Liang Wan

Zhaohu Xing, Liang Wan, Huazhu Fu et al.

In recent years, Denoising Diffusion Models have demonstrated remarkable success in generating semantically valuable pixel-wise representations for image generative modeling. In this study, we propose a novel end-to-end framework, called Diff-UNet, for medical volumetric segmentation. Our approach integrates the diffusion model into a standard U-shaped architecture to extract semantic information from the input volume effectively, resulting in excellent pixel-level representations for medical volumetric segmentation. To enhance the robustness of the diffusion model's prediction results, we also introduce a Step-Uncertainty based Fusion (SUF) module during inference to combine the outputs of the diffusion models at each step. We evaluate our method on three datasets, including multimodal brain tumors in MRI, liver tumors, and multi-organ CT volumes, and demonstrate that Diff-UNet outperforms other state-of-the-art methods significantly. Our experimental results also indicate the universality and effectiveness of the proposed model. The proposed framework has the potential to facilitate the accurate diagnosis and treatment of medical conditions by enabling more precise segmentation of anatomical structures. The codes of Diff-UNet are available at https://github.com/ge-xing/Diff-UNet

Zhaohu Xing, Lequan Yu, Liang Wan et al.

Multi-modal MR imaging is routinely used in clinical practice to diagnose and investigate brain tumors by providing rich complementary information. Previous multi-modal MRI segmentation methods usually perform modal fusion by concatenating multi-modal MRIs at an early/middle stage of the network, which hardly explores non-linear dependencies between modalities. In this work, we propose a novel Nested Modality-Aware Transformer (NestedFormer) to explicitly explore the intra-modality and inter-modality relationships of multi-modal MRIs for brain tumor segmentation. Built on the transformer-based multi-encoder and single-decoder structure, we perform nested multi-modal fusion for high-level representations of different modalities and apply modality-sensitive gating (MSG) at lower scales for more effective skip connections. Specifically, the multi-modal fusion is conducted in our proposed Nested Modality-aware Feature Aggregation (NMaFA) module, which enhances long-term dependencies within individual modalities via a tri-orientated spatial-attention transformer, and further complements key contextual information among modalities via a cross-modality attention transformer. Extensive experiments on BraTS2020 benchmark and a private meningiomas segmentation (MeniSeg) dataset show that the NestedFormer clearly outperforms the state-of-the-arts. The code is available at https://github.com/920232796/NestedFormer.

Zhaohu Xing, Lei Zhu, Lequan Yu et al.

Masked image modeling (MIM) with transformer backbones has recently been exploited as a powerful self-supervised pre-training technique. The existing MIM methods adopt the strategy to mask random patches of the image and reconstruct the missing pixels, which only considers semantic information at a lower level, and causes a long pre-training time.This paper presents HybridMIM, a novel hybrid self-supervised learning method based on masked image modeling for 3D medical image segmentation.Specifically, we design a two-level masking hierarchy to specify which and how patches in sub-volumes are masked, effectively providing the constraints of higher level semantic information. Then we learn the semantic information of medical images at three levels, including:1) partial region prediction to reconstruct key contents of the 3D image, which largely reduces the pre-training time burden (pixel-level); 2) patch-masking perception to learn the spatial relationship between the patches in each sub-volume (region-level).and 3) drop-out-based contrastive learning between samples within a mini-batch, which further improves the generalization ability of the framework (sample-level). The proposed framework is versatile to support both CNN and transformer as encoder backbones, and also enables to pre-train decoders for image segmentation. We conduct comprehensive experiments on four widely-used public medical image segmentation datasets, including BraTS2020, BTCV, MSD Liver, and MSD Spleen. The experimental results show the clear superiority of HybridMIM against competing supervised methods, masked pre-training approaches, and other self-supervised methods, in terms of quantitative metrics, timing performance and qualitative observations. The codes of HybridMIM are available at https://github.com/ge-xing/HybridMIM

Zhijin Ge, Fanhua Shang, Hongying Liu et al.

Deep neural networks are vulnerable to adversarial examples crafted by applying human-imperceptible perturbations on clean inputs. Although many attack methods can achieve high success rates in the white-box setting, they also exhibit weak transferability in the black-box setting. Recently, various methods have been proposed to improve adversarial transferability, in which the input transformation is one of the most effective methods. In this work, we notice that existing input transformation-based works mainly adopt the transformed data in the same domain for augmentation. Inspired by domain generalization, we aim to further improve the transferability using the data augmented from different domains. Specifically, a style transfer network can alter the distribution of low-level visual features in an image while preserving semantic content for humans. Hence, we propose a novel attack method named Style Transfer Method (STM) that utilizes a proposed arbitrary style transfer network to transform the images into different domains. To avoid inconsistent semantic information of stylized images for the classification network, we fine-tune the style transfer network and mix up the generated images added by random noise with the original images to maintain semantic consistency and boost input diversity. Extensive experimental results on the ImageNet-compatible dataset show that our proposed method can significantly improve the adversarial transferability on either normally trained models or adversarially trained models than state-of-the-art input transformation-based attacks. Code is available at: https://github.com/Zhijin-Ge/STM.

Tianling Liu, Wennan Liu, Lequan Yu et al.

