CVPR 2025 论文列表

Recent studies have focused on introducing pre-trained foundation models into semi-supervised learning (SSL) tasks. Nevertheless, these foundation models can exhibit biases toward different classes and tend to generate imbalanced pseudo-labels for SSL. Thus, efforts have been made to introduce the logit adjustment offset to reduce the inherent bias in foundation models for SSL tasks.Despite their success, existing foundation model-based SSL methods face challenges: 1) unreliability in the estimated logit adjustment offset, 2) overlooking the potential of linguistic knowledge in capturing model biases and 3) fail to fully exploit the unlabeled samples. To address these issues, we propose Language-Assisted Debiasing and Smoothing framework, namely LADaS, for foundation model-based SSL. It consists of two components: 1) Language-assisted Pseudo-Label Debiasing (LPLD) to reduce biases in foundation models, and 2) Language-aware Pseudo-Label Smoothing (LPLS) to fully exploit low-confidence samples to facilitate SSL training. In particular, LPLD introduces a reliability score to dynamically assess the reliability of the logit adjustment. Additionally, it incorporates a language-oriented preference to reduce model biases using linguistic knowledge derived from pre-trained language models. Finally, LPLS introduces language-aware soft labels and devises language-aware pseudo-label smoothing loss to guide the learning of unlabeled samples with low-quality pseudo-labels. Extensive experiments demonstrate the superiority of our proposed LADaS.

With the rapid advancement of multimodal information retrieval, increasingly complex retrieval tasks have emerged. Existing methods predominately rely on task-specific fine-tuning of vision-language models, often those trained with image-text contrastive learning. In this paper, we explore the possibility of re-purposing generative Large Multimodal Models (LMMs) for retrieval. This approach enables unifying all retrieval tasks under the same formulation and, more importantly, allows for extrapolation towards unseen retrieval tasks without additional training. Our contributions can be summarised in the following aspects: (i) We introduce LamRA, a versatile framework designed to empower LMMs with sophisticated retrieval and reranking capabilities. (ii) For retrieval, we adopt a two-stage training strategy comprising language-only pre-training and multimodal instruction tuning to progressively enhance LMM's retrieval performance. (iii) For reranking, we employ joint training of both pointwise and listwise reranking, offering two distinct ways to further boost the retrieval performance. (iv) Extensive experiments underscore the efficacy of our method in handling more than ten retrieval tasks, demonstrating robust performance in both supervised and zero-shot settings, including scenarios involving previously unseen retrieval tasks.

This paper presents the Large Vision Diffusion Transformer (LaVin-DiT), a scalable and unified foundation model designed to tackle over 20 computer vision tasks in a generative framework. Unlike existing large vision models directly adapted from natural language processing architectures, which rely on less efficient autoregressive techniques and disrupt spatial relationships essential for vision data, LaVin-DiT introduces key innovations to optimize generative performance for vision tasks. First, to address the high dimensionality of visual data, we incorporate a spatial-temporal variational autoencoder that encodes data into a continuous latent space. Second, for generative modeling, we develop a joint diffusion transformer that progressively produces vision outputs. Third, for unified multi-task training, in-context learning is implemented. Input-target pairs serve as task context, which guides the diffusion transformer to align outputs with specific tasks within the latent space. During inference, a task-specific context set and test data as queries allow LaVin-DiT to generalize across tasks without fine-tuning. Trained on extensive vision datasets, the model is scaled from 0.1B to 3.4B parameters, demonstrating substantial scalability and state-of-the-art performance across diverse vision tasks. This work introduces a novel pathway for large vision foundation models, underscoring the promising potential of diffusion transformers. The code and models are available at https://derrickwang005.github.io/LaVin-DiT.

Research on 3D Vision-Language Models (3D-VLMs) is gaining increasing attention, which is crucial for developing embodied AI within 3D scenes, such as visual navigation and embodied question answering. Due to the high density of visual features, especially in large 3D scenes, accurately locating task-relevant visual information is challenging. Existing works attempt to segment all objectsand consider their features as scene representations. However, these task-agnostic object features include much redundant information and missing details for the task-relevant area. To tackle these problems, we propose LSceneLLM, an adaptive framework that automatically identifies task-relevant areas by leveraging LLM's visual preference for different tasks, followed by a plug-and-play scene magnifier module to capture fine-grained details in focused areas. Specifically, a dense token selector examines the attention map of LLM to identify visual preferences for the instruction input. It then magnifies fine-grained details of the focusing area. An adaptive self-attention module is leveraged to fuse the coarse-grained and selected fine-grained visual information. To comprehensively evaluate the large scene understanding ability of 3D-VLMs, we further introduce a cross-room understanding benchmark, XR-Scene, which contains a series of large scene understanding tasks including XR-QA, XR-EmbodiedPlanning, and XR-SceneCaption. Experiments show that our method surpasses existing methods on both large scene understanding and existing scene understanding benchmarks. Plunging our scene magnifier module into the existing 3D-VLMs also brings significant improvement.

