CVPR 2026 论文列表

Multimodal large language models (MLLMs) have demonstrated remarkable capabilities in aligning visual inputs with natural language outputs. Yet, the extent to which generated tokens depend on visual modalities remains poorly understood, limiting interpretability and reliability. In this work, we present EAGLE, a lightweight black-box framework for explaining autoregressive token generation in MLLMs. EAGLE attributes any selected tokens to compact perceptual regions while quantifying the relative influence of language priors and perceptual evidence. The framework introduces an objective function that unifies sufficiency (insight score) and indispensability (necessity score), optimized via greedy search over sparsified image regions for faithful and efficient attribution. Beyond spatial attribution, EAGLE performs modality-aware analysis that disentangles what tokens rely on, providing fine-grained interpretability of model decisions. Extensive experiments across open-source MLLMs show that EAGLE consistently outperforms existing methods in faithfulness, localization, and hallucination diagnosis, while requiring substantially less GPU memory. These results highlight its effectiveness and practicality for advancing the interpretability of MLLMs.

Recent Large Vision-Language Models (LVLMs) have shown impressive capabilities in multimodal understanding and generation. Despite this progress, they remain prone to *hallucination*, where model outputs conflict with the visual input due to an over-reliance on textual priors. Existing inference-time mitigation approaches frequently depend on multi-pass or contrastive decoding, which increases latency and limits their applicability in real-time settings. To address this limitation, we propose **CausalLens**, a training-free and single-pass intervention that directly adjusts the decoder hidden states to strengthen visual grounding. By decomposing attention heads into visual, textual, and system prompt pathways, CausalLens identifies visually reliable heads using a sensitivity measure and selectively adjusts their mid-layer hidden-state contributions. A projection-aligned correction further stabilizes these adjusted states after multi-head fusion, ensuring that the enhanced visual information is preserved throughout decoding. Extensive experiments across multiple hallucination benchmarks and LVLM architectures demonstrate that CausalLens consistently improves visual fidelity while adding negligible computational overhead. The method requires no fine-tuning or architectural changes, making it well-suited for practical, latency-sensitive applications.

Existing anomaly detection methods often treat the modality and class as independent factors. Although this paradigm has enriched the development of AD research branches and produced many specialized models, it has also led to fragmented solutions and excessive memory overhead. Moreover, reconstruction-based multi-class approaches typically rely on shared decoding paths, which struggle to handle large variations across domains, resulting in distorted normality boundaries, domain interference, and high false alarm rates.To address these limitations, we propose UniMMAD, a unified framework for multi-modal and multi-class anomaly detection. At the core of UniMMAD is a Mixture-of-Experts (MoE)-driven feature decompression mechanism, which enables adaptive and disentangled reconstruction tailored to specific domains.This process is guided by a "general - specific" paradigm. In the encoding stage, multi-modal inputs of varying combinations are compressed into compact, general-purpose features. The encoder incorporates a feature compression module to suppress latent anomalies, encourage cross-modal interaction, and avoid shortcut learning.In the decoding stage, the general features are decompressed into modality-specific and class-specific forms via a sparsely-gated cross MoE, which dynamically selects expert pathways based on input modality and class. To further improve efficiency, we design a grouped dynamic filtering mechanism and an MoE-in-MoE structure, reducing MoE parameter usage by approximately 75% while maintaining sparse activation and fast inference. UniMMAD achieves state-of-the-art performance on 9 anomaly detection datasets, spanning 3 fields, 12 modalities, and 66 classes. Code is publicly available at https://github.com/yuanzhao-CVLAB/UniMMAD.

