Rehearsal-based video incremental learning often employs knowledge distillation to mitigate catastrophic forgetting of previously learned data. However, this method faces two major challenges for video task: substantial computing resources from loading teacher model and limited replay capability from performance-limited teacher model. To address these problems, we first propose a knowledge distillation-free framework for rehearsal-based video incremental learning called \textit{Teacher Agent}. Instead of loading parameter-heavy teacher networks, we introduce an agent generator that is either parameter-free or uses only a few parameters to obtain accurate and reliable soft labels. This method not only greatly reduces the computing requirement but also circumvents the problem of knowledge misleading caused by inaccurate predictions of the teacher model. Moreover, we put forward a self-correction loss which provides an effective regularization signal for the review of old knowledge, which in turn alleviates the problem of catastrophic forgetting. Further, to ensure that the samples in the memory buffer are memory-efficient and representative, we introduce a unified sampler for rehearsal-based video incremental learning to mine fixed-length key video frames. Interestingly, based on the proposed strategies, the network exhibits a high level of robustness against spatial resolution reduction when compared to the baseline. Extensive experiments demonstrate the advantages of our method, yielding significant performance improvements while utilizing only half the spatial resolution of video clips as network inputs in the incremental phases.

While Large Vision-Language Models (LVLMs) achieves remarkable success, hallucinations remain a significant barrier to their reliable deployment. Recent studies primarily attribute these defects to cross-modal attention imbalances, with most solutions focusing on re-weighting visual tokens or suppressing language priors. Such approaches often overlook the spectral characteristics of the visual information flow and frequently rely on Contrastive Decoding (CD), which doubles the inference time. Instead of following conventional approaches, we identify two distinct hallucination patterns—Perceptual-Semantic Dissociation and Localized Fixation—and accordingly develop FLASH (Frequency-Localized Attention SHaping), a training-free and CD-free framework. FLASH utilizes a Spectral Vortex Score to detect visual heads within multi-head attention layers, applying adaptive spectral modulation to rectify the visual information flow during the decoding phase. Empirical results demonstrate that FLASH offers a superior balance between performance and efficiency compared to SOTA methods.

Due to the enormous technical challenges and wide range of applications, scene text recognition (STR) has been an active research topic in computer vision for years. To tackle this tough problem, numerous innovative methods have been successively proposed, and incorporating linguistic knowledge into STR models has recently become a prominent trend. In this work, we first draw inspiration from the recent progress in Vision Transformer (ViT) to construct a conceptually simple yet functionally powerful vision STR model, which is built upon ViT and a tailored Adaptive Addressing and Aggregation (A$^3$) module. It already outperforms most previous state-of-the-art models for scene text recognition, including both pure vision models and language-augmented methods. To integrate linguistic knowledge, we further propose a Multi-Granularity Prediction strategy to inject information from the language modality into the model in an implicit way, \ie, subword representations (BPE and WordPiece) widely used in NLP are introduced into the output space, in addition to the conventional character level representation, while no independent language model (LM) is adopted. To produce the final recognition results, two strategies for effectively fusing the multi-granularity predictions are devised. The resultant algorithm (termed MGP-STR) is able to push the performance envelope of STR to an even higher level. Specifically, MGP-STR achieves an average recognition accuracy of $94\%$ on standard benchmarks for scene text recognition. Moreover, it also achieves state-of-the-art results on widely-used handwritten benchmarks as well as more challenging scene text datasets, demonstrating the generality of the proposed MGP-STR algorithm. The source code and models will be available at: \url{https://github.com/AlibabaResearch/AdvancedLiterateMachinery/tree/main/OCR/MGP-STR}.

Low-Rank Adaptation (LoRA) enables foundation models to efficiently adapt to downstream tasks with minimal trainable parameters, yet it lacks the capability of continual learning (CL) from non-stationary data streams. While existing CL methods attempt to address this by storing and retrieving task-specific parameter modules, they suffer from non-trivial challenges of task identification and inconsistency in input-output feature mappings, resulting in suboptimal feature representations. Instead of keeping multiple task-specific parameter modules, we propose a novel continual LoRA mechanism that parameterizes all task-specific adaptations with a single adaptation module, naturally obviating the task identification and mapping inconsistency issues. To this end, we reformulate LoRA-based CL as a consistent feature mapping problem that mimics the behavior of the joint-training upper bound, wherein a unified adaptation parameter matrix is learned to simultaneously capture the input-output relationships established by all task-specific LoRAs. By solving this formulation, our approach facilitates approximate unified optimization over sequential tasks via progressively updating cumulative feature statistics, achieving efficient knowledge retention and renewal. Notably, the proposed arithmetic training mechanism can be applied both to the backbone of the foundation model for parameter-efficient fine-tuning and the classifier optimization, thereby enabling continual learning within an integrated architecture that closely approximates the joint learning paradigm. Extensive experiments across conventional and long-sequence class-incremental benchmarks demonstrate that our method achieves new state-of-the-art performance.

