. Experiments on large-scale, real-world e-commerce datasets with full-pool retrieval show that TGQ-Former consistently outperforms strong connector baselines and end-to-end MLLMs. On average, it improves Hit Rate@100 (H@100) by 6.04%, demonstrating the effectiveness of text-guided visual encoding for robust multimodal retrieval.

Yi Liu, Chengxin Li, Xiaohui Dong et al.

Achieving joint learning of Salient Object Detection (SOD) and Camouflaged Object Detection (COD) is extremely challenging due to their distinct object characteristics, i.e., saliency and camouflage. The only preliminary research treats them as two contradictory tasks, training models on large-scale labeled data alternately for each task and assessing them independently. However, such task-specific mechanisms fail to meet real-world demands for addressing unknown tasks effectively. To address this issue, in this paper, we pioneer a task-agnostic framework to unify SOD and COD. To this end, inspired by the agreeable nature of binary segmentation for SOD and COD, we propose a Contrastive Distillation Paradigm (CDP) to distil the foreground from the background, facilitating the identification of salient and camouflaged objects amidst their surroundings. To probe into the contribution of our CDP, we design a simple yet effective contextual decoder involving the interval-layer and global context, which achieves an inference speed of 67 fps. Besides the supervised setting, our CDP can be seamlessly integrated into unsupervised settings, eliminating the reliance on extensive human annotations. Experiments on public SOD and COD datasets demonstrate the superiority of our proposed framework in both supervised and unsupervised settings, compared with existing state-of-the-art approaches. Code is available on https://github.com/liuyi1989/Seamless-Detection.

Ao Wang, Hui Chen, Zijia Lin et al.

Vision network designs, including Convolutional Neural Networks and Vision Transformers, have significantly advanced the field of computer vision. Yet, their complex computations pose challenges for practical deployments, particularly in real-time applications. To tackle this issue, researchers have explored various lightweight and efficient network designs. However, existing lightweight models predominantly leverage self-attention mechanisms and convolutions for token mixing. This dependence brings limitations in effectiveness and efficiency in the perception and aggregation processes of lightweight networks, hindering the balance between performance and efficiency under limited computational budgets. In this paper, we draw inspiration from the dynamic heteroscale vision ability inherent in the efficient human vision system and propose a ``See Large, Focus Small'' strategy for lightweight vision network design. We introduce LS (\textbf{L}arge-\textbf{S}mall) convolution, which combines large-kernel perception and small-kernel aggregation. It can efficiently capture a wide range of perceptual information and achieve precise feature aggregation for dynamic and complex visual representations, thus enabling proficient processing of visual information. Based on LS convolution, we present LSNet, a new family of lightweight models. Extensive experiments demonstrate that LSNet achieves superior performance and efficiency over existing lightweight networks in various vision tasks. Codes and models are available at https://github.com/jameslahm/lsnet.

Xiaoqi Zhuang, Jefersson A. Dos Santos, Jungong Han

Satellite image composition plays a critical role in remote sensing applications such as data augmentation, disaste simulation, and urban planning. We propose HarmoniDiff-RS, a training-free diffusion-based framework for harmonizing composite satellite images under diverse domain conditions. Our method aligns the source and target domains through a Latent Mean Shift operation that transfers radiometric characteristics between them. To balance harmonization and content preservation, we introduce a Timestep-wise Latent Fusion strategy by leveraging early inverted latents for high harmonization and late latents for semantic consistency to generate a set of composite candidates. A lightweight harmony classifier is trained to further automatically select the most coherent result among them. We also construct RSIC-H, a benchmark dataset for satellite image harmonization derived from fMoW, providing 500 paired composition samples. Experiments demonstrate that our method effectively performs satellite image composition, showing strong potential for scalable remote-sensing synthesis and simulation tasks. Code is available at: https://github.com/XiaoqiZhuang/HarmoniDiff-RS.

Xingyu Miao, Haoran Duan, Yang Long et al.