Estimating human gaze targets from images in-the-wild is an important and formidable task. Existing approaches primarily employ brittle, multi-stage pipelines that require explicit inputs, like head bounding boxes and human pose, in order to identify the subject of gaze analysis. As a result, detection errors can cascade and lead to failure. Moreover, these prior works lack the flexibility of specifying the gaze analysis task via natural language prompting, an approach which has been shown to have significant benefits in convenience and scalability for other image analysis tasks. To overcome these limitations, we introduce the Promptable Gaze Target Estimation (PGE) task, a new end-to-end, concept-driven paradigm for gaze analysis. PGE conditions gaze prediction on flexible user text or visual prompts (e.g., "the boy in the red shirt" or "person in point [0.52, 0.48]") to identify a specific subject for gaze analysis. This approach integrates subject localization with gaze estimation, and eliminates the rigid dependency on intermediate analysis stages. We develop a scalable data engine to generate Gaze-Co (Gaze Estimation with Concepts), a dataset and benchmark of 120K high-quality, prompt-annotated image pairs. We also propose GazeAnywhere, the first model designed for PGE. GazeAnywhere uses a transformer-based detector to fuse features from frozen encoders and simultaneously solves subject localization, in/out-of-frame presence, and gaze target heatmap estimation. GazeAnywhere achieves state-of-the-art performance on multiple PGE benchmarks, setting a strong baseline for this new problem even on a difficult out-of-domain, real-world clinical dataset. GazeAnywhere is open-sourced in github.com/IrohXu/GazeAnywhere.

Personalized generation models for a single subject have demonstrated remarkable effectiveness, highlighting their significant potential. However, when extended to multiple subjects, existing models often exhibit degraded performance, particularly in maintaining subject consistency and adhering to textual prompts. We attribute these limitations to the absence of high-quality multi-subject datasets and refined post-training strategies. To address these challenges, we propose a scalable multi-subject data generation pipeline that leverages powerful single-subject generation models to construct diverse and high-quality multi-subject training data. Through this dataset, we first enable single-subject personalization models to acquire knowledge of synthesizing multi-image and multi-subject scenarios. Furthermore, to enhance both subject consistency and text controllability, we design a set of Pairwise Subject-Consistency Rewards and general-purpose rewards, which are incorporated into a refined reinforcement learning stage. To comprehensively evaluate multi-subject personalization, we introduce a new benchmark that assesses model performance using seven subsets across three dimensions. Extensive experiments demonstrate the effectiveness of our approach in advancing multi-subject personalized image generation. Github Link: https://github.com/wang-shulei.

Chain-of-Thought (CoT) is a key technique for enhancing the reasoning capabilities of Vision Language Models (VLMs). Existing methods often employ Reinforcement Learning (RL) with external constraints to align the model's reasoning process with human cognitive patterns. However, we argue that the model's intrinsic reasoning paths may differ from human cognition, and that forcing such alignment can constrain the model's potential and even degrade its performance. To address this, we propose leveraging the model's intrinsic self-evaluation to guide its optimization. We hypothesize that a model's self-generated confidence scores are effective indicators of its reasoning quality. Based on this evaluation metric, we design two novel reward functions: (1) Sequential Confidence Rigorous Evaluation (SCRE) for challenging problems that demand strict logical reasoning, and (2) Intra-group Score Re-ranking (IGSR) for general-purpose, open-ended scenarios. We name our method Video-RAISE (Reasoning Alignment through Intrinsic Self-Evaluation). Comprehensive experiments on six video understanding benchmarks demonstrate that Video-RAISE achieves state-of-the-art (SOTA) performance, significantly outperforming previous methods and even proprietary models, e,g. GPT-4o and Gemini 1.5 pro. For instance, on the VideoMMMU benchmark, our Video-RAISE achieves a new SOTA accuracy of 52.8%, outperforming the previous best model by a significant 3.0%. In addition, our method achieves a reasoning path consistency of 90%, which is double that of the Qwen2.5-VL-Instruct and even surpasses the performance of supervised fine-tuning.

We introduce RefTon, a flux-based person-to-person virtual try-on framework that enhances garment realism through unpaired visual references. Unlike conventional approaches that rely on complex auxiliary inputs such as body parsing and warped mask or require finely designed extract branches to process various input conditions, RefTon streamlines the process by directly generating try-on results from a source image and a target garment, without the need for structural guidance or auxiliary components to handle diverse inputs. Moreover, inspired by human clothing selection behavior, RefTon leverages additional reference images (the target garment worn on different individuals) to provide powerful guidance for refining texture alignment and maintaining the garment details. To enable this capability, we built a dataset containing unpaired reference images for training. Extensive experiments on public benchmarks demonstrate that RefTon achieves competitive or superior performance compared to state-of-the-art methods, while maintaining a simple and efficient person-to-person design. \thanks Code and dataset is available at: https://github.com/360CVGroup/RefTon.

