Jinqiao Wang

Xu Zhao, Wenchao Ding, Yongqi An et al.

The recently proposed segment anything model (SAM) has made a significant influence in many computer vision tasks. It is becoming a foundation step for many high-level tasks, like image segmentation, image caption, and image editing. However, its huge computation costs prevent it from wider applications in industry scenarios. The computation mainly comes from the Transformer architecture at high-resolution inputs. In this paper, we propose a speed-up alternative method for this fundamental task with comparable performance. By reformulating the task as segments-generation and prompting, we find that a regular CNN detector with an instance segmentation branch can also accomplish this task well. Specifically, we convert this task to the well-studied instance segmentation task and directly train the existing instance segmentation method using only 1/50 of the SA-1B dataset published by SAM authors. With our method, we achieve a comparable performance with the SAM method at 50 times higher run-time speed. We give sufficient experimental results to demonstrate its effectiveness. The codes and demos will be released at https://github.com/CASIA-IVA-Lab/FastSAM.

Zhaopeng Gu, Bingke Zhu, Guibo Zhu et al.

Large Vision-Language Models (LVLMs) such as MiniGPT-4 and LLaVA have demonstrated the capability of understanding images and achieved remarkable performance in various visual tasks. Despite their strong abilities in recognizing common objects due to extensive training datasets, they lack specific domain knowledge and have a weaker understanding of localized details within objects, which hinders their effectiveness in the Industrial Anomaly Detection (IAD) task. On the other hand, most existing IAD methods only provide anomaly scores and necessitate the manual setting of thresholds to distinguish between normal and abnormal samples, which restricts their practical implementation. In this paper, we explore the utilization of LVLM to address the IAD problem and propose AnomalyGPT, a novel IAD approach based on LVLM. We generate training data by simulating anomalous images and producing corresponding textual descriptions for each image. We also employ an image decoder to provide fine-grained semantic and design a prompt learner to fine-tune the LVLM using prompt embeddings. Our AnomalyGPT eliminates the need for manual threshold adjustments, thus directly assesses the presence and locations of anomalies. Additionally, AnomalyGPT supports multi-turn dialogues and exhibits impressive few-shot in-context learning capabilities. With only one normal shot, AnomalyGPT achieves the state-of-the-art performance with an accuracy of 86.1%, an image-level AUC of 94.1%, and a pixel-level AUC of 95.3% on the MVTec-AD dataset. Code is available at https://github.com/CASIA-IVA-Lab/AnomalyGPT.

Yandu Sun, Zhiyan Hou, Haokai Ma et al.

Model merging offers a training-free way to combine multiple post-trained expert models, but merging experts obtained through reinforcement learning (RL) remains challenging. Existing spectral merging methods often assume that leading singular directions contain the main task signal, while lower-energy residual components can be compressed, selected, or attenuated to reduce interference. We find that this assumption does not hold for RL task vectors: after decomposing each task vector into a leading spectral head and a residual component, both parts can independently recover substantial behavior knowledge, while exhibiting different merging properties. The head is highly concentrated and informative but more prone to sharp cross-expert conflicts, whereas the residual component is more dispersed and provides a more stable basis for aggregation. Based on this observation, we propose ResMerge, a residual-based spectral merging framework for RL experts. ResMerge first constructs a stable residual backbone with Spherical Residual Consensus Adaptation, which estimates a reliability-weighted consensus direction on the Frobenius sphere. It then reintroduces leading-head information through a Lightweight Head Correction module gated by positive cross-expert agreement. Experiments across multiple RL expert groups and capability domains show that ResMerge better preserves expert capabilities than representative task-vector and spectral merging baselines. The implementation of ResMerge is publicly available at https://github.com/sunyd0303-cpu/ResMerge-release.

Yongqi An, Xu Zhao, Tao Yu et al.

Background subtraction (BGS) aims to extract all moving objects in the video frames to obtain binary foreground segmentation masks. Deep learning has been widely used in this field. Compared with supervised-based BGS methods, unsupervised methods have better generalization. However, previous unsupervised deep learning BGS algorithms perform poorly in sophisticated scenarios such as shadows or night lights, and they cannot detect objects outside the pre-defined categories. In this work, we propose an unsupervised BGS algorithm based on zero-shot object detection called Zero-shot Background Subtraction (ZBS). The proposed method fully utilizes the advantages of zero-shot object detection to build the open-vocabulary instance-level background model. Based on it, the foreground can be effectively extracted by comparing the detection results of new frames with the background model. ZBS performs well for sophisticated scenarios, and it has rich and extensible categories. Furthermore, our method can easily generalize to other tasks, such as abandoned object detection in unseen environments. We experimentally show that ZBS surpasses state-of-the-art unsupervised BGS methods by 4.70% F-Measure on the CDnet 2014 dataset. The code is released at https://github.com/CASIA-IVA-Lab/ZBS.

Zhiyang Chen, Yousong Zhu, Zhaowen Li et al.

Visual tasks vary a lot in their output formats and concerned contents, therefore it is hard to process them with an identical structure. One main obstacle lies in the high-dimensional outputs in object-level visual tasks. In this paper, we propose an object-centric vision framework, Obj2Seq. Obj2Seq takes objects as basic units, and regards most object-level visual tasks as sequence generation problems of objects. Therefore, these visual tasks can be decoupled into two steps. First recognize objects of given categories, and then generate a sequence for each of these objects. The definition of the output sequences varies for different tasks, and the model is supervised by matching these sequences with ground-truth targets. Obj2Seq is able to flexibly determine input categories to satisfy customized requirements, and be easily extended to different visual tasks. When experimenting on MS COCO, Obj2Seq achieves 45.7% AP on object detection, 89.0% AP on multi-label classification and 65.0% AP on human pose estimation. These results demonstrate its potential to be generally applied to different visual tasks. Code has been made available at: https://github.com/CASIA-IVA-Lab/Obj2Seq.

Kuan Zhu, Haiyun Guo, Tianyi Yan et al.

In person re-identification (ReID), very recent researches have validated pre-training the models on unlabelled person images is much better than on ImageNet. However, these researches directly apply the existing self-supervised learning (SSL) methods designed for image classification to ReID without any adaption in the framework. These SSL methods match the outputs of local views (e.g., red T-shirt, blue shorts) to those of the global views at the same time, losing lots of details. In this paper, we propose a ReID-specific pre-training method, Part-Aware Self-Supervised pre-training (PASS), which can generate part-level features to offer fine-grained information and is more suitable for ReID. PASS divides the images into several local areas, and the local views randomly cropped from each area are assigned with a specific learnable [PART] token. On the other hand, the [PART]s of all local areas are also appended to the global views. PASS learns to match the output of the local views and global views on the same [PART]. That is, the learned [PART] of the local views from a local area is only matched with the corresponding [PART] learned from the global views. As a result, each [PART] can focus on a specific local area of the image and extracts fine-grained information of this area. Experiments show PASS sets the new state-of-the-art performances on Market1501 and MSMT17 on various ReID tasks, e.g., vanilla ViT-S/16 pre-trained by PASS achieves 92.2\%/90.2\%/88.5\% mAP accuracy on Market1501 for supervised/UDA/USL ReID. Our codes are available at https://github.com/CASIA-IVA-Lab/PASS-reID.

Yufei Zhan, Yousong Zhu, Zhiyang Chen et al.

Replicating the innate human ability to detect all objects based on free-form texts at any granularity remains a formidable challenge for Large Vision Language Models (LVLMs). Current LVLMs are predominantly constrained to locate a single, pre-existing object. This limitation leads to a compromise in model design, necessitating the introduction of visual expert models or customized head structures. Beyond these constraints, our research uncovers LVLMs' capability for basic object perception, allowing them to accurately identify and locate objects of interest. Building on this insight, we introduce a novel Language-prompted Localization Dataset to fully unleash the capabilities of LVLMs in fine-grained object perception and precise location awareness. More importantly, we present Griffon, a purely LVLM-based baseline, which does not introduce any special tokens, expert models, or additional detection modules. It simply maintains a consistent structure with popular LVLMs by unifying data formats across various localization-related scenarios and is trained end-to-end through a well-designed pipeline. Comprehensive experiments demonstrate that Griffon not only achieves state-of-the-art performance on the fine-grained RefCOCO series and Flickr30K Entities but also approaches the capabilities of the expert model Faster RCNN on the detection benchmark MSCOCO. Data, codes, and models are released at https://github.com/jefferyZhan/Griffon.

