Label assignment is a critical component in object detectors, particularly within DETR-style frameworks where the one-to-one matching strategy, despite its end-to-end elegance, suffers from slow convergence due to sparse supervision. While recent works have explored one-to-many assignments to enrich supervisory signals, they often introduce complex, architecture-specific modifications and typically focus on a single auxiliary strategy, lacking a unified and scalable design. In this paper, we first systematically investigate the effects of “one-to-many” supervision and reveal a surprising insight that performance gains are driven not by the sheer quantity of supervision, but by the diversity of the assignment strategies employed. This finding suggests that a more elegant, parameter-efficient approach is attainable. Building on this insight, we propose LoRA-DETR, a flexible and lightweight framework that seamlessly integrates diverse assignment strategies into any DETR-style detector. Our method augments the primary network with multiple Low-Rank Adaptation (LoRA) branches during training, each instantiating a different one-to-many assignment rule. These branches act as auxiliary modules that inject rich, varied supervisory gradients into the main model and are discarded during inference, thus incurring no additional computational cost. This design promotes robust joint optimization while maintaining the architectural simplicity of the original detector. Extensive experiments on different baselines validate the effectiveness of our approach. Our work presents a new paradigm for enhancing detectors, demonstrating that diverse “one-to-many” supervision can be integrated to achieve state-of-the-art results without compromising model elegance.

In the past decade, image restoration has witnessed a remarkable performance leap. However, existing methods still face two critical limitations: i) many methods train distortion-specific models, which struggle to address mixed distortions; ii) most methods employ reconstruction loss during training, with little consideration for the intended use of the restored images (e.g., human viewing, detection, or segmentation), leading to a mismatch between training objectives and application goals. In this paper, we propose a scalable framework for image restoration and enhancement built upon the reinforcement learning (RL) paradigm. At the core of our framework, the RL agent is capable of accurately perceiving distortion contexts and dynamically adjusting diverse restoration operators, which ensures our framework to effectively handle mixed-distortions, and generalize well to new tasks with minor additional cost. Furthermore, RL relaxes the differentiability constraints on optimization objectives, which enables our framework to directly optimize the restoration process toward application goals, thereby bridging the gap between training objectives and downstream demands. Extensive objective and subjective evaluations demonstrate that our method not only generates visually pleasing results for human perception but also significantly improves the performance of machine vision tasks such as detection and segmentation. We further validate the framework’s scalability through generalization evaluations on unseen distortions and multi-objective optimization experiments. The results confirm the model’s superior few-shot and zero-shot capabilities compared to existing methods, as well as its flexibility in addressing multiple competing objectives. Finally, comprehensive ablation studies and hyperparameter analyses provide valuable insights into the design choices and behavior of our framework.

Previous feed-forward 4D reconstruction methods either predict per-frame static point clouds, ignoring foreground motion, or estimate point cloud trajectories while being limited to small camera motions. This restricts their ability to aggregate observations over time and reconstruct complete dynamic scenes under large viewpoint changes. To address this limitation, we propose OmniX, a feed-forward 4D reconstruction framework that predicts dense 3D point trajectories for every pixel from videos with large camera motion. OmniX decouples dynamic motion modeling from static geometry prediction and represents motion using a compact set of dynamic tokens. By leveraging the sparse and low-rank structure of 3D motion, these tokens generate trajectory fields for all pixels across all images while efficiently preserving global interactions. To facilitate training, we further build an automatic UE5-based 4D data engine and introduce a large-scale dataset containing 80K scenes and 1.28M multi-view videos with full geometric annotations. OmniX achieves state-of-the-art performance on dense 3D point trajectory prediction and 3D point tracking, while also demonstrating competitive results on video depth estimation and camera pose estimation.