Preoperative and noninvasive prediction of the meningioma grade is important in clinical practice, as it directly influences the clinical decision making. What's more, brain invasion in meningioma (i.e., the presence of tumor tissue within the adjacent brain tissue) is an independent criterion for the grading of meningioma and influences the treatment strategy. Although efforts have been reported to address these two tasks, most of them rely on hand-crafted features and there is no attempt to exploit the two prediction tasks simultaneously. In this paper, we propose a novel task-aware contrastive learning algorithm to jointly predict meningioma grade and brain invasion from multi-modal MRIs. Based on the basic multi-task learning framework, our key idea is to adopt contrastive learning strategy to disentangle the image features into task-specific features and task-common features, and explicitly leverage their inherent connections to improve feature representation for the two prediction tasks. In this retrospective study, an MRI dataset was collected, for which 800 patients (containing 148 high-grade, 62 invasion) were diagnosed with meningioma by pathological analysis. Experimental results show that the proposed algorithm outperforms alternative multi-task learning methods, achieving AUCs of 0:8870 and 0:9787 for the prediction of meningioma grade and brain invasion, respectively. The code is available at https://github.com/IsDling/predictTCL.

Zhihao Chen, Yang Zhou, Anh Tran et al.

Medical phrase grounding (MPG) aims to locate the most relevant region in a medical image, given a phrase query describing certain medical findings, which is an important task for medical image analysis and radiological diagnosis. However, existing visual grounding methods rely on general visual features for identifying objects in natural images and are not capable of capturing the subtle and specialized features of medical findings, leading to sub-optimal performance in MPG. In this paper, we propose MedRPG, an end-to-end approach for MPG. MedRPG is built on a lightweight vision-language transformer encoder and directly predicts the box coordinates of mentioned medical findings, which can be trained with limited medical data, making it a valuable tool in medical image analysis. To enable MedRPG to locate nuanced medical findings with better region-phrase correspondences, we further propose Tri-attention Context contrastive alignment (TaCo). TaCo seeks context alignment to pull both the features and attention outputs of relevant region-phrase pairs close together while pushing those of irrelevant regions far away. This ensures that the final box prediction depends more on its finding-specific regions and phrases. Experimental results on three MPG datasets demonstrate that our MedRPG outperforms state-of-the-art visual grounding approaches by a large margin. Additionally, the proposed TaCo strategy is effective in enhancing finding localization ability and reducing spurious region-phrase correlations.

Zhihao Chen, Liang Wan, Yefan Xiao et al.

Shadow removal in a single image has received increasing attention in recent years. However, removing shadows over dynamic scenes remains largely under-explored. In this paper, we propose the first data-driven video shadow removal model, termed PSTNet, by exploiting three essential characteristics of video shadows, i.e., physical property, spatio relation, and temporal coherence. Specifically, a dedicated physical branch was established to conduct local illumination estimation, which is more applicable for scenes with complex lighting and textures, and then enhance the physical features via a mask-guided attention strategy. Then, we develop a progressive aggregation module to enhance the spatio and temporal characteristics of features maps, and effectively integrate the three kinds of features. Furthermore, to tackle the lack of datasets of paired shadow videos, we synthesize a dataset (SVSRD-85) with aid of the popular game GTAV by controlling the switch of the shadow renderer. Experiments against 9 state-of-the-art models, including image shadow removers and image/video restoration methods, show that our method improves the best SOTA in terms of RMSE error for the shadow area by 14.7. In addition, we develop a lightweight model adaptation strategy to make our synthetic-driven model effective in real world scenes. The visual comparison on the public SBU-TimeLapse dataset verifies the generalization ability of our model in real scenes.

Qing Sun, Fan Lyu, Fanhua Shang et al.

Continual Learning (CL) sequentially learns new tasks like human beings, with the goal to achieve better Stability (S, remembering past tasks) and Plasticity (P, adapting to new tasks). Due to the fact that past training data is not available, it is valuable to explore the influence difference on S and P among training examples, which may improve the learning pattern towards better SP. Inspired by Influence Function (IF), we first study example influence via adding perturbation to example weight and computing the influence derivation. To avoid the storage and calculation burden of Hessian inverse in neural networks, we propose a simple yet effective MetaSP algorithm to simulate the two key steps in the computation of IF and obtain the S- and P-aware example influence. Moreover, we propose to fuse two kinds of example influence by solving a dual-objective optimization problem, and obtain a fused influence towards SP Pareto optimality. The fused influence can be used to control the update of model and optimize the storage of rehearsal. Empirical results show that our algorithm significantly outperforms state-of-the-art methods on both task- and class-incremental benchmark CL datasets.

Tianling Liu, Hongying Liu, Fanhua Shang et al.