Jinzhao Luo, Lu Zhou, Guibo Zhu et al.

Skeleton-based action recognition has become popular in recent years due to its efficiency and robustness. Most current methods adopt graph convolutional network (GCN) for topology modeling, but GCN-based methods are limited in long-distance correlation modeling and generalizability. In contrast, the potential of convolutional neural network (CNN) for topology modeling has not been fully explored. In this paper, we propose a novel CNN architecture, Temporal-Channel Topology Enhanced Network (TCTE-Net), to learn spatial and temporal topologies for skeleton-based action recognition. The TCTE-Net consists of two modules: the Temporal-Channel Focus module, which learns a temporal-channel focus matrix to identify the most critical feature representations, and the Dynamic Channel Topology Attention module, which dynamically learns spatial topological features, and fuses them with an attention mechanism to model long-distance channel-wise topology. We conduct experiments on NTU RGB+D, NTU RGB+D 120, and FineGym datasets. TCTE-Net shows state-of-the-art performance compared to CNN-based methods and achieves superior performance compared to GCN-based methods. The code is available at https://github.com/aikuniverse/TCTE-Net.

Zhaowen Li, Yousong Zhu, Fan Yang et al.

Self-supervised learning (SSL) holds promise in leveraging large amounts of unlabeled data. However, the success of popular SSL methods has limited on single-centric-object images like those in ImageNet and ignores the correlation among the scene and instances, as well as the semantic difference of instances in the scene. To address the above problems, we propose a Unified Self-supervised Visual Pre-training (UniVIP), a novel self-supervised framework to learn versatile visual representations on either single-centric-object or non-iconic dataset. The framework takes into account the representation learning at three levels: 1) the similarity of scene-scene, 2) the correlation of scene-instance, 3) the discrimination of instance-instance. During the learning, we adopt the optimal transport algorithm to automatically measure the discrimination of instances. Massive experiments show that UniVIP pre-trained on non-iconic COCO achieves state-of-the-art transfer performance on a variety of downstream tasks, such as image classification, semi-supervised learning, object detection and segmentation. Furthermore, our method can also exploit single-centric-object dataset such as ImageNet and outperforms BYOL by 2.5% with the same pre-training epochs in linear probing, and surpass current self-supervised object detection methods on COCO dataset, demonstrating its universality and potential.

Tianyue Wang, Leigang Qu, Tianyu Yang et al.

Zero-Shot Composed Image Retrieval (ZS-CIR) aims to retrieve target images given a multimodal query (comprising a reference image and a modification text), without training on annotated triplets. Existing methods typically convert the multimodal query into a single modality-either as an edited caption for Text-to-Image retrieval (T2I) or as an edited image for Image-to-Image retrieval (I2I). However, each paradigm has inherent limitations: T2I often loses fine-grained visual details, while I2I struggles with complex semantic modifications. To effectively leverage their complementary strengths under diverse query intents, we propose WISER, a training-free framework that unifies T2I and I2I via a "retrieve-verify-refine" pipeline, explicitly modeling intent awareness and uncertainty awareness. Specifically, WISER first performs Wider Search by generating both edited captions and images for parallel retrieval to broaden the candidate pool. Then, it conducts Adaptive Fusion with a verifier to assess retrieval confidence, triggering refinement for uncertain retrievals, and dynamically fusing the dual-path for reliable ones. For uncertain retrievals, WISER generates refinement suggestions through structured self-reflection to guide the next retrieval round toward Deeper Thinking. Extensive experiments demonstrate that WISER significantly outperforms previous methods across multiple benchmarks, achieving relative improvements of 45% on CIRCO (mAP@5) and 57% on CIRR (Recall@1) over existing training-free methods. Notably, it even surpasses many training-dependent methods, highlighting its superiority and generalization under diverse scenarios. Code will be released at https://github.com/Physicsmile/WISER.

Zhiyang Chen, Yousong Zhu, Yufei Zhan et al.

Large vision-language models (LVLMs) suffer from hallucination a lot, generating responses that apparently contradict to the image content occasionally. The key problem lies in its weak ability to comprehend detailed content in a multi-modal context, which can be mainly attributed to two factors in training data and loss function. The vision instruction dataset primarily focuses on global description, and the auto-regressive loss function favors text modeling rather than image understanding. In this paper, we bring more detailed vision annotations and more discriminative vision models to facilitate the training of LVLMs, so that they can generate more precise responses without encounter hallucination. On one hand, we generate image-text pairs with detailed relationship annotations in panoptic scene graph dataset (PSG). These conversations pay more attention on detailed facts in the image, encouraging the model to answer questions based on multi-modal contexts. On the other hand, we integrate SAM and mask prediction loss as auxiliary supervision, forcing the LVLMs to have the capacity to identify context-related objects, so that they can generate more accurate responses, mitigating hallucination. Moreover, to provide a deeper evaluation on the hallucination in LVLMs, we propose a new benchmark, RAH-Bench. It divides vision hallucination into three different types that contradicts the image with wrong categories, attributes or relations, and introduces False Positive Rate as detailed sub-metric for each type. In this benchmark, our approach demonstrates an +8.4% enhancement compared to original LLaVA and achieves widespread performance improvements across other models.

Gengsheng Li, Jinghan He, Shijie Wang et al.

Self-play bootstraps LLM reasoning through an iterative Challenger-Solver loop: the Challenger is trained to generate questions that target the Solver's capabilities, and the Solver is optimized on the generated data to expand its reasoning skills. However, existing frameworks like R-Zero often exhibit non-sustained improvement, where early gains degrade as self-play continues. We identify a key failure mode, Diversity Illusion, where the Solver's training signals appear diverse yet collapse into recurring underlying patterns. It manifests as (1) Local Diversity Illusion, where diversity is enforced only within-batch, inducing cross-iteration mode cycling; and (2) Surface Diversity Illusion, where questions vary superficially but require near-identical reasoning skills. To mitigate them, we propose R-Diverse with two aligned innovations: Memory-Augmented Penalty (MAP), which uses a persistent memory bank to discourage recycling across iterations, and Skill-Aware Measurement (SAM), which evaluates diversity by the reasoning skills exercised rather than surface variation of questions. Across 10 math and general reasoning benchmarks, R-Diverse sustains gains over more iterations and consistently outperforms prior self-play methods. Code is available at https://github.com/Gengsheng-Li/R-Diverse.

Jinghan He, Haiyun Guo, Ming Tang et al.

Instruction tuning is now a widely adopted approach to aligning large multimodal models (LMMs) to follow human intent. It unifies the data format of vision-language tasks, enabling multi-task joint training. However, vision-language tasks are constantly being created in practice. Instead of always re-training LMMs when new tasks arrive, continual learning offers flexibility for models to continually and efficiently exploit the evolving data. This work aims to explore the following two questions: 1) Do LMMs still suffer from catastrophic forgetting in continual instruction tuning? 2) Are the existing three classes of continual learning methods still applicable to the continual instruction tuning of LMMs? An extensive study is conducted to address the above questions. First, we establish the first benchmark in this setting and reveal that catastrophic forgetting is still observed when continually instruction-tuning LMMs. However, the multi-task joint instruction tuning can facilitate the model's continual learning ability and mitigate forgetting. Second, we integrate and adapt classic continual learning methods to our context, demonstrating the efficacy of data replay and model expansion strategies across diverse scenarios. In contrast, regularization-based methods only perform well on models that have been jointly instruction-tuned on multiple tasks. Third, we delve into the correlation and forgetting dynamics between vision-language task pairs and propose task-similarity-informed regularization and model expansion methods for continual instruction tuning of LMMs. Experimental results show that our approach consistently boosts the model's performance.