Invisible image watermarking enables copyright protection by embedding ownership signals into the images without causing perceptual degradation. However, existing watermarking methods are primarily evaluated under single perturbations, which fail to benchmark their practical robustness. To address this evaluation gap, we introduce the Composite Perturbation Benchmark (CPBench) that systematically evaluates watermark robustness under compositions of multiple perturbations, as commonly encountered in real-world scenarios. Our analysis reveals that existing methods suffer significant performance degradation under composite perturbations—a critical robustness challenge that is often overlooked by current benchmarks. To address this limitation, we propose CPMark, a perturbation-aware diffusion watermarking framework that embeds structured watermark signals into the hidden features of the U-Net and employs a differentiable perturbation simulator to significantly enhance robustness during training. Extensive experiments demonstrate that CPMark significantly outperforms existing methods under composite perturbations and maintains high visual fidelity, while also achieving state-of-the-art performance under single distortions.

Multimodal Large Language Models (MLLMs) are increasingly being deployed as automated content moderators. Within this landscape, we uncover a critical threat: Adversarial Smuggling Attacks. Unlike adversarial perturbations (for misclassification) and adversarial jailbreaks (for harmful output generation), adversarial smuggling exploits the Human-AI capability gap. It encodes harmful content into human-readable visual formats that remain AI-unreadable, thereby evading automated detection and enabling the dissemination of harmful content. We classify smuggling attacks into two pathways: (1) Perceptual Blindness, disrupting text recognition; and (2) Reasoning Blockade, inhibiting semantic understanding despite successful text recognition. To evaluate this threat, we constructed SmuggleBench, the first comprehensive benchmark comprising 1,700 adversarial smuggling attack instances. Evaluations on SmuggleBench reveal that both proprietary (e.g., GPT-5) and open-source (e.g., Qwen3-VL) state-of-the-art models are vulnerable to this threat, producing Attack Success Rates (ASR) exceeding 90

Current image quality assessment methods are heavily biased towards global distortions (e.g., noise, blur), neglecting local perceptual artifacts such as ghosting, lens flare, and moire effects. Although significant progress has been made in artifact removal, the fundamental problem of automatic artifact detection remains largely unexplored. In this paper, we formalize the Image Perceptual Artifact Detection (IPAD) task to address this gap. We contribute a benchmark dataset comprising 3,520 artifact images, including 520 real-captured and 3,000 synthetic samples, each paired with pixel-level masks across three representative artifact categories. The core challenge of IPAD lies in the localized, subtle, and semantically weak nature of these artifacts, which makes them prone to missed detection. To overcome this, we introduce IPAD-CLIP, a novel framework built upon CLIP that enhances artifact discrimination in both textual and visual spaces while preserving generalization capabilities. Our key insight is that local artifacts often exhibit strong correlations with specific semantic contexts. Accordingly, we learn artifact-aware text embeddings to explicitly model the object-artifact relationships, resulting in enhanced representations that clear differentiate between clean and artifact prompts. These text embeddings are then used as anchors to shift the visual encoder's attention from high-level semantics to subtle, low-level artifacts. Extensive experiments demonstrate that IPAD-CLIP offers a resource-efficient adaptation of CLIP for detection, significantly outperforming advanced image anomaly detection and manipulation detection methods on our benchmark. To the best of our knowledge, this is the first study addressing multi-class local perceptual artifact detection in terms of both dataset and model.

Spiking Neural Networks (SNNs) have attracted increasing attention due to their brain-inspired computing paradigm and superior energy efficiency on specialized hardware. However, the discrete and non-differentiable nature of spike signals inevitably leads to information loss during signal propagation, which limits the performance of SNNs compared with Artificial Neural Networks (ANNs). To address this issue, we propose ICE-Net, an information-complementary enhancement method for SNNs. Specifically, it employs a customized simplified ICE branch to learn complementary features and delivers them to the corresponding stages of the SNN branch, thereby mitigating information loss during propagation. Extensive experiments on multiple benchmark datasets demonstrate that ICE-Net consistently outperforms existing advanced methods. In particular, ICE-Net achieves 81.62% accuracy on CIFAR-100 with only 4 timesteps, surpassing the previous best result by 2.11%, while reducing energy consumption by approximately 25% compared with ANNs. In addition, we develop several simplified complementary modules that provide flexible trade-offs between accuracy and energy consumption.