In clinical practice, crossmodal information including medical images and tabular data is essential for disease diagnosis. There exists a significant modality gap between these data types, which obstructs advancements in crossmodal diagnostic accuracy. Most existing crossmodal learning (CML) methods primarily focus on exploring relationships among high-level encoder outputs, leading to the neglect of local information in images. Additionally, these methods often overlook the extraction of task-relevant information. In this paper, we propose a novel coarse-to-fine crossmodal learning (CFCML) framework to progressively reduce the modality gap between multimodal images and tabular data, by thoroughly exploring inter-modal relationships. At the coarse stage, we explore the relationships between multi-granularity features from various image encoder stages and tabular information, facilitating a preliminary reduction of the modality gap. At the fine stage, we generate unimodal and crossmodal prototypes that incorporate class-aware information, and establish hierarchical anchor-based relationship mining (HRM) strategy to further diminish the modality gap and extract discriminative crossmodal information. This strategy utilize modality samples, unimodal prototypes, and crossmodal prototypes as anchors to develop contrastive learning approaches, effectively enhancing inter-class disparity while reducing intra-class disparity from multiple perspectives. Experimental results indicate that our method outperforms the state-of-the-art (SOTA) methods, achieving improvements of 1.53% and 0.91% in AUC metrics on the MEN and Derm7pt datasets, respectively. The code is available at https://github.com/IsDling/CFCML.

Yupeng Zhang, Ruize Han, Wei Feng et al.

Object detection is an important task in computer vision, which aims to detect the objects of interest. through the given category list or query images. In this work, we propose a new problem of language-visual-complementary open-set object detection (LV-OSD), i.e., using the flexible text-based and/or image-based prompts to specify the desired object categories. This setting is more common and practical in real-world applications. For this purpose, we design a dual-branch detection framework, LVDor, which can simultaneously accept both text and image prompts. Specifically, we first build the Multi-modal Prompts (MPr) containing various text descriptions and image samples for each category. Subsequently, to bridge the semantic gap among the input image, text prompts, and image prompts, we design a Target-guided Prompt Dynamic Weighting (TPDW) module. Guided by the prior information of the target image, this module dynamically produces the text and image prompts that best align with the target semantics, achieving precise alignment and effectively reducing the discrepancy between the two modalities, thereby accommodating the LV-OSD setting. We also propose a simple Prompt Random Masking (PRM) mechanism during training to simulate the arbitrary combination of text and/or image prompts in testing. Extensive experimental results verify our problem formulation's reasonability and our method's effectiveness. Prompts and code will be released publicly.

Yupeng Zhang, Ruize Han, Yuzhong Feng et al.

Open-vocabulary object detection (OVD) has made significant progress, enabling detectors to generalize from seen to unseen categories. However, real-world category spaces continually evolve, and existing OVD models still struggle with newly emerging concepts, while repeated full retraining is prohibitively expensive. To this end, we introduce a new task setting, termed Continual OVD with Novel Concept Injection (COVD), where models sequentially learn incoming novel concept groups while preserving prior concepts and original open-vocabulary knowledge, along with a new benchmark, Novel-114. Our key observation is that pretrained visual encoders often already perceive and represent many novel concepts, and the main bottleneck lies in the lack of stable semantic alignment between visual representations and textual concepts. Based on this, we propose NoIn-Det, an efficient continual injection framework without additional parameters. NoIn-Det freezes the visual encoder, preserves the text representation space using only texts of common concepts and previously injected concepts, and injects novel concepts by updating only a small subset of text-branch parameters beneficial to novel concept learning. Extensive experiments show that NoIn-Det effectively learns novel concepts, preserves old knowledge, and consistently outperforms existing continual learning methods for VLMs without introducing additional parameters.Novel-114 and the code will be released.

Tianling Liu, Lequan Yu, Tong Han et al.

While multimodal data integrating diverse imaging and clinical tabular records is crucial for accurate medical diagnosis, the arbitrary absence of specific modalities is prevalent in clinical practice, severely degrading the performance of multimodal models. Existing methods either discard missing modalities, leading to information loss, or struggle to synthesize them without capturing complex inter-modal dependencies. To address these limitations, we propose a novel Context-driven Missing-Modality Learning (CMML) framework, which sequentially performs modality synthesis and semantic alignment to achieve robust diagnosis under arbitrary missing conditions. Specifically, we design a Cascade Residual Transformer-based Autoencoder (CRTA) that leverages learnable context tokens acting as dataset-level semantic prior to capture inter-modal dependencies and synthesize key missing representations. These representations are further enriched by modality-specific memory banks. To resolve the discrepancy between original available and synthesized representations, we transform the learned context tokens into instance-adaptive semantic references by infusing multimodal representations from the CRTA's outputs. This reference guides the alignment of heterogeneous modality representations into a unified space, where class-aware contrastive refinement is finally applied to explore discriminative diagnostic cues. Extensive evaluations on skin lesion (Derm7pt), ocular disease (ODIR), and meningioma (MEN) datasets demonstrate that CMML significantly outperforms state-of-the-art (SOTA) methods, yielding AVG AUC improvements of 1.26%, 0.97%, and 1.32%, respectively.

Yupeng Zhang, Ruize Han, Ningnan Guo et al.