Fangxun Shu, Biaolong Chen, Yue Liao et al.

We present a simple yet effective end-to-end Video-language Pre-training (VidLP) framework, Masked Contrastive Video-language Pretraining (MAC), for video-text retrieval tasks. Our MAC aims to reduce video representation's spatial and temporal redundancy in the VidLP model by a mask sampling mechanism to improve pre-training efficiency. Comparing conventional temporal sparse sampling, we propose to randomly mask a high ratio of spatial regions and only feed visible regions into the encoder as sparse spatial sampling. Similarly, we adopt the mask sampling technique for text inputs for consistency. Instead of blindly applying the mask-then-prediction paradigm from MAE, we propose a masked-then-alignment paradigm for efficient video-text alignment. The motivation is that video-text retrieval tasks rely on high-level alignment rather than low-level reconstruction, and multimodal alignment with masked modeling encourages the model to learn a robust and general multimodal representation from incomplete and unstable inputs. Coupling these designs enables efficient end-to-end pre-training: reduce FLOPs (60% off), accelerate pre-training (by 3x), and improve performance. Our MAC achieves state-of-the-art results on various video-text retrieval datasets, including MSR-VTT, DiDeMo, and ActivityNet. Our approach is omnivorous to input modalities. With minimal modifications, we achieve competitive results on image-text retrieval tasks.

Zhaowen Li, Xu Zhao, Chaoyang Zhao et al.

Previous unsupervised domain adaptation methods did not handle the cross-domain problem from the perspective of frequency for computer vision. The images or feature maps of different domains can be decomposed into the low-frequency component and high-frequency component. This paper proposes the assumption that low-frequency information is more domain-invariant while the high-frequency information contains domain-related information. Hence, we introduce an approach, named low-frequency module (LFM), to extract domain-invariant feature representations. The LFM is constructed with the digital Gaussian low-pass filter. Our method is easy to implement and introduces no extra hyperparameter. We design two effective ways to utilize the LFM for domain adaptation, and our method is complementary to other existing methods and formulated as a plug-and-play unit that can be combined with these methods. Experimental results demonstrate that our LFM outperforms state-of-the-art methods for various computer vision tasks, including image classification and object detection.

Zhaowen Li, Xu Zhao, Peigeng Ding et al.

Recent researches indicate that utilizing the frequency information of input data can enhance the performance of networks. However, the existing popular convolutional structure is not designed specifically for utilizing the frequency information contained in datasets. In this paper, we propose a novel and effective module, named FreConv (frequency branch-and-integration convolution), to replace the vanilla convolution. FreConv adopts a dual-branch architecture to extract and integrate high- and low-frequency information. In the high-frequency branch, a derivative-filter-like architecture is designed to extract the high-frequency information while a light extractor is employed in the low-frequency branch because the low-frequency information is usually redundant. FreConv is able to exploit the frequency information of input data in a more reasonable way to enhance feature representation ability and reduce the memory and computational cost significantly. Without any bells and whistles, experimental results on various tasks demonstrate that FreConv-equipped networks consistently outperform state-of-the-art baselines.

Zhaowen Li, Yousong Zhu, Zhiyang Chen et al.

Inspired by the masked language modeling (MLM) in natural language processing tasks, the masked image modeling (MIM) has been recognized as a strong self-supervised pre-training method in computer vision. However, the high random mask ratio of MIM results in two serious problems: 1) the inadequate data utilization of images within each iteration brings prolonged pre-training, and 2) the high inconsistency of predictions results in unreliable generations, $i.e.$, the prediction of the identical patch may be inconsistent in different mask rounds, leading to divergent semantics in the ultimately generated outcomes. To tackle these problems, we propose the efficient masked autoencoders with self-consistency (EMAE) to improve the pre-training efficiency and increase the consistency of MIM. In particular, we present a parallel mask strategy that divides the image into K non-overlapping parts, each of which is generated by a random mask with the same mask ratio. Then the MIM task is conducted parallelly on all parts in an iteration and the model minimizes the loss between the predictions and the masked patches. Besides, we design the self-consistency learning to further maintain the consistency of predictions of overlapping masked patches among parts. Overall, our method is able to exploit the data more efficiently and obtains reliable representations. Experiments on ImageNet show that EMAE achieves the best performance on ViT-Large with only 13% of MAE pre-training time using NVIDIA A100 GPUs. After pre-training on diverse datasets, EMAE consistently obtains state-of-the-art transfer ability on a variety of downstream tasks, such as image classification, object detection, and semantic segmentation.

Tianyi Yan, Kuan Zhu, Haiyun guo et al.

Clustering-based methods, which alternate between the generation of pseudo labels and the optimization of the feature extraction network, play a dominant role in both unsupervised learning (USL) and unsupervised domain adaptive (UDA) person re-identification (Re-ID). To alleviate the adverse effect of noisy pseudo labels, the existing methods either abandon unreliable labels or refine the pseudo labels via mutual learning or label propagation. However, a great many erroneous labels are still accumulated because these methods mostly adopt traditional unsupervised clustering algorithms which rely on certain assumptions on data distribution and fail to capture the distribution of complex real-world data. In this paper, we propose the plug-and-play graph-based pseudo label correction network (GLC) to refine the pseudo labels in the manner of supervised clustering. GLC is trained to perceive the varying data distribution at each epoch of the self-training with the supervision of initial pseudo labels generated by any clustering method. It can learn to rectify the initial noisy labels by means of the relationship constraints between samples on the k Nearest Neighbor (kNN) graph and early-stop training strategy. Specifically, GLC learns to aggregate node features from neighbors and predict whether the nodes should be linked on the graph. Besides, GLC is optimized with 'early stop' before the noisy labels are severely memorized to prevent overfitting to noisy pseudo labels. Consequently, GLC improves the quality of pseudo labels though the supervision signals contain some noise, leading to better Re-ID performance. Extensive experiments in USL and UDA person Re-ID on Market-1501 and MSMT17 show that our method is widely compatible with various clustering-based methods and promotes the state-of-the-art performance consistently.

Ruiming Liang, Yinan Zheng, Kexin Zheng et al. · tsinghua

Diffusion-based policies have gained growing popularity in solving a wide range of decision-making tasks due to their superior expressiveness and controllable generation during inference. However, effectively training large diffusion policies using reinforcement learning (RL) remains challenging. Existing methods either suffer from unstable training due to directly maximizing value objectives, or face computational issues due to relying on crude Gaussian likelihood approximation, which requires a large amount of sufficiently small denoising steps. In this work, we propose DIPOLE (Dichotomous diffusion Policy improvement), a novel RL algorithm designed for stable and controllable diffusion policy optimization. We begin by revisiting the KL-regularized objective in RL, which offers a desirable weighted regression objective for diffusion policy extraction, but often struggles to balance greediness and stability. We then formulate a greedified policy regularization scheme, which naturally enables decomposing the optimal policy into a pair of stably learned dichotomous policies: one aims at reward maximization, and the other focuses on reward minimization. Under such a design, optimized actions can be generated by linearly combining the scores of dichotomous policies during inference, thereby enabling flexible control over the level of greediness.Evaluations in offline and offline-to-online RL settings on ExORL and OGBench demonstrate the effectiveness of our approach. We also use DIPOLE to train a large vision-language-action (VLA) model for end-to-end autonomous driving (AD) and evaluate it on the large-scale real-world AD benchmark NAVSIM, highlighting its potential for complex real-world applications.

Tuan-Hung Vu, Eduardo Valle, Andrei Bursuc et al.