Score Distillation Sampling (SDS) has emerged as a prominent method for text-to-3D generation by leveraging the strengths of 2D diffusion models. However, SDS is limited to generation tasks and lacks the capability to edit existing 3D assets. Conversely, variants of SDS that introduce editing capabilities often can not generate new 3D assets effectively. In this work, we observe that the processes of generation and editing within SDS and its variants have unified underlying gradient terms. Building on this insight, we propose Unified Distillation Sampling (UDS), a method that seamlessly integrates both the generation and editing of 3D assets. Essentially, UDS refines the gradient terms used in vanilla SDS methods, unifying them to support both tasks. Extensive experiments demonstrate that UDS not only outperforms baseline methods in generating 3D assets with richer details but also excels in editing tasks, thereby bridging the gap between 3D generation and editing. The code is available on: https://github.com/xingy038/UDS.

Lihao Liu, Juexiao Feng, Hui Chen et al.

Traditional object detection models are constrained by the limitations of closed-set datasets, detecting only categories encountered during training. While multimodal models have extended category recognition by aligning text and image modalities, they introduce significant inference overhead due to cross-modality fusion and still remain restricted by predefined vocabulary, leaving them ineffective at handling unknown objects in open-world scenarios. In this work, we introduce Universal Open-World Object Detection (Uni-OWD), a new paradigm that unifies open-vocabulary and open-world object detection tasks. To address the challenges of this setting, we propose YOLO-UniOW, a novel model that advances the boundaries of efficiency, versatility, and performance. YOLO-UniOW incorporates Adaptive Decision Learning to replace computationally expensive cross-modality fusion with lightweight alignment in the CLIP latent space, achieving efficient detection without compromising generalization. Additionally, we design a Wildcard Learning strategy that detects out-of-distribution objects as "unknown" while enabling dynamic vocabulary expansion without the need for incremental learning. This design empowers YOLO-UniOW to seamlessly adapt to new categories in open-world environments. Extensive experiments validate the superiority of YOLO-UniOW, achieving achieving 34.6 AP and 30.0 APr on LVIS with an inference speed of 69.6 FPS. The model also sets benchmarks on M-OWODB, S-OWODB, and nuScenes datasets, showcasing its unmatched performance in open-world object detection. Code and models are available at https://github.com/THU-MIG/YOLO-UniOW.

Chenghao Xiao, Zhuoxu Huang, Danlu Chen et al.

Pretrained language models are long known to be subpar in capturing sentence and document-level semantics. Though heavily investigated, transferring perturbation-based methods from unsupervised visual representation learning to NLP remains an unsolved problem. This is largely due to the discreteness of subword units brought by tokenization of language models, limiting small perturbations of inputs to form semantics-preserved positive pairs. In this work, we conceptualize the learning of sentence-level textual semantics as a visual representation learning process. Drawing from cognitive and linguistic sciences, we introduce an unsupervised visual sentence representation learning framework, employing visually-grounded text perturbation methods like typos and word order shuffling, resonating with human cognitive patterns, and enabling perturbation to texts to be perceived as continuous. Our approach is further bolstered by large-scale unsupervised topical alignment training and natural language inference supervision, achieving comparable performance in semantic textual similarity (STS) to existing state-of-the-art NLP methods. Additionally, we unveil our method's inherent zero-shot cross-lingual transferability and a unique leapfrogging pattern across languages during iterative training. To our knowledge, this is the first representation learning method devoid of traditional language models for understanding sentence and document semantics, marking a stride closer to human-like textual comprehension. Our code is available at https://github.com/gowitheflow-1998/Pixel-Linguist

Ao Wang, Hui Chen, Jiaxin Li et al.

Recently, large vision-language models (LVLMs) have rapidly gained popularity for their strong generation and reasoning capabilities given diverse multimodal inputs. However, these models incur significant computational and memory overhead during inference, which greatly hinders the efficient deployment in practical scenarios. The extensive key-value (KV) cache, necessitated by the lengthy input and output sequences, notably contributes to the high inference cost. Based on this, recent works have investigated ways to reduce the KV cache size for higher efficiency. Although effective, they generally overlook the distinct importance distributions of KV vectors across layers and maintain the same cache size for each layer during the next token prediction. This results in the significant contextual information loss for certain layers, leading to notable performance decline. To address this, we present PrefixKV, where "Prefix" means the top-ranked KV based on importance rather than position in the original sequence. It reframes the challenge of determining KV cache sizes for all layers into the task of searching for the optimal global prefix configuration. With an adaptive layer-wise KV retention recipe based on binary search, the maximum contextual information can thus be preserved in each layer, facilitating the generation. Extensive experiments demonstrate that our method achieves the state-of-the-art performance compared with others. It exhibits superior inference efficiency and generation quality trade-offs, showing promising potential for practical applications. Code is available at https://github.com/THU-MIG/PrefixKV.