Continual Visual Question Answering (Continual VQA) poses unique challenges for multimodal continual learning, requiring models to incrementally acquire new knowledge while preserving visual-semantic grounding across tasks. However, existing benchmarks hinder fair and robust evaluation of such capabilities, as they allow models to exploit dataset biases rather than demonstrate genuine continual reasoning. We identify two structural flaws in current benchmark design. First, shared answer vocabularies across tasks encourage answer memorization, inflating performance and underestimating forgetting. Second, static answer priors within each task make the training and test answer distributions nearly identical, obscuring robustness under distribution shifts. To address these issues, we introduce UCo-VQA, an Unbiased benchmark suite that enforces token-level disjoint answer spaces across tasks and introduces intra-task train-test distribution shifts, enabling fairer assessment of forgetting and generalization in multimodal continual learning. We further provide a parameter-efficient baseline that mitigates forgetting and enhances grounding through question-only replay and dual-level distillation, offering a lightweight and memory-efficient framework for continual adaptation. Extensive experiments on UCo-VQA reveal that prior methods substantially overestimate performance under biased setups, while our approach achieves state-of-the-art results, improving robustness and retention by up to 4.18% and 2.21%, respectively.

Advances in generative modeling have made it increasingly easy to fabricate realistic portrayals of individuals, creating serious risks for security, communication, and public trust. Detecting such person-driven manipulations requires systems that not only distinguish altered content from authentic media but also provide clear and reliable reasoning. In this paper, we introduce TriDF, a comprehensive benchmark for interpretable DeepFake detection. TriDF contains high-quality forgeries from advanced synthesis models, covering 16 DeepFake types across image, video, and audio modalities. The benchmark evaluates three key aspects: Perception, which measures the ability of a model to identify fine-grained manipulation artifacts using human-annotated evidence; Detection, which assesses classification performance across diverse forgery families and generators; and Hallucination, which quantifies the reliability of model-generated explanations. Experiments on state-of-the-art multimodal large language models show that accurate perception is essential for reliable detection, but hallucination can severely disrupt decision-making, revealing the interdependence of these three aspects. TriDF provides a unified framework for understanding the interaction between detection accuracy, evidence identification, and explanation reliability, offering a foundation for building trustworthy systems that address real-world synthetic media threats.

Embodied Question Answering (EQA) connects perception, reasoning, and interaction within embodied environments. However, existing datasets and benchmarks remain fragmented, each focusing on a limited subset of reasoning skills such as spatial understanding or procedural reasoning, without offering a unified large-scale framework for comprehensive evaluation. We present EQA-Decision, a large-scale embodied QA dataset that systematically covers four complementary dimensions of embodied reasoning: static scene construction, spatial understanding, task dynamics reasoning, and instant decision. The dataset contains over four million question-answer pairs with hierarchical annotations across diverse embodied scenarios. In addition, we develop RoboDecision, a strong baseline model aligned with the EQA-Decision Benchmark, providing a unified framework that jointly evaluates perception, reasoning, and action-level decision-making in embodied environments. Results demonstrate that EQA-Decision effectively benchmarks and enhances VLM capabilities in spatial and interaction reasoning, providing a solid foundation for advancing embodied intelligence research.

Synthesizing text-driven 3D human motion within realistic scenes requires learning both semantic intent ("walk to the couch") and physical feasibility (e.g., avoiding collisions). Current methods use generative frameworks that simultaneously learn high-level planning and low-level contact reasoning, and rely on computationally expensive 3D scene data such as point clouds or voxel occupancy grids. We propose SceMoS, a scene-aware motion synthesis framework that shows that structured 2D scene representations can serve as a powerful alternative to full 3D supervision in physically grounded motion synthesis. SceMoS disentangles global planning from local execution using lightweight 2D cues and relying on (1) a text-conditioned autoregressive global motion planner that operates on a bird's-eye-view (BEV) image rendered from an elevated corner of the scene, encoded with DINOv2 features, as the scene representation, and (2) a geometry-grounded motion tokenizer trained via a conditional VQ-VAE, that uses 2D local scene heightmap, thus embedding surface physics directly into a discrete vocabulary. This 2D factorization reaches an efficiency-fidelity trade-off: BEV semantics capture spatial layout and affordance for global reasoning, while local heightmaps enforce fine-grained physical adherence without full 3D volumetric reasoning. SceMoS achieves state-of-the-art motion realism and contact accuracy on the TRUMANS benchmark, reducing the number of trainable parameters for scene encoding by over 50%, showing that 2D scene cues can effectively ground 3D human-scene interaction. Project URL: https://anindita127.github.io/SceMoS.