We propose the unified BRAVO challenge to benchmark the reliability of semantic segmentation models under realistic perturbations and unknown out-of-distribution (OOD) scenarios. We define two categories of reliability: (1) semantic reliability, which reflects the model's accuracy and calibration when exposed to various perturbations; and (2) OOD reliability, which measures the model's ability to detect object classes that are unknown during training. The challenge attracted nearly 100 submissions from international teams representing notable research institutions. The results reveal interesting insights into the importance of large-scale pre-training and minimal architectural design in developing robust and reliable semantic segmentation models.

Jianghao Chen, Pu Jian, Tengxiao Xi et al.

During the development of large language models (LLMs), the scale and quality of the pre-training data play a crucial role in shaping LLMs' capabilities. To accelerate the research of LLMs, several large-scale datasets, such as C4 [1], Pile [2], RefinedWeb [3] and WanJuan [4], have been released to the public. However, most of the released corpus focus mainly on English, and there is still lack of complete tool-chain for extracting clean texts from web data. Furthermore, fine-grained information of the corpus, e.g. the quality of each text, is missing. To address these challenges, we propose in this paper a new complete tool-chain EvalWeb to extract Chinese clean texts from noisy web data. First, similar to previous work, manually crafted rules are employed to discard explicit noisy texts from the raw crawled web contents. Second, a well-designed evaluation model is leveraged to assess the remaining relatively clean data, and each text is assigned a specific quality score. Finally, we can easily utilize an appropriate threshold to select the high-quality pre-training data for Chinese. Using our proposed approach, we release the largest and latest large-scale high-quality Chinese web text ChineseWebText, which consists of 1.42 TB and each text is associated with a quality score, facilitating the LLM researchers to choose the data according to the desired quality thresholds. We also release a much cleaner subset of 600 GB Chinese data with the quality exceeding 90%.

Yongqi An, Chang Lu, Kuan Zhu et al.

Large language models (LLMs) face growing challenges in efficient generative inference due to the increasing memory demands of Key-Value (KV) caches, especially for long sequences. Existing eviction methods typically retain KV pairs with high attention weights but overlook the impact of attention redistribution caused by token removal, as well as the spatial-temporal dynamics in KV selection. In this paper, we propose ReST-KV, a robust KV eviction method that combines layer-wise output Reconstruction and Spatial-Temporal smoothing to provide a more comprehensive perspective for the KV cache eviction task. Specifically, ReST-KV formulates KV cache eviction as an optimization problem that minimizes output discrepancies through efficient layer-wise reconstruction. By directly modeling how each token's removal affects the model output, our method naturally captures attention redistribution effects, going beyond simplistic reliance on raw attention weights. To further enhance robustness, we design exponential moving average smoothing to handle temporal variations and an adaptive window-based mechanism to capture spatial patterns. Our method, ReST-KV, significantly advances performance on long-context benchmarks. It surpasses state-of-the-art baselines by 2.58% on LongBench and 15.2% on RULER. Additionally, ReST-KV consistently outperforms existing methods on Needle-in-a-Haystack and InfiniteBench, all while achieving a remarkable 10.61$\times$ reduction in decoding latency at 128k context length. The code is publicly available at https://github.com/an-yongqi/rest-kv to facilitate reproducibility and further research.

Xiangzhao Hao, Shijie Wang, Tianyu Yang et al.

Universal Multimodal Retrieval requires unified embedding models capable of interpreting diverse user intents, ranging from simple keywords to complex compositional instructions. While Multimodal Large Language Models (MLLMs) possess strong reasoning capabilities, prevailing adaptations confine them to static encoders, underutilizing their generative potential. This encoder-only paradigm struggles with complex intents that demand logical deduction rather than superficial pattern matching. To address this, we introduce TRACE (Task-adaptive Reasoning And Compressing Embeddings). TRACE unifies generative reasoning with discriminative representation learning. It first generates a structured Chain-of-Thought (CoT) to explicitly reason about the query, and subsequently compresses this reasoning trace into a compact embedding via a dedicated token. To train this framework, we construct M-BEIR-CoT, a large-scale dataset featuring a difficulty-aware routing strategy. Experiments on the M-BEIR benchmark establish TRACE as the new state-of-the-art. Crucially, TRACE demonstrates a learned implicit routing behavior. It autonomously activates reasoning for complex queries while bypassing it for simpler ones, achieving an optimal balance between retrieval accuracy and inference throughput. Furthermore, by internalizing the deductive process, TRACE exhibits remarkable zero-shot transferability to unseen domains and novel constraints.

Yongqi An, Xu Zhao, Tao Yu et al.

Network Pruning is a promising way to address the huge computing resource demands of the deployment and inference of Large Language Models (LLMs). Retraining-free is important for LLMs' pruning methods. However, almost all of the existing retraining-free pruning approaches for LLMs focus on unstructured pruning, which requires specific hardware support for acceleration. In this paper, we propose a novel retraining-free structured pruning framework for LLMs, named FLAP (FLuctuation-based Adaptive Structured Pruning). It is hardware-friendly by effectively reducing storage and enhancing inference speed. For effective structured pruning of LLMs, we highlight three critical elements that demand the utmost attention: formulating structured importance metrics, adaptively searching the global compressed model, and implementing compensation mechanisms to mitigate performance loss. First, FLAP determines whether the output feature map is easily recoverable when a column of weight is removed, based on the fluctuation pruning metric. Then it standardizes the importance scores to adaptively determine the global compressed model structure. At last, FLAP adds additional bias terms to recover the output feature maps using the baseline values. We thoroughly evaluate our approach on a variety of language benchmarks. Without any retraining, our method significantly outperforms the state-of-the-art methods, including LLM-Pruner and the extension of Wanda in structured pruning. The code is released at https://github.com/CASIA-IVA-Lab/FLAP.

Hao Chang, Zhihui Wang, Lingxiang Wu et al.

Low-altitude vision systems are becoming a critical infrastructure for smart city governance. However, existing object-centric perception paradigms and loosely coupled vision-language pipelines are still difficult to support management-oriented anomaly understanding required in real-world urban governance. To bridge this gap, we introduce GovLA-10K, the first management-oriented multi-modal benchmark for low-altitude intelligence, along with GovLA-Reasoner, a unified vision-language reasoning framework tailored for governance-aware aerial perception. Unlike existing studies that aim to exhaustively annotate all visible objects, GovLA-10K is deliberately designed around functionally salient targets that directly correspond to practical management needs, and further provides actionable management suggestions grounded in these observations. To effectively coordinate the fine-grained visual grounding with high-level contextual language reasoning, GovLA-Reasoner introduces an efficient feature adapter that implicitly coordinates discriminative representation sharing between the visual detector and the large language model (LLM). Extensive experiments show that our method significantly improves performance while avoiding the need of fine-tuning for any task-specific individual components. We believe our work offers a new perspective and foundation for future studies on management-aware low-altitude vision-language systems.

Zhaopeng Gu, Bingke Zhu, Guibo Zhu et al.

Zero-shot anomaly detection (ZSAD) methods entail detecting anomalies directly without access to any known normal or abnormal samples within the target item categories. Existing approaches typically rely on the robust generalization capabilities of multimodal pretrained models, computing similarities between manually crafted textual features representing "normal" or "abnormal" semantics and image features to detect anomalies and localize anomalous patches. However, the generic descriptions of "abnormal" often fail to precisely match diverse types of anomalies across different object categories. Additionally, computing feature similarities for single patches struggles to pinpoint specific locations of anomalies with various sizes and scales. To address these issues, we propose a novel ZSAD method called FiLo, comprising two components: adaptively learned Fine-Grained Description (FG-Des) and position-enhanced High-Quality Localization (HQ-Loc). FG-Des introduces fine-grained anomaly descriptions for each category using Large Language Models (LLMs) and employs adaptively learned textual templates to enhance the accuracy and interpretability of anomaly detection. HQ-Loc, utilizing Grounding DINO for preliminary localization, position-enhanced text prompts, and Multi-scale Multi-shape Cross-modal Interaction (MMCI) module, facilitates more accurate localization of anomalies of different sizes and shapes. Experimental results on datasets like MVTec and VisA demonstrate that FiLo significantly improves the performance of ZSAD in both detection and localization, achieving state-of-the-art performance with an image-level AUC of 83.9% and a pixel-level AUC of 95.9% on the VisA dataset. Code is available at https://github.com/CASIA-IVA-Lab/FiLo.