Leqi Shen, Guoqiang Gong, Tianxiang Hao et al.

The parameter-efficient adaptation of the image-text pretraining model CLIP for video-text retrieval is a prominent area of research. While CLIP is focused on image-level vision-language matching, video-text retrieval demands comprehensive understanding at the video level. Three key discrepancies emerge in the transfer from image-level to video-level: vision, language, and alignment. However, existing methods mainly focus on vision while neglecting language and alignment. In this paper, we propose Discrepancy Reduction in Vision, Language, and Alignment (DiscoVLA), which simultaneously mitigates all three discrepancies. Specifically, we introduce Image-Video Features Fusion to integrate image-level and video-level features, effectively tackling both vision and language discrepancies. Additionally, we generate pseudo image captions to learn fine-grained image-level alignment. To mitigate alignment discrepancies, we propose Image-to-Video Alignment Distillation, which leverages image-level alignment knowledge to enhance video-level alignment. Extensive experiments demonstrate the superiority of our DiscoVLA. In particular, on MSRVTT with CLIP (ViT-B/16), DiscoVLA outperforms previous methods by 1.5% in R@1, reaching a final score of 50.5% R@1. The code is available at https://github.com/LunarShen/DsicoVLA.

Yizhe Xiong, Wei Huang, Xin Ye et al.

Post-training is essential for adapting Large Language Models (LLMs) to real-world applications. Deploying post-trained models faces significant challenges due to substantial memory overhead and noticeable inference latency. Existing work has identified significant redundancies in LLMs and proposed efficient architectures, namely intra-layer KV sharing and cross-layer KV sharing. However, intra-layer KV sharing still results in high inference costs, while cross-layer KV sharing leads to significant performance degradation. As a result, both methods remain suboptimal for post-training pre-trained LLMs. In this paper, we identify that the \texttt{Softmax} operation is a primary bottleneck for LLM inference and discover that it is actually highly redundant during post-training. We propose Softmax \textbf{Uni}fication in \textbf{Att}e\textbf{n}tion (\textbf{UniAttn}), a novel post-training method that unifies Softmax activations across transformer blocks to reduce LLM inference costs. Additionally, UniAttn adopts a linear projection to compensate for the errors induced by Softmax unification. Experiments show that UniAttn matches the performance of standard post-training while significantly reducing inference costs, outperforming existing efficient architectures during post-training. Our code will be available at \url{https://github.com/Bostoncake/UniAttn}.

Sijie Li, Biao Qian, Jungong Han

Network pruning is an effective technique for enabling lightweight Large Vision-Language Models (LVLMs), which primarily incorporates both weights and activations into the importance metric. However, existing efforts typically process calibration data from different modalities in a unified manner, overlooking modality-specific behaviors. This raises a critical challenge: how to address the divergent behaviors of textual and visual tokens for accurate pruning of LVLMs. To this end, we systematically investigate the sensitivity of visual and textual tokens to the pruning operation by decoupling their corresponding weights, revealing that: (i) the textual pathway should be calibrated via text tokens, since it exhibits higher sensitivity than the visual pathway; (ii) the visual pathway exhibits high redundancy, permitting even 50% sparsity. Motivated by these insights, we propose a simple yet effective Asymmetric Text-Visual Weight Pruning method for LVLMs, dubbed ATV-Pruning, which establishes the importance metric for accurate weight pruning by selecting the informative tokens from both textual and visual pathways. Specifically, ATV-Pruning integrates two primary innovations: first, a calibration pool is adaptively constructed by drawing on all textual tokens and a subset of visual tokens; second, we devise a layer-adaptive selection strategy to yield important visual tokens. Finally, extensive experiments across standard multimodal benchmarks verify the superiority of our ATV-Pruning over state-of-the-art methods.