In real-world scenarios, continual changes in weather, illumination, and imaging conditions cause significant domain shifts, leading detectors trained on a single source domain to degrade severely in unseen environments. Existing single-domain generalized object detection (SDGOD) methods mainly rely on data augmentation or domain-invariant representation learning, but pay limited attention to detector mechanisms, leaving clear limitations under complex domain shifts. Through analytical experiments, we find that performance degradation is dominated by increasing missed detections, which fundamentally arises from reduced cross-domain stability of the detector: object-background and inter-instance relations become less stable in the encoding stage, while semantic-spatial alignment of query representations also becomes harder to maintain in the decoding stage. To this end, we propose VFM$^{4}$SDG, a dual-prior learning framework for SDGOD, which introduces a frozen vision foundation model (VFM) as a transferable cross-domain stability prior into detector representation learning and query modeling. In the encoding stage, we propose Cross-domain Stable Relational Prior Distillation to enhance the robustness of object-background and inter-instance relational modeling. In the decoding stage, we propose Semantic-Contextual Prior-based Query Enhancement, which injects category-level semantic prototypes and global visual context into queries to improve their semantic recognition and spatial localization stability in unseen domains. Extensive experiments show that the proposed method consistently outperforms existing SOTA methods on standard SDGOD benchmarks and two mainstream DETR-based detectors, demonstrating its effectiveness, robustness, and generality.

Tianling Liu, Hongying Liu, Fanhua Shang et al.

Multimodal MRIs play a crucial role in clinical diagnosis and treatment. Feature disentanglement (FD)-based methods, aiming at learning superior feature representations for multimodal data analysis, have achieved significant success in multimodal learning (MML). Typically, existing FD-based methods separate multimodal data into modality-shared and modality-specific features, and employ concatenation or attention mechanisms to integrate these features. However, our preliminary experiments indicate that these methods could lead to a loss of shared information among subsets of modalities when the inputs contain more than two modalities, and such information is critical for prediction accuracy. Furthermore, these methods do not adequately interpret the relationships between the decoupled features at the fusion stage. To address these limitations, we propose a novel Complete Feature Disentanglement (CFD) strategy that recovers the lost information during feature decoupling. Specifically, the CFD strategy not only identifies modality-shared and modality-specific features, but also decouples shared features among subsets of multimodal inputs, termed as modality-partial-shared features. We further introduce a new Dynamic Mixture-of-Experts Fusion (DMF) module that dynamically integrates these decoupled features, by explicitly learning the local-global relationships among the features. The effectiveness of our approach is validated through classification tasks on three multimodal MRI datasets. Extensive experimental results demonstrate that our approach outperforms other state-of-the-art MML methods with obvious margins, showcasing its superior performance.

Tingliang Feng, Hao Shi, Xueyang Liu et al.

Many methods of semantic image segmentation have borrowed the success of open compound domain adaptation. They minimize the style gap between the images of source and target domains, more easily predicting the accurate pseudo annotations for target domain's images that train segmentation network. The existing methods globally adapt the scene style of the images, whereas the object styles of different categories or instances are adapted improperly. This paper proposes the Object Style Compensation, where we construct the Object-Level Discrepancy Memory with multiple sets of discrepancy features. The discrepancy features in a set capture the style changes of the same category's object instances adapted from target to source domains. We learn the discrepancy features from the images of source and target domains, storing the discrepancy features in memory. With this memory, we select appropriate discrepancy features for compensating the style information of the object instances of various categories, adapting the object styles to a unified style of source domain. Our method enables a more accurate computation of the pseudo annotations for target domain's images, thus yielding state-of-the-art results on different datasets.

Weiqi Li, Fan Lyu, Fanhua Shang et al.

Real-world data is extremely imbalanced and presents a long-tailed distribution, resulting in models that are biased towards classes with sufficient samples and perform poorly on rare classes. Recent methods propose to rebalance classes but they undertake the seesaw dilemma (what is increasing performance on tail classes may decrease that of head classes, and vice versa). In this paper, we argue that the seesaw dilemma is derived from gradient imbalance of different classes, in which gradients of inappropriate classes are set to important for updating, thus are prone to overcompensation or undercompensation on tail classes. To achieve ideal compensation, we formulate the long-tailed recognition as an multi-objective optimization problem, which fairly respects the contributions of head and tail classes simultaneously. For efficiency, we propose a Gradient-Balancing Grouping (GBG) strategy to gather the classes with similar gradient directions, thus approximately make every update under a Pareto descent direction. Our GBG method drives classes with similar gradient directions to form more representative gradient and provide ideal compensation to the tail classes. Moreover, We conduct extensive experiments on commonly used benchmarks in long-tailed learning and demonstrate the superiority of our method over existing SOTA methods.

Yupeng Zhang, Ruize Han, Zhiwei Chen et al.

Despite the remarkable progress in open-vocabulary object detection (OVD), a significant gap remains between the training and testing phases. During training, the RPN and RoI heads often misclassify unlabeled novel-category objects as background, causing some proposals to be prematurely filtered out by the RPN while others are further misclassified by the RoI head. During testing, these proposals again receive low scores and are removed in post-processing, leading to a significant drop in recall and ultimately weakening novel-category detection performance.To address these issues, we propose a novel training framework-NoOVD-which innovatively integrates a self-distillation mechanism grounded in the knowledge of frozen vision-language models (VLMs). Specifically, we design K-FPN, which leverages the pretrained knowledge of VLMs to guide the model in discovering novel-category objects and facilitates knowledge distillation-without requiring additional data-thus preventing forced alignment of novel objects with background.Additionally, we introduce R-RPN, which adjusts the confidence scores of proposals during inference to improve the recall of novel-category objects. Cross-dataset evaluations on OV-LVIS, OV-COCO, and Objects365 demonstrate that our approach consistently achieves superior performance across multiple metrics.