Tao Yu, Yongqi An, Kuan Zhu et al.

Large Language Models (LLMs) demonstrate impressive performance across natural language tasks but incur substantial computational and storage costs due to their scale. Post-training structured pruning offers an efficient solution. However, when few-shot calibration sets fail to adequately reflect the pretraining data distribution, existing methods exhibit limited generalization to downstream tasks. To address this issue, we propose Function-Aware Neuron Grouping (FANG), a post-training pruning framework that alleviates calibration bias by identifying and preserving neurons critical to specific function. FANG groups neurons with similar function based on the type of semantic context they process and prunes each group independently. During importance estimation within each group, tokens that strongly correlate with the functional role of the neuron group are given higher weighting. Additionally, FANG also preserves neurons that contribute across multiple context types. To achieve a better trade-off between sparsity and performance, it allocates sparsity to each block adaptively based on its functional complexity. Experiments show that FANG improves downstream accuracy while preserving language modeling performance. It achieves the state-of-the-art (SOTA) results when combined with FLAP and OBC, two representative pruning methods. Specifically, FANG outperforms FLAP and OBC by 1.5%--8.5% in average accuracy under 30% and 40% sparsity.

Kexin Ma, Jing Xiao, Chaofeng Chen et al.

Token pruning has emerged as an effective approach to reduce the substantial computational overhead of Large Vision-Language Models (LVLMs) by discarding less informative visual tokens while preserving performance. However, existing methods typically rely on individual attention sources from different LVLM components, resulting in incomplete and suboptimal pruning decisions due to biased attention distributions. To address this problem, we propose DeSAP, a novel Decoupled Similarity-Aware Pruning method for precise, task-aware token pruning within the visual encoder. Specifically, DeSAP introduces a decoupled similarity to capture fine-grained cross-modal relevance between visual features and text tokens, providing explicit task-related guidance for pruning. By integrating decoupled similarity with visual saliency signals derived from visual attention, DeSAP performs token pruning under the guidance of both task-related and visual cues, enabling robust pruning even under aggressive pruning ratios. Extensive experiments across diverse benchmarks and architectures show that DeSAP consistently outperforms SOTA methods in both accuracy and efficiency. On LLaVA-1.5-7B, DeSAP achieves a 10 times FLOPs reduction and a 2.3 times prefill speedup by retaining only 11.1% of visual tokens, while maintaining 98.1% of the original performance.

Yuanbing Zhu, Bingke Zhu, Yingying Chen et al.

Pretrained vision-language models (VLMs), \eg CLIP, are increasingly used to bridge the gap between open- and close-vocabulary recognition in open-vocabulary image segmentation. As VLMs are generally pretrained with low-resolution images (e.g. $224\times224$), most previous methods operate only on downscaled images. We question this design as low resolution features often fail to preserve fine details. A typical solution is to employ additional image backbones for high-resolution inputs, but it also introduce significant computation overhead. Therefore, we propose MROVSeg, a multi-resolution training framework for open-vocabulary image segmentation with a single pretrained CLIP backbone, that uses sliding windows to slice the high-resolution input into uniform patches, each matching the input size of the well-trained image encoder. Its key components include a Multi-Res Adapter, which restores the spatial geometry and grasps local-global correspondences across patches by interacting with multi-resolution features. To achieve accurate segmentation, we introduce Multi-grained Masked Attention scheme to aggregate multi-grained semantics from multi-resolution CLIP features to object queries. Through comprehensive experiments, we demonstrate the superiority of MROVSeg on well-established open-vocabulary image segmentation benchmarks, establishing new standards for open-vocabulary image segmentation.

Pengxiang Cai, Zhiwei Liu, Guibo Zhu et al.

Pixel-level fine-grained image editing remains an open challenge. Previous works fail to achieve an ideal trade-off between control granularity and inference speed. They either fail to achieve pixel-level fine-grained control, or their inference speed requires optimization. To address this, this paper for the first time employs a regression-based network to learn the variation patterns of StyleGAN latent codes during the image dragging process. This method enables pixel-level precision in dragging editing with little time cost. Users can specify handle points and their corresponding target points on any GAN-generated images, and our method will move each handle point to its corresponding target point. Through experimental analysis, we discover that a short movement distance from handle points to target points yields a high-fidelity edited image, as the model only needs to predict the movement of a small portion of pixels. To achieve this, we decompose the entire movement process into multiple sub-processes. Specifically, we develop a transformer encoder-decoder based network named 'Latent Predictor' to predict the latent code motion trajectories from handle points to target points in an autoregressive manner. Moreover, to enhance the prediction stability, we introduce a component named 'Latent Regularizer', aimed at constraining the latent code motion within the distribution of natural images. Extensive experiments demonstrate that our method achieves state-of-the-art (SOTA) inference speed and image editing performance at the pixel-level granularity.

Tianyu Yang, ChenWei He, Xiangzhao Hao et al.

Composed Image Retrieval (CIR) aims to retrieve target images based on a hybrid query comprising a reference image and a modification text. Early dual-tower Vision-Language Models (VLMs) struggle with cross-modality compositional reasoning required for this task. Recently, adapting generative Multimodal Large Language Models (MLLMs) for retrieval offers a promising direction. However, we identify that this adaptation strategy overlooks a fundamental issue: adapting a generative MLLM into a single-embedding discriminative retriever triggers a paradigm conflict, which leads to Capability Degradation - the deterioration of native fine-grained reasoning after retrieval adaptation. To address this challenge, we propose ReCALL (Recalibrating Capability Degradation), a model-agnostic framework that follows a diagnose-generate-refine pipeline: Firstly, we diagnose cognitive blind spots of the retriever via self-guided informative instance mining. Next, we generate corrective instructions and triplets by CoT prompting the foundation MLLM and conduct quality control with VQA-based consistency filtering. Finally, we refine the retriever through continual training on these triplets with a grouped contrastive scheme, thereby internalizing fine-grained visual-semantic distinctions and realigning the discriminative embedding space of retriever with intrinsic compositional reasoning within the MLLM. Extensive experiments on CIRR and FashionIQ show that ReCALL consistently recalibrates degraded capabilities and achieves state-of-the-art performance. Code will be released soon.

Zhen Chen, Yuhao Zhai, Jun Zhang et al.

To assist surgeons in the operating theatre, surgical phase recognition is critical for developing computer-assisted surgical systems, which requires comprehensive understanding of surgical videos. Although existing studies made great progress, there are still two significant limitations worthy of improvement. First, due to the compromise of resource consumption, frame-wise visual features are extracted by 2D networks and disregard spatial and temporal knowledge of surgical actions, which hinders subsequent inter-frame modeling for phase prediction. Second, these works simply utilize ordinary classification loss with one-hot phase labels to optimize the phase predictions, and cannot fully explore surgical videos under inadequate supervision. To overcome these two limitations, we propose a Surgical Temporal Action-aware Network with sequence Regularization, named STAR-Net, to recognize surgical phases more accurately from input videos. Specifically, we propose an efficient multi-scale surgical temporal action (MS-STA) module, which integrates visual features with spatial and temporal knowledge of surgical actions at the cost of 2D networks. Moreover, we devise the dual-classifier sequence regularization (DSR) to facilitate the training of STAR-Net by the sequence guidance of an auxiliary classifier with a smaller capacity. Our STAR-Net with MS-STA and DSR can exploit visual features of surgical actions with effective regularization, thereby leading to the superior performance of surgical phase recognition. Extensive experiments on a large-scale gastrectomy surgery dataset and the public Cholec80 benchmark prove that our STAR-Net significantly outperforms state-of-the-arts of surgical phase recognition.