Large-kernel Convolutional Neural Networks (CNNs) deliver remarkable performance in vision tasks by significantly expanding receptive fields, yet their quadratic parameter growth critically impedes storage-efficient edge deployment. While existing efficient architectures adopt parameter-efficient depthwise separable convolution backbones that leverage techniques like low-rank approximation and weight sharing to compress depthwise convolutions, we identify a critical oversight: pointwise convolutions dominate parameter volume (>87

With the widespread deployment of visible-infrared multimodal perception systems in safety-critical domains such as autonomous driving, evaluating their cross-modal adversarial robustness has become increasingly vital. However, existing approaches exhibit significant limitations in approximating the intrinsic laws of imaging. Most studies either focus on a single modality, failing to bypass cross-modal verification, or simplify infrared modeling into heuristic pixel-intensity distributions, neglecting the impact of ambient temperature fluctuations on adversarial stability. To bridge this gap, this paper proposes MEC-Patch, a cross-modal adversarial attack framework driven by intrinsic physical laws. By leveraging the Stefan-Boltzmann Law, we establish a physics-grounded cross-spectral mapping that explicitly links material emissivity to thermal radiation. Building on this formulation, we reveal that, under a fixed emissivity distribution, ambient temperature variations induce consistent global scaling while preserving relative emissivity-induced contrast. We exploit this property to construct temperature-robust adversarial perturbations whose discriminative patterns remain stable in the infrared modality, thereby fundamentally mitigating environmental sensitivity. Furthermore, we employ the physics-constrained NSGA-II algorithm to synergistically optimize the material-distribution-based patch parameters effective across both modalities, while enhancing generalization through a Dynamic Adversarial Resampling (DAR) strategy. Experimental results demonstrate that MEC-Patch effectively deceives state-of-the-art multimodal detectors and exhibits high robustness within high-fidelity, physically-consistent, and multi-scene simulation environments. This research provides a physical-law-driven perspective for the security assessment of multimodal perception systems.

Person re-identification aims to match the same individual across non-overlapping camera views. In real-world scenarios, ReID models are expected to continuously learn from newly arriving data while mitigating the catastrophic forgetting problem, namely the loss of previously acquired knowledge, which defines the lifelong ReID task. Existing lifelong ReID methods face limitations in balancing performance between old and new domains. To address this, we propose a Dynamic Knowledge Adaptation and Retention framework, which achieves a dynamic learning mechanism. During the forward propagation process, input images are first fed into a shared backbone network to extract basic features. Subsequently, the Dynamic Adaptation module performs instance normalization on the features to mitigate inter-domain discrepancies and dynamically generates convolutional kernel parameters based on domain-specific information within the input images, enabling adaptive feature optimization tailored to specific domains. During the training phase, we adopt an Adaptability Retention strategy to restrict model updates. When encountering new incoming data, we utilize the frozen model trained on previous domains as a knowledge anchor to constrain parameter updates during fine-tuning on the new data. Specifically, a constraint loss is applied to maintain parameter consistency between the knowledge anchor and the trainable model, enabling an adaptive balance between knowledge retention and adaptation. This design ensures the model’s adaptability to new domains while effectively mitigating catastrophic forgetting. Experiments on four mainstream person re-identification datasets demonstrate that our method achieves outstanding performance in the lifelong learning setting, with an average Rank-1 accuracy of 65.7% and an average mAP of 55.4%, significantly surpassing existing methods.