Lianyu Hu, Wei Feng, Liqing Gao et al.

In sign language, the conveyance of human body trajectories predominantly relies upon the coordinated movements of hands and facial expressions across successive frames. Despite the recent advancements of sign language understanding methods, they often solely focus on individual frames, inevitably overlooking the inter-frame correlations that are essential for effectively modeling human body trajectories. To address this limitation, this paper introduces a spatial-temporal correlation network, denoted as CorrNet+, which explicitly identifies body trajectories across multiple frames. In specific, CorrNet+ employs a correlation module and an identification module to build human body trajectories. Afterwards, a temporal attention module is followed to adaptively evaluate the contributions of different frames. The resultant features offer a holistic perspective on human body movements, facilitating a deeper understanding of sign language. As a unified model, CorrNet+ achieves new state-of-the-art performance on two extensive sign language understanding tasks, including continuous sign language recognition (CSLR) and sign language translation (SLT). Especially, CorrNet+ surpasses previous methods equipped with resource-intensive pose-estimation networks or pre-extracted heatmaps for hand and facial feature extraction. Compared with CorrNet, CorrNet+ achieves a significant performance boost across all benchmarks while halving the computational overhead. A comprehensive comparison with previous spatial-temporal reasoning methods verifies the superiority of CorrNet+. Code is available at https://github.com/hulianyuyy/CorrNet_Plus.

Lianyu Hu, Fanhua Shang, Liang Wan et al.

In this paper, we introduce iLLaVA, a simple method that can be seamlessly deployed upon current Large Vision-Language Models (LVLMs) to greatly increase the throughput with nearly lossless model performance, without a further requirement to train. iLLaVA achieves this by finding and gradually merging the redundant tokens with an accurate and fast algorithm, which can merge hundreds of tokens within only one step. While some previous methods have explored directly pruning or merging tokens in the inference stage to accelerate models, our method excels in both performance and throughput by two key designs. First, while most previous methods only try to save the computations of Large Language Models (LLMs), our method accelerates the forward pass of both image encoders and LLMs in LVLMs, which both occupy a significant part of time during inference. Second, our method recycles the beneficial information from the pruned tokens into existing tokens, which avoids directly dropping context tokens like previous methods to cause performance loss. iLLaVA can nearly 2$\times$ the throughput, and reduce the memory costs by half with only a 0.2\% - 0.5\% performance drop across models of different scales including 7B, 13B and 34B. On tasks across different domains including single-image, multi-images and videos, iLLaVA demonstrates strong generalizability with consistently promising efficiency. We finally offer abundant visualizations to show the merging processes of iLLaVA in each step, which show insights into the distribution of computing resources in LVLMs. Code is available at https://github.com/hulianyuyy/iLLaVA.

Xuzhi Wang, Wei Feng, Lingdong Kong et al.

LiDAR semantic segmentation plays a vital role in autonomous driving. Existing voxel-based methods for LiDAR semantic segmentation apply uniform partition to the 3D LiDAR point cloud to form a structured representation based on cartesian/cylindrical coordinates. Although these methods show impressive performance, the drawback of existing voxel-based methods remains in two aspects: (1) it requires a large enough input voxel resolution, which brings a large amount of computation cost and memory consumption. (2) it does not well handle the unbalanced point distribution of LiDAR point cloud. In this paper, we propose a non-uniform cylindrical partition network named NUC-Net to tackle the above challenges. Specifically, we propose the Arithmetic Progression of Interval (API) method to non-uniformly partition the radial axis and generate the voxel representation which is representative and efficient. Moreover, we propose a non-uniform multi-scale aggregation method to improve contextual information. Our method achieves state-of-the-art performance on SemanticKITTI and nuScenes datasets with much faster speed and much less training time. And our method can be a general component for LiDAR semantic segmentation, which significantly improves both the accuracy and efficiency of the uniform counterpart by $4 \times$ training faster and $2 \times$ GPU memory reduction and $3 \times$ inference speedup. We further provide theoretical analysis towards understanding why NUC is effective and how point distribution affects performance. Code is available at \href{https://github.com/alanWXZ/NUC-Net}{https://github.com/alanWXZ/NUC-Net}.

Liang Liao, Liang Wan, Mingsheng Liu et al.

When some application scenarios need to use semantic segmentation technology, like automatic driving, the primary concern comes to real-time performance rather than extremely high segmentation accuracy. To achieve a good trade-off between speed and accuracy, two-branch architecture has been proposed in recent years. It treats spatial information and semantics information separately which allows the model to be composed of two networks both not heavy. However, the process of fusing features with two different scales becomes a performance bottleneck for many nowaday two-branch models. In this research, we design a new fusion mechanism for two-branch architecture which is guided by attention computation. To be precise, we use the Dual-Guided Attention (DGA) module we proposed to replace some multi-scale transformations with the calculation of attention which means we only use several attention layers of near linear complexity to achieve performance comparable to frequently-used multi-layer fusion. To ensure that our module can be effective, we use Residual U-blocks (RSU) to build one of the two branches in our networks which aims to obtain better multi-scale features. Extensive experiments on Cityscapes and CamVid dataset show the effectiveness of our method.

Xinmeng Hou, Ziting Chang, Zhouquan Lu et al.