Yicheng Xiao, Yu Chen, Haoxuan Ma et al.

While the Contrastive Language-Image Pretraining(CLIP) model has achieved remarkable success in a variety of downstream vison language understanding tasks, enhancing its capability for fine-grained image-text alignment remains an active research focus. To this end, most existing works adopt the strategy of explicitly increasing the granularity of visual information processing, e.g., incorporating visual prompts to guide the model focus on specific local regions within the image. Meanwhile, researches on Multimodal Large Language Models(MLLMs) have demonstrated that training with long and detailed textual descriptions can effectively improve the model's fine-grained vision-language alignment. However, the inherent token length limitation of CLIP's text encoder fundamentally limits CLIP to process more granular textual information embedded in long text sequences. To synergistically leverage the advantages of enhancing both visual and textual content processing granularity, we propose PixCLIP, a novel framework designed to concurrently accommodate visual prompt inputs and process lengthy textual descriptions. Specifically, we first establish an automated annotation pipeline capable of generating pixel-level localized, long-form textual descriptions for images. Utilizing this pipeline, we construct LongGRIT, a high-quality dataset comprising nearly 1.5 million samples. Secondly, we replace CLIP's original text encoder with the LLM and propose a three-branch pixel-text alignment learning framework, facilitating fine-grained alignment between image regions and corresponding textual descriptions at arbitrary granularity. Experiments demonstrate that PixCLIP showcases breakthroughs in pixel-level interaction and handling long-form texts, achieving state-of-the-art performance.

Zhaopeng Gu, Bingke Zhu, Guibo Zhu et al.

Visual Anomaly Detection (VAD) aims to identify abnormal samples in images that deviate from normal patterns, covering multiple domains, including industrial, logical, and medical fields. Due to the domain gaps between these fields, existing VAD methods are typically tailored to each domain, with specialized detection techniques and model architectures that are difficult to generalize across different domains. Moreover, even within the same domain, current VAD approaches often follow a "one-category-one-model" paradigm, requiring large amounts of normal samples to train class-specific models, resulting in poor generalizability and hindering unified evaluation across domains. To address this issue, we propose a generalized few-shot VAD method, UniVAD, capable of detecting anomalies across various domains, such as industrial, logical, and medical anomalies, with a training-free unified model. UniVAD only needs few normal samples as references during testing to detect anomalies in previously unseen objects, without training on the specific domain. Specifically, UniVAD employs a Contextual Component Clustering ($C^3$) module based on clustering and vision foundation models to segment components within the image accurately, and leverages Component-Aware Patch Matching (CAPM) and Graph-Enhanced Component Modeling (GECM) modules to detect anomalies at different semantic levels, which are aggregated to produce the final detection result. We conduct experiments on nine datasets spanning industrial, logical, and medical fields, and the results demonstrate that UniVAD achieves state-of-the-art performance in few-shot anomaly detection tasks across multiple domains, outperforming domain-specific anomaly detection models. Code is available at https://github.com/FantasticGNU/UniVAD.

Xin He, Haiyun Guo, Kuan Zhu et al.

Lane detection plays an important role in autonomous driving perception systems. As deep learning algorithms gain popularity, monocular lane detection methods based on them have demonstrated superior performance and emerged as a key research direction in autonomous driving perception. The core designs of these algorithmic frameworks can be summarized as follows: (1) Task paradigm, focusing on lane instance-level discrimination; (2) Lane modeling, representing lanes as a set of learnable parameters in the neural network; (3) Global context supplementation, enhancing inference on the obscure lanes; (4) Perspective effect elimination, providing accurate 3D lanes for downstream applications. From these perspectives, this paper presents a comprehensive overview of existing methods, encompassing both the increasingly mature 2D lane detection approaches and the developing 3D lane detection works. Besides, this paper compares the performance of mainstream methods on different benchmarks and investigates their inference speed under a unified setting for fair comparison. Moreover, we present some extended works on lane detection, including multi-task perception, video lane detection, online high-definition map construction, and lane topology reasoning, to offer readers a comprehensive roadmap for the evolution of lane detection. Finally, we point out some potential future research directions in this field. We exhaustively collect the papers and codes of existing works at https://github.com/Core9724/Awesome-Lane-Detection and will keep tracing the research.

Yongqi An, Xu Zhao, Tao Yu et al.

Outliers have been widely observed in Large Language Models (LLMs), significantly impacting model performance and posing challenges for model compression. Understanding the functionality and formation mechanisms of these outliers is critically important. Existing works, however, largely focus on reducing the impact of outliers from an algorithmic perspective, lacking an in-depth investigation into their causes and roles. In this work, we provide a detailed analysis of the formation process, underlying causes, and functions of outliers in LLMs. We define and categorize three types of outliers-activation outliers, weight outliers, and attention outliers-and analyze their distributions across different dimensions, uncovering inherent connections between their occurrences and their ultimate influence on the attention mechanism. Based on these observations, we hypothesize and explore the mechanisms by which these outliers arise and function, demonstrating through theoretical derivations and experiments that they emerge due to the self-attention mechanism's softmax operation. These outliers act as implicit context-aware scaling factors within the attention mechanism. As these outliers stem from systematic influences, we term them systematic outliers. Our study not only enhances the understanding of Transformer-based LLMs but also shows that structurally eliminating outliers can accelerate convergence and improve model compression. The code is avilable at https://github.com/an-yongqi/systematic-outliers.

Yufei Zhan, Ziheng Wu, Yousong Zhu et al.

Despite notable advancements in multimodal reasoning, leading Multimodal Large Language Models (MLLMs) still underperform on vision-centric multimodal reasoning tasks in general scenarios. This shortfall stems from their predominant reliance on logic- and knowledge-based slow thinking strategies, while effective for domains like math and science, fail to integrate visual information effectively during reasoning. Consequently, these models often fail to adequately ground visual cues, resulting in suboptimal performance in tasks that require multiple plausible visual interpretations and inferences. To address this, we present GThinker (General Thinker), a novel reasoning MLLM excelling in multimodal reasoning across general scenarios, mathematics, and science. GThinker introduces Cue-Rethinking, a flexible reasoning pattern that grounds inferences in visual cues and iteratively reinterprets these cues to resolve inconsistencies. Building on this pattern, we further propose a two-stage training pipeline, including pattern-guided cold start and incentive reinforcement learning, designed to enable multimodal reasoning capabilities across domains. Furthermore, to support the training, we construct GThinker-11K, comprising 7K high-quality, iteratively-annotated reasoning paths and 4K curated reinforcement learning samples, filling the data gap toward general multimodal reasoning. Extensive experiments demonstrate that GThinker achieves 81.5% on the challenging comprehensive multimodal reasoning benchmark M$^3$CoT, surpassing the latest O4-mini model. It also shows an average improvement of 2.1% on general scenario multimodal reasoning benchmarks, while maintaining on-par performance in mathematical reasoning compared to counterpart advanced reasoning models. The code, model, and data will be released soon at https://github.com/jefferyZhan/GThinker.

Yunfang Niu, Lingxiang Wu, Dong Yi et al.

Fashion image editing aims to modify a person's appearance based on a given instruction. Existing methods require auxiliary tools like segmenters and keypoint extractors, lacking a flexible and unified framework. Moreover, these methods are limited in the variety of clothing types they can handle, as most datasets focus on people in clean backgrounds and only include generic garments such as tops, pants, and dresses. These limitations restrict their applicability in real-world scenarios. In this paper, we first extend an existing dataset for human generation to include a wider range of apparel and more complex backgrounds. This extended dataset features people wearing diverse items such as tops, pants, dresses, skirts, headwear, scarves, shoes, socks, and bags. Additionally, we propose AnyDesign, a diffusion-based method that enables mask-free editing on versatile areas. Users can simply input a human image along with a corresponding prompt in either text or image format. Our approach incorporates Fashion DiT, equipped with a Fashion-Guidance Attention (FGA) module designed to fuse explicit apparel types and CLIP-encoded apparel features. Both Qualitative and quantitative experiments demonstrate that our method delivers high-quality fashion editing and outperforms contemporary text-guided fashion editing methods.