Spiking Neural Networks (SNNs) have garnered increasing attention for their biological plausibility, energy efficiency, and temporal modeling capability. Due to the non-differentiability of spike generation, a widely used supervised training method for SNNs is backpropagation through time with surrogate gradients, which achieves competitive performance with a small number of timesteps. Intuitively, scaling timesteps should further improve performance by enriching temporal dynamics. However, we observe timestep scaling paradox (TSP), a counter-intuitive degradation in accuracy when scaling timesteps. We investigate TSP and link it to long-term temporal gradient vanishing and weakened cross-timestep dependencies. To address this, we propose the Timestep-Scalable (TS) neuron model. It introduces long-term memory reconsolidation to enhance cross-timestep information flow and enable effective learning with more timesteps. In parallel, a temporal forgetting mechanism periodically truncates the accumulation path, suppressing excessive temporal information buildup and improving training stability. Supported by theoretical analysis and extensive experiments, TS consistently improves performance when scaling timesteps. Beyond gains from timestep scaling, it attains state-of-the-art results on EEG signals, event-based recognition, and time-series forecasting, while remaining strong on conventional image classification and object detection datasets.

Adapting vision-language models to remote sensing imagery presents a fundamental challenge: both the visual and linguistic distributions of satellite data lie far outside natural image pretraining corpora. Despite this, prompting remains the dominant deployment paradigm, driven by the assumption that domain-specific language can guide frozen model representations toward specialized tasks. We test this assumption directly on a domain where the mismatch is prominent: cloud segmentation for satellite imagery. Using CLIPSeg on the CloudSEN12+ cloud segmentation benchmark, we evaluate 60 prompt variants spanning simple labels, domain terminology, appearance descriptors, and contextual cues, finding that every variant underperforms the zero-shot baseline (0.255 mIoU), with engineered prompts scoring as low as 0.07 mIoU. No amount of linguistic refinement bridges the gap between CLIP's natural image representations and satellite spectral imagery. In contrast, supervised fine-tuning with just 0.1

The past year has seen over 20 open-source document parsing models, yet thefield still benchmarks almost exclusively on OmniDocBench, a 1,355-pagemanually annotated dataset whose top scores have saturated above 90

High-performance Face Anti-Spoofing (FAS) systems depend critically on extensive labeled data, while Multi-Modal FAS (MMFAS) exacerbates this dependency due to the increased complexity of collecting and annotating multi-modal data. To address this challenge, we propose a novel framework called Multi-Modal Self-Supervised FAS (M2S2FAS), which introduces a new network architecture alongside three carefully designed pretext tasks. The primary objective of M2S2FAS is to model inter-modal correspondences, thereby generating versatile pre-trained weights that can be effectively utilized across different modalities. Our framework demonstrates its efficacy through minimal fine-tuning, achieving high-performance levels. Extensive experimental evaluations on widely used multi-modal FAS datasets validate the superiority of the proposed method.

Large Vision-Language Models (LVLMs) face significant challenges in long video understanding due to the excessive computational cost and information loss associated with uniform sampling. Existing keyframe selection methods often treat video frames as atomic entities and allocate visual budgets equally, thereby overlooking high-level semantic structures and introducing substantial redundancy. To address these limitations, we propose GMM-EVA (Gaussian Mixture Modeling for Event-Aware Visual Allocation), which leverages Gaussian Mixture Models to model event-level structure from discrete frame-wise observations. A differentiated allocation strategy is then applied to preserve one primary high-resolution keyframe per event for high-fidelity detail, while utilizing lower-resolution secondary keyframes to maintain temporal context and optimize token budgets. GMM-EVA is a training-free, plug-and-play framework that generalizes robustly across various relevance measures and downstream LVLMs. Extensive experiments on multiple long video benchmarks demonstrate that our method significantly outperforms uniform sampling. Notably, GMM-EVA achieves comparable performance to baseline selection methods while utilizing only approximately half of the visual token budget, highlighting its superior efficiency and effectiveness.