Yiwen Liang, Hui Chen, Yizhe Xiong et al.

Vision-language models (VLMs) exhibit remarkable zero-shot capabilities but struggle with distribution shifts in downstream tasks when labeled data is unavailable, which has motivated the development of Test-Time Adaptation (TTA) to improve VLMs' performance during inference without annotations. Among various TTA approaches, cache-based methods show promise by preserving historical knowledge from low-entropy samples in a dynamic cache and fostering efficient adaptation. However, these methods face two critical reliability challenges: (1) entropy often becomes unreliable under distribution shifts, causing error accumulation in the cache and degradation in adaptation performance; (2) the final predictions may be unreliable due to inflexible decision boundaries that fail to accommodate large downstream shifts. To address these challenges, we propose a Reliable Test-time Adaptation (ReTA) method that integrates two complementary strategies to enhance reliability from two perspectives. First, to mitigate the unreliability of entropy as a sample selection criterion for cache construction, we introduce Consistency-aware Entropy Reweighting (CER), which incorporates consistency constraints to weight entropy during cache updating. While conventional approaches rely solely on low entropy for cache prioritization and risk introducing noise, our method leverages predictive consistency to maintain a high-quality cache and facilitate more robust adaptation. Second, we present Diversity-driven Distribution Calibration (DDC), which models class-wise text embeddings as multivariate Gaussian distributions, enabling adaptive decision boundaries for more accurate predictions across visually diverse content. Extensive experiments demonstrate that ReTA consistently outperforms state-of-the-art methods, particularly under real-world distribution shifts. Code: https://github.com/Evelyn1ywliang/ReTA.

Jingkun Chen, Haoran Duan, Xiao Zhang et al.

Medical image segmentation remains challenging due to the high cost of pixel-level annotations for training. In the context of weak supervision, clinician gaze data captures regions of diagnostic interest; however, its sparsity limits its use for segmentation. In contrast, vision-language models (VLMs) provide semantic context through textual descriptions but lack the explanation precision required. Recognizing that neither source alone suffices, we propose a teacher-student framework that integrates both gaze and language supervision, leveraging their complementary strengths. Our key insight is that gaze data indicates where clinicians focus during diagnosis, while VLMs explain why those regions are significant. To implement this, the teacher model first learns from gaze points enhanced by VLM-generated descriptions of lesion morphology, establishing a foundation for guiding the student model. The teacher then directs the student through three strategies: (1) Multi-scale feature alignment to fuse visual cues with textual semantics; (2) Confidence-weighted consistency constraints to focus on reliable predictions; (3) Adaptive masking to limit error propagation in uncertain areas. Experiments on the Kvasir-SEG, NCI-ISBI, and ISIC datasets show that our method achieves Dice scores of 80.78%, 80.53%, and 84.22%, respectively-improving 3-5% over gaze baselines without increasing the annotation burden. By preserving correlations among predictions, gaze data, and lesion descriptions, our framework also maintains clinical interpretability. This work illustrates how integrating human visual attention with AI-generated semantic context can effectively overcome the limitations of individual weak supervision signals, thereby advancing the development of deployable, annotation-efficient medical AI systems. Code is available at: https://github.com/jingkunchen/FGI.

Yan Zhang, Zhong Ji, Changxu Meng et al.

Recent studies focus on the Remote Sensing Image-Text Retrieval (RSITR), which aims at searching for the corresponding targets based on the given query. Among these efforts, the application of Foundation Models (FMs), such as CLIP, to the domain of remote sensing has yielded encouraging outcomes. However, existing FM based methodologies neglect the negative impact of weakly correlated sample pairs and fail to account for the key distinctions among remote sensing texts, leading to biased and superficial exploration of sample pairs. To address these challenges, we propose an approach named iEBAKER (an Improved Eliminate Before Align strategy with Keyword Explicit Reasoning framework) for RSITR. Specifically, we propose an innovative Eliminate Before Align (EBA) strategy to filter out the weakly correlated sample pairs, thereby mitigating their deviations from optimal embedding space during alignment.Further, two specific schemes are introduced from the perspective of whether local similarity and global similarity affect each other. On this basis, we introduce an alternative Sort After Reversed Retrieval (SAR) strategy, aims at optimizing the similarity matrix via reverse retrieval. Additionally, we incorporate a Keyword Explicit Reasoning (KER) module to facilitate the beneficial impact of subtle key concept distinctions. Without bells and whistles, our approach enables a direct transition from FM to RSITR task, eliminating the need for additional pretraining on remote sensing data. Extensive experiments conducted on three popular benchmark datasets demonstrate that our proposed iEBAKER method surpasses the state-of-the-art models while requiring less training data. Our source code will be released at https://github.com/zhangy0822/iEBAKER.