Shurong Zheng, Yousong Zhu, Hongyin Zhao et al.

Multimodal Large Language Models (MLLMs) have demonstrated impressive progress in single-image grounding and general multi-image understanding. Recently, some methods begin to address multi-image grounding. However, they are constrained by single-target localization and limited types of practical tasks, due to the lack of unified modeling for generalized grounding tasks. Therefore, we propose GeM-VG, an MLLM capable of Generalized Multi-image Visual Grounding. To support this, we systematically categorize and organize existing multi-image grounding tasks according to their reliance of cross-image cues and reasoning, and introduce the MG-Data-240K dataset, addressing the limitations of existing datasets regarding target quantity and image relation. To tackle the challenges of robustly handling diverse multi-image grounding tasks, we further propose a hybrid reinforcement finetuning strategy that integrates chain-of-thought (CoT) reasoning and direct answering, considering their complementary strengths. This strategy adopts an R1-like algorithm guided by a carefully designed rule-based reward, effectively enhancing the model's overall perception and reasoning capabilities. Extensive experiments demonstrate the superior generalized grounding capabilities of our model. For multi-image grounding, it outperforms the previous leading MLLMs by 2.0% and 9.7% on MIG-Bench and MC-Bench, respectively. In single-image grounding, it achieves a 9.1% improvement over the base model on ODINW. Furthermore, our model retains strong capabilities in general multi-image understanding.

Chen Wang, Tianyu Peng, Wen Yang et al.

Empathetic interaction is a cornerstone of human-machine communication, due to the need for understanding speech enriched with paralinguistic cues and generating emotional and expressive responses. However, the most powerful empathetic LSLMs are increasingly closed off, leaving the crucial details about the architecture, data and development opaque to researchers. Given the critical need for transparent research into the LSLMs and empathetic behavior, we present OpenS2S, a fully open-source, transparent and end-to-end LSLM designed to enable empathetic speech interactions. Based on our empathetic speech-to-text model BLSP-Emo, OpenS2S further employs a streaming interleaved decoding architecture to achieve low-latency speech generation. To facilitate end-to-end training, OpenS2S incorporates an automated data construction pipeline that synthesizes diverse, high-quality empathetic speech dialogues at low cost. By leveraging large language models to generate empathetic content and controllable text-to-speech systems to introduce speaker and emotional variation, we construct a scalable training corpus with rich paralinguistic diversity and minimal human supervision. We release the fully open-source OpenS2S model, including the dataset, model weights, pre-training and fine-tuning codes, to empower the broader research community and accelerate innovation in empathetic speech systems. The project webpage can be accessed at https://casia-lm.github.io/OpenS2S

Jiachen Yu, Yufei Zhan, Ziheng Wu et al.

Recent extensive works have demonstrated that by introducing long CoT, the capabilities of MLLMs to solve complex problems can be effectively enhanced. However, the reasons for the effectiveness of such paradigms remain unclear. It is challenging to analysis with quantitative results how much the model's specific extraction of visual cues and its subsequent so-called reasoning during inference process contribute to the performance improvements. Therefore, evaluating the faithfulness of MLLMs' reasoning to visual information is crucial. To address this issue, we first present a cue-driven automatic and controllable editing pipeline with the help of GPT-Image-1. It enables the automatic and precise editing of specific visual cues based on the instruction. Furthermore, we introduce VFaith-Bench, the first benchmark to evaluate MLLMs' visual reasoning capabilities and analyze the source of such capabilities with an emphasis on the visual faithfulness. Using the designed pipeline, we constructed comparative question-answer pairs by altering the visual cues in images that are crucial for solving the original reasoning problem, thereby changing the question's answer. By testing similar questions with images that have different details, the average accuracy reflects the model's visual reasoning ability, while the difference in accuracy before and after editing the test set images effectively reveals the relationship between the model's reasoning ability and visual perception. We further designed specific metrics to expose this relationship. VFaith-Bench includes 755 entries divided into five distinct subsets, along with an additional human-labeled perception task. We conducted in-depth testing and analysis of existing mainstream flagship models and prominent open-source model series/reasoning models on VFaith-Bench, further investigating the underlying factors of their reasoning capabilities.

Zhaopeng Gu, Bingke Zhu, Guibo Zhu et al.

Anomaly detection methods typically require extensive normal samples from the target class for training, limiting their applicability in scenarios that require rapid adaptation, such as cold start. Zero-shot and few-shot anomaly detection do not require labeled samples from the target class in advance, making them a promising research direction. Existing zero-shot and few-shot approaches often leverage powerful multimodal models to detect and localize anomalies by comparing image-text similarity. However, their handcrafted generic descriptions fail to capture the diverse range of anomalies that may emerge in different objects, and simple patch-level image-text matching often struggles to localize anomalous regions of varying shapes and sizes. To address these issues, this paper proposes the FiLo++ method, which consists of two key components. The first component, Fused Fine-Grained Descriptions (FusDes), utilizes large language models to generate anomaly descriptions for each object category, combines both fixed and learnable prompt templates and applies a runtime prompt filtering method, producing more accurate and task-specific textual descriptions. The second component, Deformable Localization (DefLoc), integrates the vision foundation model Grounding DINO with position-enhanced text descriptions and a Multi-scale Deformable Cross-modal Interaction (MDCI) module, enabling accurate localization of anomalies with various shapes and sizes. In addition, we design a position-enhanced patch matching approach to improve few-shot anomaly detection performance. Experiments on multiple datasets demonstrate that FiLo++ achieves significant performance improvements compared with existing methods. Code will be available at https://github.com/CASIA-IVA-Lab/FiLo.

Fan Yang, Shurong Zheng, Hongyin Zhao et al.

Recent Large Vision-Language Models (LVLMs) demonstrate remarkable capabilities in image understanding and natural language generation. However, current approaches focus predominantly on global image understanding, struggling to simulate human visual attention trajectories and explain associations between descriptions and specific regions. We propose TraceVision, a unified vision-language model integrating trajectory-aware spatial understanding in an end-to-end framework. TraceVision employs a Trajectory-aware Visual Perception (TVP) module for bidirectional fusion of visual features and trajectory information. We design geometric simplification to extract semantic keypoints from raw trajectories and propose a three-stage training pipeline where trajectories guide description generation and region localization. We extend TraceVision to trajectory-guided segmentation and video scene understanding, enabling cross-frame tracking and temporal attention analysis. We construct the Reasoning-based Interactive Localized Narratives (RILN) dataset to enhance logical reasoning and interpretability. Extensive experiments on trajectory-guided captioning, text-guided trajectory prediction, understanding, and segmentation demonstrate that TraceVision achieves state-of-the-art performance, establishing a foundation for intuitive spatial interaction and interpretable visual understanding.

Xingyu Ma, Xin Tian, Lingxiang Wu et al.

Text-to-SQL is a subtask in semantic parsing that has seen rapid progress with the evolution of Large Language Models (LLMs). However, LLMs face challenges due to hallucination issues and a lack of domain-specific database knowledge(such as table schema and cell values). As a result, they can make errors in generating table names, columns, and matching values to the correct columns in SQL statements. This paper introduces a method of knowledge injection to enhance LLMs' ability to understand schema contents by incorporating prior knowledge. This approach improves their performance in Text-to-SQL tasks. Experimental results show that pre-training LLMs on domain-specific database knowledge and fine-tuning them on downstream Text-to-SQL tasks significantly improves the Execution Match (EX) and Exact Match (EM) metrics across various models. This effectively reduces errors in generating column names and matching values to the columns. Furthermore, the knowledge-injected models can be applied to many downstream Text-to-SQL tasks, demonstrating the generalizability of the approach presented in this paper.