Spiking Neural Networks (SNNs) naturally process visual inputs across multiple timesteps, offering rich temporal dynamics and energy-efficient computation. However, the temporally invariant supervision commonly used in training tends to reinforce the same dominant response patterns across timesteps, leading to redundant representations and limiting temporal discriminability.To overcome this limitation, we introduce \emph{Temporal Representation Enhancement} (TRE), a novel learning-to-forget paradigm that encourages more diverse and complementary temporal representations. TRE identifies high-contribution semantic patterns through class-specific contribution estimation and temporal accumulation, and selectively suppresses them using a dynamic modulation strategy. By redirecting the model’s attention toward alternative yet informative semantic cues, TRE promotes the learning of complementary features across timesteps.This approach not only strengthens the temporal discriminative capacity of SNNs but also enables more effective multi-timestep learning by leveraging richer semantic information. Extensive experiments on both static image datasets and dynamic neuromorphic datasets demonstrate that TRE consistently improves classification accuracy and feature diversity across different SNN backbones.

The detection of face forgery has become increasingly vital due to the severe security concerns posed by face manipulation techniques. While recent studies on forgery detection have demonstrated promising results when the training and testing samples come from the same domains, the problem remains challenging when attempting to extend the detector to unseen methods. In this work, we propose an innovative approach to enhance the generalization capability of forgery detection methods by exploring degradation inconsistency clues interspersed between the background and the manipulated face regions. Our motivation stems from the observation that digital photos undergo different degradation during acquisition and transmission, resulting in backgrounds and faces from different sources containing distinct degradation patterns in the forged faces. The proposed framework, termed the Degradation Consistency Learning Framework, integrates two core components: a data generation network that modulates degradation transformations to obtain tampered facial images, and a detection network that mines degradation inconsistency clues from both spatial and frequency domains. These two components are tightly coupled through adversarial training, forming a dynamic architecture akin to a Generative Adversarial Network (GAN). Experimental results on different benchmark and evaluation protocols (i.e., indataset and cross-dataset) have demonstrated the effectiveness of our method.

Multimodal foundation models have demonstrated impressive capabilities across diverse tasks. However, their potential as plug-and-play solutions for missing modality re construction remains underexplored. To bridge this gap, we identify and formalize three potential paradigms for missing modality reconstruction, and perform a comprehensive evaluation across these paradigms, covering 42 model variants in terms of reconstruction accuracy and adaptability to downstream tasks. Our analysis reveals that current foundation models often fall short in two critical aspects: (i) fine-grained semantic extraction from the available modalities, and (ii) robust validation of generated modalities. These limitations lead to suboptimal and, at times, misaligned generations. To address these challenges, we propose anagentic framework tailored for missing modality reconstruction. This framework dynamically formulates modality-aware mining strategies based on the input context, facilitating the extraction of richer and more discriminative semantic features. In addition, we introduce a self-refinement mechanism, which iteratively verifies and enhances the quality of generated modalities through internal feedback. Experimental results show that our method reduces FID for missing image reconstruction by at least 14% and MER for missing text reconstruction by at least 10% compared to baselines. Code is released at: https://github.com/Guanzhou-Ke/AFM2.