We study instruction-guided editing of egocentric videos for interactive AR applications. While recent AI video editors perform well on third-person footage, egocentric views present unique challenges -- including rapid egomotion, and frequent hand-object interactions -- that create a significant domain gap. Moreover, existing offline editing pipelines suffer from high latency, limiting real-time interaction.To address these issues, we present a complete ecosystem for egocentric video editing. First, we construct EgoEditData, a carefully designed and manually curated dataset specifically designed for egocentric editing scenarios, featuring rich hand-object interactions, while explicitly preserving hands. Second, we develop EgoEdit, an instruction-following egocentric video editor that supports real-time streaming inference on a single GPU. Finally, we introduce EgoEditBench, an evaluation suite targeting instruction faithfulness, hand and interaction preservation, and temporal stability under egomotion.Across both egocentric and general editing tasks, EgoEdit produces temporally stable, instruction-faithful results with interactive latency. It achieves clear gains on egocentric editing benchmarks--where existing methods struggle--while maintaining performance comparable to the strongest baselines on general editing tasks. EgoEditData and EgoEditBench will be made public for the research community.

Cardiac magnetic resonance (CMR) is the clinical gold standard for assessing cardiovascular diseases, but its interpretation relies on expert experience and remains challenging, particularly for identifying rare diseases. Existing automated methods lack interpretable reasoning processes, limiting clinical adoption. Although vision-language models (VLMs) possess basic visual understanding and text generation capabilities, they still lack verifiable reasoning chains in medical diagnosis and underperform on minority classes in long-tail distributions. To address these challenges, we propose CMR-RD, to our knowledge the first VLM for interpretable diagnosis in CMR, capable of generating explicit diagnostic chains aligned with imaging evidence. We construct a CMR dataset that reflects real-world clinical distributions, comprising five disease categories (including two rare conditions) plus normal controls. Building on this, the general-purpose VLM is aligned to medical and CMR semantics using large-scale medical vision-text data, and cold-start training is used to enhance its understanding of medical concepts and basic reasoning. To enhance reasoning and performance on rare samples, we propose Group Phase Policy Optimization (GPPO), which combines online multi-stage reinforcement learning (RL) with adaptive sampling. GPPO enables the model to proactively explore rare and underperforming classes, thereby effectively mitigating long-tail bias. Experiments demonstrate that CMR-RD achieves highly competitive accuracy and reasoning-chain correctness compared with medical and general VLMs, shows stronger recognition of rare categories.

Recent advancements in Video Large Language Models (VideoLLMs) have enabled strong performance across diverse multimodal video tasks. To reduce the high computational cost of processing dense video frames, efficiency-oriented methods such as frame selection have been widely adopted. While effective at minimizing redundancy, these methods often cause notable performance drops on tasks requiring temporal reasoning. Unlike humans, who can infer event progression from sparse visual cues, VideoLLMs frequently misinterpret temporal relations when intermediate frames are omitted. To address this limitation, we explore visual prompting (VP) as a lightweight yet effective way to enhance temporal understanding in VideoLLMs. Our analysis reveals that simply annotating each frame with explicit ordinal information helps the model perceive temporal continuity. This visual cue also supports frame-level referencing and mitigates positional ambiguity within a sparsely sampled sequence. Building on these insights, we introduce ViKey, a training-free framework that combines VP with a lightweight Keyword-Frame Mapping (KFM) module. KFM leverages frame indices as dictionary-like keys to link textual cues to the most relevant frames, providing explicit temporal anchors during inference. Despite its simplicity, our approach substantially improves temporal reasoning and, on some datasets, preserves dense-frame baseline performance with as few as 20% of frames.