Linge Wang, Yingying Chen, Bingke Zhu et al.

Recent advances in audio-visual representation learning have shown the value of combining contrastive alignment with masked reconstruction. However, jointly optimizing these objectives in a single forward pass forces the contrastive branch to rely on randomly visible patches designed for reconstruction rather than cross-modal alignment, introducing semantic noise and optimization interference. We propose TG-DP, a Teacher-Guided Dual-Path framework that decouples reconstruction and alignment into separate optimization paths. By disentangling the masking regimes of the two branches, TG-DP enables the contrastive pathway to use a visibility pattern better suited to cross-modal alignment. A teacher model further provides auxiliary guidance for organizing visible tokens in this branch, helping reduce interference and stabilize cross-modal representation learning. TG-DP achieves state-of-the-art performance in zero-shot retrieval. On AudioSet, it improves R@1 from 35.2\% to 37.4\% for video-to-audio retrieval and from 27.9\% to 37.1\% for audio-to-video retrieval. The learned representations also remain semantically robust, achieving state-of-the-art linear-probe performance on AS20K and VGGSound. Taken together, our results suggest that decoupling multimodal objectives and introducing teacher-guided structure into the contrastive pathway provide an effective framework for improving large-scale audio-visual pretraining. Code is available at https://github.com/wanglg20/TG-DP.

Langyu Wang, Bingke Zhu, Yingying Chen et al.

The weakly-supervised audio-visual video parsing (AVVP) aims to predict all modality-specific events and locate their temporal boundaries. Despite significant progress, due to the limitations of the weakly-supervised and the deficiencies of the model architecture, existing methods are lacking in simultaneously improving both the segment-level prediction and the event-level prediction. In this work, we propose a audio-visual Mamba network with pseudo labeling aUGmentation (MUG) for emphasising the uniqueness of each segment and excluding the noise interference from the alternate modalities. Specifically, we annotate some of the pseudo-labels based on previous work. Using unimodal pseudo-labels, we perform cross-modal random combinations to generate new data, which can enhance the model's ability to parse various segment-level event combinations. For feature processing and interaction, we employ a audio-visual mamba network. The AV-Mamba enhances the ability to perceive different segments and excludes additional modal noise while sharing similar modal information. Our extensive experiments demonstrate that MUG improves state-of-the-art results on LLP dataset in all metrics (e.g,, gains of 2.1% and 1.2% in terms of visual Segment-level and audio Segment-level metrics). Our code is available at https://github.com/WangLY136/MUG.

Yufei Zhan, Hongyin Zhao, Yousong Zhu et al.

Large Multimodal Models (LMMs) have recently demonstrated remarkable visual understanding performance on both vision-language and vision-centric tasks. However, they often fall short in integrating advanced, task-specific capabilities for compositional reasoning, which hinders their progress toward truly competent general vision models. To address this, we present a unified visual reasoning mechanism that enables LMMs to solve complicated compositional problems by leveraging their intrinsic capabilities (e.g. grounding and visual understanding capabilities). Different from the previous shortcut learning mechanism, our approach introduces a human-like understanding-thinking-answering process, allowing the model to complete all steps in a single pass forwarding without the need for multiple inferences or external tools. This design bridges the gap between foundational visual capabilities and general question answering, encouraging LMMs to generate faithful and traceable responses for complex visual reasoning. Meanwhile, we curate 334K visual instruction samples covering both general scenes and text-rich scenes and involving multiple foundational visual capabilities. Our trained model, Griffon-R, has the ability of end-to-end automatic understanding, self-thinking, and reasoning answers. Comprehensive experiments show that Griffon-R not only achieves advancing performance on complex visual reasoning benchmarks including VSR and CLEVR, but also enhances multimodal capabilities across various benchmarks like MMBench and ScienceQA. Data, models, and codes will be release at https://github.com/jefferyZhan/Griffon/tree/master/Griffon-R soon.

Guojing Ge, Qi Song, Guibo Zhu et al.

Blind face restoration is a challenging task due to the unknown and complex degradation. Although face prior-based methods and reference-based methods have recently demonstrated high-quality results, the restored images tend to contain over-smoothed results and lose identity-preserved details when the degradation is severe. It is observed that this is attributed to short-range dependencies, the intrinsic limitation of convolutional neural networks. To model long-range dependencies, we propose a Transformer-based blind face restoration method, named BFRFormer, to reconstruct images with more identity-preserved details in an end-to-end manner. In BFRFormer, to remove blocking artifacts, the wavelet discriminator and aggregated attention module are developed, and spectral normalization and balanced consistency regulation are adaptively applied to address the training instability and over-fitting problem, respectively. Extensive experiments show that our method outperforms state-of-the-art methods on a synthetic dataset and four real-world datasets. The source code, Casia-Test dataset, and pre-trained models are released at https://github.com/s8Znk/BFRFormer.

Yuqi Pan, Yongqi An, Zheng Li et al.

The Transformer architecture, despite its widespread success, struggles with long-context scenarios due to quadratic computation and linear memory growth. While various linear attention variants mitigate these efficiency constraints by compressing context into fixed-size states, they often degrade performance in tasks such as in-context retrieval and reasoning. To address this limitation and achieve more effective context compression, we propose two key innovations. First, we introduce a row-sparse update formulation for linear attention by conceptualizing state updating as information classification. This enables sparse state updates via softmax-based top-$k$ hard classification, thereby extending receptive fields and reducing inter-class interference. Second, we present Sparse State Expansion (SSE) within the sparse framework, which expands the contextual state into multiple partitions, effectively decoupling parameter size from state capacity while maintaining the sparse classification paradigm. Supported by efficient parallelized implementations, our design achieves effective classification and highly discriminative state representations. We extensively validate SSE in both pure linear and hybrid (SSE-H) architectures across language modeling, in-context retrieval, and mathematical reasoning benchmarks. SSE demonstrates strong retrieval performance and scales favorably with state size. Moreover, after reinforcement learning (RL) training, our 2B SSE-H model achieves state-of-the-art mathematical reasoning performance among small reasoning models, scoring 64.5 on AIME24 and 50.2 on AIME25, significantly outperforming similarly sized open-source Transformers. These results highlight SSE as a promising and efficient architecture for long-context modeling.

Xiangzhao Hao, Kuan Zhu, Hongyu Guo et al.

Using natural language to query visual information is a fundamental need in real-world applications. Text-Image Retrieval (TIR) retrieves a target image from a gallery based on an image-level description, while Referring Expression Comprehension (REC) localizes a target object within a given image using an instance-level description. However, real-world applications often present more complex demands. Users typically query an instance-level description across a large gallery and expect to receive both relevant image and the corresponding instance location. In such scenarios, TIR struggles with fine-grained descriptions and object-level localization, while REC is limited in its ability to efficiently search large galleries and lacks an effective ranking mechanism. In this paper, we introduce a new task called \textbf{Referring Expression Instance Retrieval (REIR)}, which supports both instance-level retrieval and localization based on fine-grained referring expressions. First, we propose a large-scale benchmark for REIR, named REIRCOCO, constructed by prompting advanced vision-language models to generate high-quality referring expressions for instances in the MSCOCO and RefCOCO datasets. Second, we present a baseline method, Contrastive Language-Instance Alignment with Relation Experts (CLARE), which employs a dual-stream architecture to address REIR in an end-to-end manner. Given a referring expression, the textual branch encodes it into a query embedding. The visual branch detects candidate objects and extracts their instance-level visual features. The most similar candidate to the query is selected for bounding box prediction. CLARE is first trained on object detection and REC datasets to establish initial grounding capabilities, then optimized via Contrastive Language-Instance Alignment (CLIA) for improved retrieval across images. We will release our code and benchmark publicly.