Cross-view geo-localization between UAV and satellite imagery remains a fundamental yet highly challenging task, especially under large off-nadir views where drastic perspective distortions, occlusions, and appearance gaps occur. Existing benchmarks and methods primarily focus on near-nadir scenarios and often overlook the importance of structural scene understanding and intra-domain relational constraints, limiting their performance in real-world deployments. In this work, we introduce OffNadirLoc, a new benchmark for large off-nadir UAV-to-satellite geo-localization. To tackle the unique challenges posed by off-nadir perspectives, we further propose ONLoc, a framework that incorporates a structure-aware contextual weighting mechanism to dynamically emphasize reliable local features while suppressing ambiguous or repetitive regions. Additionally, we design a view-coherent learning strategy, which treats one satellite image and the corresponding UAV images from multiple views as a cohesive semantic group. This set-level supervision enables the model to learn viewpoint-invariant and discriminative features, making it more effective at capturing multi-view consistency than conventional pairwise contrastive learning. Extensive experiments on the OffNadirLoc benchmark and four near-nadir datasets demonstrate that our method consistently outperforms state-of-the-art approaches while exhibiting strong zero-shot generalization to unseen datasets without additional training. The code will be released at https://montalario.github.io/offnadirloc/.

Incremental Object Detection (IOD) aims to continuously learn new object classes without forgetting previously learned ones. A persistent challenge is catastrophic forgetting, primarily attributed to background shift in conventional detectors. While pseudo-labeling mitigates this in dense detectors, we identify a novel, distinct source of forgetting specific to DETR-like architectures: background foregrounding. This arises from the exhaustiveness constraint of the Hungarian matcher, which forcibly assigns every ground truth target to one prediction, even when predictions primarily cover background regions (i.e., low IoU). This erroneous supervision compels the model to misclassify background features as specific foreground classes, disrupting learned representations and accelerating forgetting. To address this, we propose a Quality-guided Min-Cost Max-Flow (Q-MCMF) matcher. To avoid forced assignments, Q-MCMF builds a flow graph and prunes implausible matches based on geometric quality. It then optimizes for the final matching that minimizes cost and maximizes valid assignments. This strategy eliminates harmful supervision from background foregrounding while maximizing foreground learning signals. Extensive experiments on the COCO dataset under various incremental settings demonstrate that our method consistently outperforms existing state-of-the-art approaches.

Conventional video anomaly detection approaches struggle with increasingly sophisticated fine-grained analysis requirements in real-world applications, establishing Temporal Video Anomaly Grounding (TVAG) as one of the pivotal research frontiers in advanced anomaly video comprehension systems. Targeting the scarcity of precise temporal annotations, this work develops a single-frame supervision based framework, Glance-guided Cross-modal Proposal Generation (GCPG), which offers competitive grounding performance, surpassing some fully supervised methods under specific metrics, while substantially reducing annotation costs. The framework consists of a Cross-Modal Collaborative Pseudo-Glance Localization module (PGL) and a Glance-Guided Gaussian Proposal Optimization module (GPO). PGL employs a semantic-aware dual-branch mechanism that jointly performs cross-modal feature fusion classification and textual semantic verification to generate reliable pseudo-frame supervision, forming the foundation for cross-modal alignment learning. GPO enhances proposal quality by reconstructing Gaussian mask composition weights based on glance-keyword alignment and distributional consistency. Comprehensive experiments and ablation analyses on two challenging TVAG benchmarks validate the efficacy of our single-frame supervised approach.

Accurate High-Definition (HD) map construction is critical for autonomous driving, yet existing methods face a fundamental trade-off: vectorization-based approaches preserve topology but struggle with geometric fidelity, while rasterization-based approaches enable precise geometric supervision but produce unstructured outputs. To bridge this gap, we propose GSMap, a novel framework that unifies both paradigms via a learnable 2D Gaussian representation. Each map element is modeled as an ordered sequence of 2D Gaussians, whose centers correspond to the vertices of the vectorized polyline/polygon. This formulation enables simultaneous optimization through: (1) Differentiable rasterization that enforces pixel-level geometric constraints, and (2) Topology-aware vectorization that maintains structural regularity. Experiments on both nuScenes and Argoverse2 demonstrate that our Gaussian-based representation effectively unifies geometric and topological learning, achieving significant performance improvements and demonstrating strong compatibility with existing HD mapping architectures. Code will be available at https://github.com/peakpang/GSMap