Large language models (LLMs) fail on over one-third of multi-hop questions with counterfactual premises and remain vulnerable to adversarial prompts that trigger biased or factually incorrect responses, which exposes a fundamental deficit in self-regulated reasoning. We propose \textbf{MetaCrit}, a multi-agent framework grounded in Nelson and Narens' metacognitive regulation theory. MetaCrit decomposes reasoning regulation into four agents: object-level generation, a \emph{monitoring} agent that assesses response validity, a \emph{control} agent that critiques logical soundness, and a meta-level synthesizer that integrates all signals into a final response. Evaluation across eight benchmarks, four model backbones, and a college-level analytical writing study shows that MetaCrit significantly improves content truthfulness and logical soundness while eliminating toxic outputs. Its modular design allows individual agents to be integrated into existing frameworks as drop-in components without architectural modifications.

Xijun Lu, Hongying Liu, Fanhua Shang et al.

Medical image anomaly detection faces unique challenges due to subtle, heterogeneous anomalies embedded in complex anatomical structures. Through systematic Grad-CAM analysis, we reveal that discriminative activation maps fail on medical data, unlike their success on industrial datasets, motivating the need for manifold-level modeling. We propose PDD (Manifold-Prior Diverse Distillation), a framework that unifies dual-teacher priors into a shared high-dimensional manifold and distills this knowledge into dual students with complementary behaviors. Specifically, frozen VMamba-Tiny and wide-ResNet50 encoders provide global contextual and local structural priors, respectively. Their features are unified through a Manifold Matching and Unification (MMU) module, while an Inter-Level Feature Adaption (InA) module enriches intermediate representations. The unified manifold is distilled into two students: one performs layer-wise distillation via InA for local consistency, while the other receives skip-projected representations through a Manifold Prior Affine (MPA) module to capture cross-layer dependencies. A diversity loss prevents representation collapse while maintaining detection sensitivity. Extensive experiments on multiple medical datasets demonstrate that PDD significantly outperforms existing state-of-the-art methods, achieving improvements of up to 11.8%, 5.1%, and 8.5% in AUROC on HeadCT, BrainMRI, and ZhangLab datasets, respectively, and 3.4% in F1 max on the Uni-Medical dataset, establishing new state-of-the-art performance in medical image anomaly detection. The implementation will be released at https://github.com/OxygenLu/PDD

Junting Zhao, Yang Zhou, Zhihao Chen et al.

Automated radiology reporting holds immense clinical potential in alleviating the burdensome workload of radiologists and mitigating diagnostic bias. Recently, retrieval-based report generation methods have garnered increasing attention due to their inherent advantages in terms of the quality and consistency of generated reports. However, due to the long-tail distribution of the training data, these models tend to learn frequently occurring sentences and topics, overlooking the rare topics. Regrettably, in many cases, the descriptions of rare topics often indicate critical findings that should be mentioned in the report. To address this problem, we introduce a Topicwise Separable Sentence Retrieval (Teaser) for medical report generation. To ensure comprehensive learning of both common and rare topics, we categorize queries into common and rare types to learn differentiated topics, and then propose Topic Contrastive Loss to effectively align topics and queries in the latent space. Moreover, we integrate an Abstractor module following the extraction of visual features, which aids the topic decoder in gaining a deeper understanding of the visual observational intent. Experiments on the MIMIC-CXR and IU X-ray datasets demonstrate that Teaser surpasses state-of-the-art models, while also validating its capability to effectively represent rare topics and establish more dependable correspondences between queries and topics.

Fan Lyu, Wei Feng, Yuepan Li et al.

The goal of Continual Learning (CL) is to continuously learn from new data streams and accomplish the corresponding tasks. Previously studied CL assumes that data are given in sequence nose-to-tail for different tasks, thus indeed belonging to Serial Continual Learning (SCL). This paper studies the novel paradigm of Parallel Continual Learning (PCL) in dynamic multi-task scenarios, where a diverse set of tasks is encountered at different time points. PCL presents challenges due to the training of an unspecified number of tasks with varying learning progress, leading to the difficulty of guaranteeing effective model updates for all encountered tasks. In our previous conference work, we focused on measuring and reducing the discrepancy among gradients in a multi-objective optimization problem, which, however, may still contain negative transfers in every model update. To address this issue, in the dynamic multi-objective optimization problem, we introduce task-specific elastic factors to adjust the descent direction towards the Pareto front. The proposed method, called Elastic Multi-Gradient Descent (EMGD), ensures that each update follows an appropriate Pareto descent direction, minimizing any negative impact on previously learned tasks. To balance the training between old and new tasks, we also propose a memory editing mechanism guided by the gradient computed using EMGD. This editing process updates the stored data points, reducing interference in the Pareto descent direction from previous tasks. Experiments on public datasets validate the effectiveness of our EMGD in the PCL setting.

Hongmei Yin, Tingliang Feng, Fan Lyu et al.