Most lightweight Single Object Tracking methods achieve real-time speed with a static matching paradigm. This design is efficient, but often struggles under heavy occlusion and large appearance variation. Stronger trackers address these challenges with dynamic templates or memory networks, yet their cost is usually too high for CPU-based edge devices. We bridge this gap with LTFTrack, a lightweight framework that introduces temporal information through feature reuse. Our Cross-frame Temporal Fusion (CTF) module extracts dynamic target cues directly from the previous search region, avoiding the cost of reprocessing additional inputs. To handle the spatial mismatch between dynamic search features and static template features, we further introduce a self-distillation alignment mechanism. During training, a ground-truth-guided teacher supervises feature alignment, while inference incurs no extra cost. Experiments on LaSOT, LaSOText, and GOT-10k show that LTFTrack performs strongly across benchmarks. On LaSOT, it reaches an AUC of $\text{6 6. 8 \%}, \text{2. 1 \%}$ higher than AsymTrack-B, while running at $\text{4 5}$ FPS on CPU.

Recently, image restoration (IR) methods have demonstrated promising potential by leveraging pre-trained cross-modal priors to provide explicit text guidance. However, these methods struggle with mixed degradations, as they cannot effectively process multi-instruction prompts, often yielding unsatisfactory outputs. They typically rely on a single text prompt applied iteratively to address the most dominant degradation. The process is computationally intensive and time-consuming. In this paper, we propose a novel IR framework capable of effectively understanding lengthy and complex instruction prompts and restoring multiple degradations simultaneously. Our key innovation is the attribute-aware attention (ATT) module, which disentangles low-level image features into shared basis attribute elements by modeling the intrinsic relationships between degradations and attributes. The ATT module extracts fine-grained degradation cues from multi-level textual descriptions and establishes cross-modal projections in the latent attribute space, generating a set of attention maps. These maps explicitly localize regions associated with each attribute, enhancing the interpretability of the restoration process. Extensive experiments demonstrate that our method outperforms existing methods across various tasks. Furthermore, our model exhibits strong generalization when handling unseen degradations.

Video diffusion transformers (vDiTs) generate high quality video but introduce extremely high compute cost due to the long diffusion timesteps and self attention computation. As diffusion timesteps are reduced, the computation cost of self attention becomes the dominant bottleneck. Existing acceleration approaches largely inherit sparse attention techniques from large language models, which fail to consider the unique spatiotemporal correlation of video data. This paper presents Kaleido, an algorithm hardware codesign that accelerates all operations in vDiTs by exploiting channel-wise spatiotemporal correlations in latent space. Based on this insight, we propose a lightweight channelwise reuse algorithm that skips redundant computations by reusing partial results while preserving higher generative quality than prior methods (>17 dB). To efficiently support this algorithm, we design a systolic array like accelerator with reconfigurable processing elements and a lightweight data dispatcher to mitigate irregular sparsity and data access patterns introduced by our reuse algorithm. Evaluations across three mainstream vDiT models show that Kaleido achieves up to 5.9x speedup and 16.0x energy savings over state of the art accelerators.

Knowledge-based Visual Question Answering (KB-VQA) requires models to retrieve visual entities matching the query image from large-scale encyclopedic knowledge bases and answer related questions. Existing multimodal Retrieval Augmented Generation (mRAG) methods rely on global visual features to match candidate entities, yet when the knowledge base contains numerous visually similar entities, the retriever struggles to distinguish them, populating the candidate set with visually similar but factually mismatched distractors. Since subsequent processing steps such as noise filtering are also confined to this fixed candidate set, errors from failed retrieval inevitably propagate to the final answer. To address these challenges, we propose MMAgent-R^2, an agentic mRAG framework that integrates visual reranking and active rejection as its internal verification mechanism. Visual reranking directly compares query and candidate images, capturing discriminative details beyond textual descriptions to precisely identify the target entity among similar candidates; active rejection discards unreliable results and retrieves additional candidates when no confident match is found, moving beyond the fixed candidate pool. We design a composite reward function with step-level verification rewards and achieve joint optimization of external retrieval, internal verification, and answer generation via GRPO training. Experiments on InfoSeek, E-VQA, and MMhops demonstrate that achieves state-of-the-art performance, with particularly notable advantages in challenging retrieval scenarios and complex multi-image multi-hop reasoning tasks.