Cloth-changing person re-identification (CC-ReID) aims to match individuals wearing varying clothes across camera views. Existing CC-ReID methods typically focus on extracting clothing-invariant features such as body shape, pose, gait, etc. However, these features are diverse and often entangled with clothing-related visual clues, posing significant challenges for comprehensively and effectively separating and leveraging them for re-identification. To address these challenges, we propose a text-guided clothing generalizable (Tex-CG) model, which employs multi-modal large language models (MLLMs) to comprehensively and explicitly decouple clothing-invariant features from pedestrian images in the textual domain. By shifting feature disentanglement to the textual domain, the interference caused by visual entanglement between clothing and clothing-invariant clues can be significantly reduced. Additionally, to ensure compatibility between offline-decoupled features from MLLMs and our online-trained Tex-CG model, we utilize CLIP-based image-text matching to train implicit clothing-invariant prompts that embed discriminative pedestrian information. A dynamic fusion module is subsequently introduced to leverage these implicit prompts for selectively integrating valuable and compatible components from the MLLM’s explicitly decoupled features, constructing robust guidance to direct our model to effectively capture clothing-invariant visual clues for re-identification. Extensive experiments demonstrate the effectiveness of our method, and the Tex-CG model achieves state-of-the-art performance on five mainstream CC-ReID benchmarks. Our code is available at https://github.com/JiaoBL1234/Tex-CG.

Multi-view Multi-object Association and Tracking (MvMoAT) associates objects across camera views and tracks them over time, supporting identity persistence and forensic trajectory reconstruction in multi-platform cooperative perception. Unlike conventional multiple object tracking, MvMoAT faces frequent viewpoint shifts that distort appearance and undermine cross-view association and temporal tracking. We propose FUSION, a viewpoint-robust Feature Unification framework for multi-view aSsociation and IdentificatiON. Its Multi-cue Adaptive Combination (MAC) module adaptively integrates viewpoint-invariant cues with appearance features to improve cross-view association, while Online Multi-view Feature Synchronization (OMFS) aggregates pedestrian features across historical and cross-view frames for temporally consistent tracking. We also introduce RealMvMoAT, a large-scale benchmark featuring substantial inter- and intra-camera viewpoint variation. It contains 504.9K frames from 7 cameras (5 UAV and 2 ground views) across 10 scenes, with over 7.3M identity-labeled bounding boxes. All cameras exhibit random and substantial motion. To the best of our knowledge, RealMvMoAT is the largest MvMoAT dataset to date. Its scale, viewpoint diversity, complex platform motion, and realistic trajectories provide a comprehensive resource for future research. Experiments on RealMvMoAT and six public benchmarks show that FUSION achieves state-of-the-art performance.

Out-of-Distribution (OOD) detection is essential for ensuring the robustness and reliability of object detection systems deployed in safety-critical applications. While prior research has mainly focused on uni-modal detectors or vision-language model (VLM) based classifiers, the potential of VLM-based object detectors in OOD scenarios remains underexplored. In this work, we take the first step toward building OOD object detection methods upon VLMs. We identify two challenges specific to VLM detectors: (i) their text-guided attention enhances foreground with ID labels but treats background uniformly, leaving potential OOD regions unexploited for separating in-distribution (ID) from OOD instances; and (ii) their sigmoid-based multi-label outputs are incompatible with softmax-based OOD scores, calling for scoring functions consistent with VLM probabilistic outputs. Hence, we introduce Negative Label Guided Attention and Scoring (NegAS). To address (i), we propose a negative label guided attention module (NegA), where LLM-generated, visually-similar but semantically-different negative labels are used to guide attention toward potential OOD background regions. To address (ii), we introduce a novel sigmoid-based OOD scoring function (NegS) that leverages both ID and negative labels, producing strong responses for ID instances and suppressed responses for OOD ones. Extensive experiments demonstrate that our approach improves OOD detection performance by a large margin while maintaining ID accuracy, e.g., reducing the FPR95 by 11.4