In this work, we focus on continual semantic segmentation (CSS), where segmentation networks are required to continuously learn new classes without erasing knowledge of previously learned ones. Although storing images of old classes and directly incorporating them into the training of new models has proven effective in mitigating catastrophic forgetting in classification tasks, this strategy presents notable limitations in CSS. Specifically, the stored and new images with partial category annotations leads to confusion between unannotated categories and the background, complicating model fitting. To tackle this issue, this paper proposes a novel Enhanced Instance Replay (EIR) method, which not only preserves knowledge of old classes while simultaneously eliminating background confusion by instance storage of old classes, but also mitigates background shifts in the new images by integrating stored instances with new images. By effectively resolving background shifts in both stored and new images, EIR alleviates catastrophic forgetting in the CSS task, thereby enhancing the model's capacity for CSS. Experimental results validate the efficacy of our approach, which significantly outperforms state-of-the-art CSS methods.

Lianyu Hu, Fanhua Shang, Wei Feng et al.

In this paper, we introduce LightVLM, a simple but effective method that can be seamlessly deployed upon existing Vision-Language Models (VLMs) to greatly accelerate the inference process in a training-free manner. We divide the inference procedure of VLMs into two stages, i.e., encoding and decoding, and propose to simultaneously accelerate VLMs in both stages to largely improve model efficiency. During encoding, we propose pyramid token merging to reduce tokens of different LLM layers in a hierarchical manner by finally only keeping a few dominant tokens to achieve high efficiency. During decoding, aimed at reducing the high latency of outputting long sequences, we propose KV Cache compression to remove unnecessary caches to increase the network throughput. Experimental results show that LightVLM successfully retains 100% performance when only preserving 35% image tokens, and maintains around 98% performance when keeping only 3% image tokens. LightVLM could 2.02$\times$ the network throughput and reduce the prefilling time by 3.65$\times$. LightVLM also makes large VLMs faster again by enabling a heavy model (e.g., InternVL2.5 26B) to infer faster than significantly smaller models (e.g., InternVL2.5 8B), hopefully facilitating the real-world deployment. When generating long text sequences (e.g., 4096 tokens), LightVLM could reduce the inference time by 3.21$\times$, largely outperforming existing methods.

Yupeng Zhang, Ruize Han, Fangnan Zhou et al.

In this work, we handle a new problem of Open-Domain Open-Vocabulary (ODOV) object detection, which considers the detection model's adaptability to the real world including both domain and category shifts. For this problem, we first construct a new benchmark OD-LVIS, which includes 46,949 images, covers 18 complex real-world domains and 1,203 categories, and provides a comprehensive dataset for evaluating real-world object detection. Besides, we develop a novel baseline method for ODOV detection.The proposed method first leverages large language models to generate the domain-agnostic text prompts for category embedding. It further learns the domain embedding from the given image, which, during testing, can be integrated into the category embedding to form the customized domain-specific category embedding for each test image. We provide sufficient benchmark evaluations for the proposed ODOV detection task and report the results, which verify the rationale of ODOV detection, the usefulness of our benchmark, and the superiority of the proposed method.

Jiaxi Li, Di Lin, Hao Chen et al.

Deep neural networks (DNNs) often struggle with out-of-distribution data, limiting their reliability in diverse realworld applications. To address this issue, domain generalization methods have been developed to learn domain-invariant features from single or multiple training domains, enabling generalization to unseen testing domains. However, existing approaches usually overlook the impact of style frequency within the training set. This oversight predisposes models to capture spurious visual correlations caused by style confounding factors, rather than learning truly causal representations, thereby undermining inference reliability. In this work, we introduce Style Deconfounding Causal Learning (SDCL), a novel causal inference-based framework designed to explicitly address style as a confounding factor. Our approaches begins with constructing a structural causal model (SCM) tailored to the domain generalization problem and applies a backdoor adjustment strategy to account for style influence. Building on this foundation, we design a style-guided expert module (SGEM) to adaptively clusters style distributions during training, capturing the global confounding style. Additionally, a back-door causal learning module (BDCL) performs causal interventions during feature extraction, ensuring fair integration of global confounding styles into sample predictions, effectively reducing style bias. The SDCL framework is highly versatile and can be seamlessly integrated with state-of-the-art data augmentation techniques. Extensive experiments across diverse natural and medical image recognition tasks validate its efficacy, demonstrating superior performance in both multi-domain and the more challenging single-domain generalization scenarios.

Ye Liu, Lei Zhu, Shunda Pei et al.

Single image dehazing is a challenging task, for which the domain shift between synthetic training data and real-world testing images usually leads to degradation of existing methods. To address this issue, we propose a novel image dehazing framework collaborating with unlabeled real data. First, we develop a disentangled image dehazing network (DID-Net), which disentangles the feature representations into three component maps, i.e. the latent haze-free image, the transmission map, and the global atmospheric light estimate, respecting the physical model of a haze process. Our DID-Net predicts the three component maps by progressively integrating features across scales, and refines each map by passing an independent refinement network. Then a disentangled-consistency mean-teacher network (DMT-Net) is employed to collaborate unlabeled real data for boosting single image dehazing. Specifically, we encourage the coarse predictions and refinements of each disentangled component to be consistent between the student and teacher networks by using a consistency loss on unlabeled real data. We make comparison with 13 state-of-the-art dehazing methods on a new collected dataset (Haze4K) and two widely-used dehazing datasets (i.e., SOTS and HazeRD), as well as on real-world hazy images. Experimental results demonstrate that our method has obvious quantitative and qualitative improvements over the existing methods.

Zhihao Chen, Liang Wan, Lei Zhu et al.