Zhuoxu Huang, Mengxi Jia, Hao Sun et al.

Reinforcement Learning with verifiable rewards (RLVR) has emerged as a primary learning paradigm for enhancing the reasoning capabilities of multi-modal large language models (MLLMs). However, during RL training, the enormous state space of MLLM and sparse rewards often leads to entropy collapse, policy degradation, or over-exploitation of suboptimal behaviors. This necessitates an exploration strategy that maintains productive stochasticity while avoiding the drawbacks of uncontrolled random sampling, yielding inefficient exploration. In this paper, we propose CalibRL, a hybrid-policy RLVR framework that supports controllable exploration with expert guidance, enabled by two key mechanisms. First, a distribution-aware advantage weighting scales updates by group rareness to calibrate the distribution, therefore preserving exploration. Meanwhile, the asymmetric activation function (LeakyReLU) leverages the expert knowledge as a calibration baseline to moderate overconfident updates while preserving their corrective direction. CalibRL increases policy entropy in a guided manner and clarifies the target distribution by estimating the on-policy distribution through online sampling. Updates are driven by these informative behaviors, avoiding convergence to erroneous patterns. Importantly, these designs help alleviate the distributional mismatch between the model's policy and expert trajectories, thereby achieving a more stable balance between exploration and exploitation. Extensive experiments across eight benchmarks, including both in-domain and out-of-domain settings, demonstrate consistent improvements, validating the effectiveness of our controllable hybrid-policy RLVR training. Code is available at https://github.com/zhh6425/CalibRL.

Mingqi Gao, Jinyu Yang, Jingnan Luo et al.

Referring video object segmentation (RVOS) has recently generated great popularity in computer vision due to its widespread applications. Existing RVOS setting contains elaborately trimmed videos, with text-referred objects always appearing in all frames, which however fail to fully reflect the realistic challenges of this task. This simplified setting requires RVOS methods to only predict where objects, with no need to show when the objects appear. In this work, we introduce a new setting towards in-the-wild RVOS. To this end, we collect a new benchmark dataset using Youtube Untrimmed videos for RVOS - YoURVOS, which contains 1,120 in-the-wild videos with 7 times more duration and scenes than existing datasets. Our new benchmark challenges RVOS methods to show not only where but also when objects appear in videos. To set a baseline, we propose Object-level Multimodal TransFormers (OMFormer) to tackle the challenges, which are characterized by encoding object-level multimodal interactions for efficient and global spatial-temporal localisation. We demonstrate that previous VOS methods struggle on our YoURVOS benchmark, especially with the increase of target-absent frames, while our OMFormer consistently performs well. Our YoURVOS dataset offers an imperative benchmark, which will push forward the advancement of RVOS methods for practical applications.

Wensheng Wu, Zheming Lu, Ziqian Lu et al.

Industrial anomaly detection faces significant challenges due to the scarcity of anomalous samples and the complexity of real-world anomalies. In this paper, we propose a foundation model-based anomaly synthesis pipeline (FMAS) that generates highly realistic anomalous samples without fine-tuning or class-specific training. Motivated by the distinct frequency-domain characteristics of anomalies, we introduce aWavelet Domain Attention Module (WDAM), which exploits adaptive sub-band processing to enhance anomaly feature extraction. The combination of FMAS and WDAM significantly improves anomaly detection sensitivity while maintaining computational efficiency. Comprehensive experiments on MVTec AD and VisA datasets demonstrate that WDAM, as a plug-and-play module, achieves substantial performance gains against existing baselines.