Continual learning (CL) enables AI models to adapt to evolving environments while mitigating catastrophic forgetting, which is a critical capability for dynamic real-world applications. With the growing popularity of pre-trained Vision Transformer (ViT) models and visual prompt tuning (VPT) technique in CL, this work explores a CL method on top of the ViT-based foundation model, through VPT mechanism with theoretical guarantees. Inspired by the orthogonal projection method, we aim to leverage this approach for VPT to enhance CL performance, particularly in long-term scenarios. However, since the orthogonal projection is originally designed for linear operations in CNNs, applying it to ViTs poses challenges induced by the non-linear self-attention mechanism and the distribution drift within LayerNorm. To address these issues, we deduced two orthogonality conditions to achieve the prompt gradient orthogonal projection, which provide a theoretical guarantee of maintaining stability. Considering the strict orthogonal constraints can diminish model capacity and reduce plasticity, we further propose an importance-aware orthogonal regularization framework. By applying varying degrees of orthogonal constraints to different parameters based on their importance to old and new tasks, the framework adaptively enhances model capacity and thereby promotes long-sequence CL while improving the stability-plasticity trade-off. To implement the proposed approach, a null-space-based approximation solution is employed to efficiently achieve the prompt gradient orthogonal projection. Extensive experiments on various class-incremental learning benchmarks demonstrate that our method achieves state-of-the-art performance across diverse CL scenarios.

Multimodal Chain-of-Thought (MCoT) models have demonstrated impressive capability in complex visual reasoning tasks. Unfortunately, recent studies reveal that they suffer from severe hallucination problems due to diminished visual attention during the generation process.However, visual attention decay is a well-studied problem in Large Vision-Language Models (LVLMs). Considering the fundamental differences in reasoning processes between MCoT models and traditional LVLMs, we raise a basic question: Whether MCoT models have unique causes of hallucinations? To answer this question, we systematically investigate the hallucination patterns of MCoT models and find that fabricated texts are primarily generated in associative reasoning steps, which we term divergent thinking. Leveraging these insights, we introduce a simple yet effective strategy that can effectively localize divergent thinking steps and intervene in the decoding process to mitigate hallucinations. Extensive experiments show that our method outperforms existing methods by a large margin. More importantly, our proposed method can conveniently integrate with other hallucination mitigation methods and further boost their performance. The code will be released.

Two-stage methods have dominated video paragraph captioning for several years. Recently, one-stage methods have gained popularity, eschewing the event detection stage and generating paragraphs directly from untrimmed videos using transformers. Despite consistently improving performance, we argue that current one-stage methods generate paragraphs less relevant to the video content, a shortcoming not adequately reflected by reference-based metrics. In this paper, we aim to enhance the video relevance of generated paragraph captions for one-stage models. Towards this goal, we identify two key issues: (i) one-stage models struggle to detect decisive video content due to insufficient attention, in the absence of an event detection stage; and (ii) current one-stage methods lose some knowledge learned from shallow layers, resulting in missed video content. To address these challenges, we propose the Perturbed Attention Self-Distillation (PASD) framework, which incorporates two key designs: (i) Perturbed Attention: Our key insight is that the removal of important video content results in a decrease in captioning performance. We design a perturbed branch containing a learnable probe to decrease performance. This learned probe then corrects the original attention, helping the model focus on important video content. (ii) Layer-Wise Self-Distillation: We design a layer-wise self-distillation mechanism that transfers knowledge from shallow layers to deep layers, enabling the model to fully exploit multi-level representations learned by different layers. Our novelty lies in integrating these two mechanisms to enhance video relevance without requiring an explicit event detection stage. Experimental results clearly demonstrate that PASD improves video-caption alignment, outperforming both traditional one-stage and two-stage methods. Additionally, PASD maintains computational efficiency, making it suitable for real-world deployment.

Understanding anomalous human behaviors at a fine-grained level remains a major challenge in complex scenarios. Existing video anomaly understanding (VAU) methods often rely on coarse frame-level cues or overlook structured modeling of individual actions, limiting their capacity for reasoning about human interactions and accountability. To address these challenges, we propose TargetVAU, a multimodal anomaly-aware reasoning framework designed for individual-level anomaly recognition and explanation. TargetVAU first extracts both global-level and human-centric visual features using a frozen Vision Transformer (ViT) encoder. An Anomaly-focused Temporal Sampler is then employed to select behaviorally informative frames via a density-aware strategy guided by predicted anomaly scores. A Spatio-Temporal Interaction Graph is constructed to explicitly model interactions among individuals across time and space. These structured representations are fused with prompt embeddings via a frozen Q-Former to form a unified semantic representation. Finally, a large language model fine-tuned with low-rank adaptation (LoRA) performs instruction-guided reasoning to identify anomalous individuals and generate natural language explanations. Extensive experiments on UCCD and HIVAU-70K demonstrate that TargetVAU significantly outperforms existing methods in both accuracy and interpretability, advancing the state of individual-level anomaly understanding in surveillance videos.