Shadow detection in a single image has received significant research interest in recent years. However, much fewer works have been explored in shadow detection over dynamic scenes. The bottleneck is the lack of a well-established dataset with high-quality annotations for video shadow detection. In this work, we collect a new video shadow detection dataset, which contains 120 videos with 11, 685 frames, covering 60 object categories, varying lengths, and different motion/lighting conditions. All the frames are annotated with a high-quality pixel-level shadow mask. To the best of our knowledge, this is the first learning-oriented dataset for video shadow detection. Furthermore, we develop a new baseline model, named triple-cooperative video shadow detection network (TVSD-Net). It utilizes triple parallel networks in a cooperative manner to learn discriminative representations at intra-video and inter-video levels. Within the network, a dual gated co-attention module is proposed to constrain features from neighboring frames in the same video, while an auxiliary similarity loss is introduced to mine semantic information between different videos. Finally, we conduct a comprehensive study on ViSha, evaluating 12 state-of-the-art models (including single image shadow detectors, video object segmentation, and saliency detection methods). Experiments demonstrate that our model outperforms SOTA competitors.

Qian Zhang, Wei Feng, Liang Wan et al.

This paper studies a new problem, namely active lighting recurrence (ALR) that physically relocalizes a light source to reproduce the lighting condition from single reference image for a same scene, which may suffer from fine-grained changes during twice observations. ALR is of great importance for fine-grained visual inspection and change detection, because some phenomena or minute changes can only be clearly observed under particular lighting conditions. Therefore, effective ALR should be able to online navigate a light source toward the target pose, which is challenging due to the complexity and diversity of real-world lighting and imaging processes. To this end, we propose to use the simple parallel lighting as an analogy model and based on Lambertian law to compose an instant navigation ball for this purpose. We theoretically prove the feasibility, i.e., equivalence and convergence, of this ALR approach for realistic near point light source and small near surface light source. Besides, we also theoretically prove the invariance of our ALR approach to the ambiguity of normal and lighting decomposition. The effectiveness and superiority of the proposed approach have been verified by both extensive quantitative experiments and challenging real-world tasks on fine-grained change detection of cultural heritages. We also validate the generality of our approach to non-Lambertian scenes.

Qing Guo, Wei Feng, Zhihao Chen et al.

Intuitively, motion blur may hurt the performance of visual object tracking. However, we lack quantitative evaluation of tracker robustness to different levels of motion blur. Meanwhile, while image deblurring methods can produce visually clearer videos for pleasing human eyes, it is unknown whether visual object tracking can benefit from image deblurring or not. In this paper, we address these two problems by constructing a Blurred Video Tracking benchmark, which contains a variety of videos with different levels of motion blurs, as well as ground truth tracking results for evaluating trackers. We extensively evaluate 23 trackers on this benchmark and observe several new interesting results. Specifically, we find that light blur may improve the performance of many trackers, but heavy blur always hurts the tracking performance. We also find that image deblurring may help to improve tracking performance on heavily blurred videos but hurt the performance on lightly blurred videos. According to these observations, we propose a new GAN based scheme to improve the tracker robustness to motion blurs. In this scheme, a finetuned discriminator is used as an adaptive assessor to selectively deblur frames during the tracking process. We use this scheme to successfully improve the accuracy and robustness of 6 trackers.

Ruize Han, Yujun Zhang, Wei Feng et al.

Video surveillance can be significantly enhanced by using both top-view data, e.g., those from drone-mounted cameras in the air, and horizontal-view data, e.g., those from wearable cameras on the ground. Collaborative analysis of different-view data can facilitate various kinds of applications, such as human tracking, person identification, and human activity recognition. However, for such collaborative analysis, the first step is to associate people, referred to as subjects in this paper, across these two views. This is a very challenging problem due to large human-appearance difference between top and horizontal views. In this paper, we present a new approach to address this problem by exploring and matching the subjects' spatial distributions between the two views. More specifically, on the top-view image, we model and match subjects' relative positions to the horizontal-view camera in both views and define a matching cost to decide the actual location of horizontal-view camera and its view angle in the top-view image. We collect a new dataset consisting of top-view and horizontal-view image pairs for performance evaluation and the experimental results show the effectiveness of the proposed method.

Mengya Gao, Yujun Shen, Quanquan Li et al.

Knowledge Distillation (KD) aims at improving the performance of a low-capacity student model by inheriting knowledge from a high-capacity teacher model. Previous KD methods typically train a student by minimizing a task-related loss and the KD loss simultaneously, using a pre-defined loss weight to balance these two terms. In this work, we propose to first transfer the backbone knowledge from a teacher to the student, and then only learn the task-head of the student network. Such a decomposition of the training process circumvents the need of choosing an appropriate loss weight, which is often difficult in practice, and thus makes it easier to apply to different datasets and tasks. Importantly, the decomposition permits the core of our method, Stage-by-Stage Knowledge Distillation (SSKD), which facilitates progressive feature mimicking from teacher to student. Extensive experiments on CIFAR-100 and ImageNet suggest that SSKD significantly narrows down the performance gap between student and teacher, outperforming state-of-the-art approaches. We also demonstrate the generalization ability of SSKD on other challenging benchmarks, including face recognition on IJB-A dataset as well as object detection on COCO dataset.