The pursuit of a universal AI-generated image (AIGI) detector often relies on aggregating data from numerous generators to improve generalization. However, this paper identifies a paradoxical phenomenon we term the "Benefit then Conflict" dilemma, where detector performance stagnates and eventually degrades as source diversity expands. Our systematic analysis, diagnoses this failure by identifying two core issues: severe data-level heterogeneity which causes the feature distributions of real and synthetic images to increasingly overlap, and a critical model-level bottleneck from fixed, pretrained encoders that cannot adapt to the rising complexity. To address these challenges, we propose Generator-Aware Prototype Learning (GAPL) , a framework that constrain representation with a structured learning paradigm. GAPL learns a compact set of canonical forgery prototypes to create a unified, low-variance feature space, effectively countering data heterogeneity. To resolve the model bottleneck, it employs a two-stage training scheme with Low-Rank Adaptation, enhancing its discriminative power while preserving valuable pretrained knowledge. This approach establishes a more robust and generalizable decision boundary. Through extensive experiments, we demonstrate that GAPL achieves state-of-the-art performance, showing superior detection accuracy across a wide variety of GAN and diffusion-based generators. Code is available at https://github.com/UltraCapture/GAPL

Object removal requires eliminating not only the target object but also its associated visual effects such as shadows and reflections. However, diffusion-based inpainting and removal methods often introduce artifacts, hallucinate contents, alter background, and struggle to remove object effects accurately. To address these challenges, we propose ObjectClear, a novel framework that decouples foreground removal from background reconstruction via an adaptive target-aware attention mechanism. This design empowers the model to precisely localize and remove both objects and their effects while maintaining high background fidelity. Moreover, the learned attention maps are leveraged for an attention-guided fusion strategy during inference, further enhancing visual consistency. To facilitate the training and evaluation, we construct OBER, a large-scale dataset for OBject-Effect Removal, which provides paired images with and without object-effects, along with precise masks for both objects and their effects. The dataset comprises high-quality captured and simulated data, covering diverse objects, effects, and complex multi-object scenes. Extensive experiments demonstrate that ObjectClear outperforms prior methods, achieving superior object-effect removal quality and background fidelity, especially in challenging scenarios.

Diffusion models have shown promising performance as data-driven priors for computational imaging, as well as some capacity to detect out-of-distribution (OOD) images. However, existing approaches to OOD detection often require some knowledge of the shifted distribution, fail to detect subtle or localized distribution shifts, and operate on full images, rather than the indirect measurements available in inverse problems. We propose an OOD detection metric based on the Kullback-Leibler divergence between the diffusion prior and the posterior distribution, that (i) does not require any calibration data or knowledge of the shifted distribution, and (ii) can detect whole images as OOD as well as localize OOD patches within an image. Experimentally, we show that this metric can detect subtle yet semantically meaningful distribution shifts, such as the shift from healthy liver CT scans to those with tumors, and generalizes across different types of diffusion models, datasets, and inverse problems.

Generating realistic conversational gestures are essential for achieving natural, socially engaging interactions with digital humans. However, existing methods typically map a single audio stream to a single speaker's motion, without considering social context or modeling the mutual dynamics between two people engaging in conversation. We present DyaDiT, a multi-modal diffusion transformer that generates contextually appropriate human motion from dyadic audio signals. Trained on Seamless Interaction Dataset, DyaDiT takes dyadic audio with optional social-context tokens to produce context-appropriate motion. It fuses information from both speakers to capture interaction dynamics, uses a motion dictionary to encode motion priors, and can optionally utilize the conversational partner's gestures to produce more responsive motion. We evaluate DyaDiT on standard motion generation metrics and conduct quantitative user studies, demonstrating that it not only surpasses existing methods on objective metrics but is also strongly preferred by users, highlighting its robustness and socially favorable motion generation. Code and models will be released upon acceptance.

We introduce the novel task of egocentric self-emotion tracking, which aims to infer an individual's evolving emotions from egocentric multimodal streams such as voice, visual surroundings, semantic subtext, and eye-tracking signals. To establish this research direction, we present: (1) OSMO dataset, a large-scale annotation effort on 110 hours of existing bilingual smart-glasses recordings, establishing the largest egocentric emotion dataset and the first with subject-wise emotion timelines; (2) OSMO benchmark, a suite of five tasks (emotion recognition, sentiment, intensity, localization, and reasoning), that redefine emotion understanding as a continuous, context-aware process rather than discrete classification of trimmed videos; (3) OSIRIS, a large multimodal model that tracks emotions over time by reasoning over the user's personal emotion history, current expressions, and egocentric observations. Extensive evaluations show that OSIRIS achieves a state-of-the-art performance, delivering, for the first time, coherent emotion timelines from egocentric data. Project website: https://osmo-emos.github.io.