We propose a training-free method for open-vocabulary semantic segmentation using Vision-and-Language Models (VLMs). Our approach enhances the initial per-patch predictions of VLMs through label propagation, which jointly optimizes predictions by incorporating patch-to-patch relationships. Since VLMs are primarily optimized for cross-modal alignment and not for intra-modal similarity, we use a Vision Model (VM) that is observed to better capture these relationships. We address resolution limitations inherent to patch-based encoders by applying label propagation at the pixel level as a refinement step, significantly improving segmentation accuracy near class boundaries. Our method, called LPOSS+, performs inference over the entire image, avoiding window-based processing and thereby capturing contextual interactions across the full image. LPOSS+ achieves state-of-the-art performance among training-free methods, across a diverse set of datasets. Code: https://github.com/vladan-stojnic/LPOSS

Compositional zero-shot learning (CZSL) aims to recognize unseen attribute-object compositions by learning the primitive concepts (*i.e.*, attribute and object) from the training set. While recent works achieve impressive results in CZSL by leveraging large vision-language models like CLIP, they ignore the rich semantic relationships between primitive concepts and their compositions. In this work, we propose LOGICZSL, a novel logic-induced learning framework to explicitly model the semantic relationships. Our logic-induced learning framework formulates the relational knowledge constructed from large language models as a set of logic rules, and grounds them onto the training data. Our logic-induced losses are complementary to the widely used CZSL losses, therefore can be employed to inject the semantic information into any existing CZSL methods. Extensive experimental results show that our method brings significant performance improvements across diverse datasets (*i.e.*, CGQA, UT-Zappos50K, MIT-States) with strong CLIP-based methods and settings (*i.e.*, Close World, Open World).

Interpretability and consistency have long been crucial factors in MRI reconstruction. While interpretability has been significantly innovated with the emerging deep unfolding networks, current solutions still suffer from inconsistency issues and produce inferior anatomical structures. Especially in out-of-distribution cases, e.g., when the acceleration rate (AR) varies, the generalization performance is often catastrophic. To counteract the dilemma, we propose an innovative Linear Mamba Operator (LMO) to ensure consistency and generalization, while still enjoying desirable interpretability. Theoretically, we argue that mapping between function spaces, rather than between signal instances, provides a solid foundation of high generalization. Technically, LMO achieves a good balance between global integration facilitated by a state space model that scans the whole function domain, and local integration engaged with an appealing property of continuous-discrete equivalence. On that basis, learning holistic features can be guaranteed, tapping the potential of maximizing data consistency. Quantitative and qualitative results demonstrate that LMO significantly outperforms other state-of-the-arts. More importantly, LMO is the unique model that, with AR changed, achieves retraining performance without retraining steps. Codes are available at https://github.com/ZhengJianwei2/LMO.

Recent advancements in multimodal large language models (MLLMs) have shown promising results, yet existing approaches struggle to effectively handle both temporal and spatial localization simultaneously. This challenge stems from two key issues: first, incorporating spatial-temporal localization introduces a vast number of coordinate combinations, complicating the alignment of linguistic and visual coordinate representations; second, encoding fine-grained temporal and spatial information during video feature compression is inherently difficult. To address these issues, we propose LLaVA-ST, a MLLM for fine-grained spatial-temporal multimodal understanding. In LLaVA-ST, we propose Language-Aligned Positional Embedding, which embeds the textual coordinate special token into the visual space, simplifying the alignment of fine-grained spatial-temporal correspondences. Additionally, we design the Spatial-Temporal Packer, which decouples the feature compression of temporal and spatial resolutions into two distinct point-to-region attention processing streams. Furthermore, we propose ST-Align dataset with 4.3M training samples for fine-grained spatial-temporal multimodal understanding. With ST-align, we present a progressive training pipeline that aligns the visual and textual feature through sequential coarse-to-fine stages.Additionally, we introduce an ST-Align benchmark to evaluate spatial-temporal interleaved fine-grained understanding tasks, which include Spatial-Temporal Video Grounding (STVG) , Event Localization and Captioning (ELC) and Spatial Video Grounding (SVG). LLaVA-ST achieves outstanding performance on 11 benchmarks requiring fine-grained temporal, spatial, or spatial-temporal interleaving multimodal understanding.

We introduce LLaVA-Critic, the first open-source large multimodal model (LMM) designed as a generalist evaluator to assess performance across a wide range of multimodal tasks. LLaVA-Critic is trained using a high-quality critic instruction-following dataset that incorporates diverse evaluation criteria and scenarios. Our experiments demonstrate the model's effectiveness in two key areas: (i) LMM-as-a-Judge, where LLaVA-Critic provides reliable evaluation scores, performing on par with or surpassing GPT models on multiple evaluation benchmarks; and (ii) Preference Learning, where it generates reward signals for preference learning, enhancing model alignment capabilities. This work underscores the potential of open-source LMMs in self-critique and evaluation, setting the stage for future research into scalable, superhuman alignment feedback mechanisms for LMMs.

Generating natural listener responses in conversational scenarios is crucial for creating engaging digital humans and avatars. Recent work has shown that large language models (LLMs) can be effectively leveraged for this task, demonstrating remarkable capabilities in generating contextually appropriate listener behaviors. However, current LLM-based methods face two critical limitations: they rely solely on speech content, overlooking other crucial communication signals, and they entangle listener identity with response generation, compromising output fidelity and generalization. In this work, we present a novel framework that addresses these limitations while maintaining the advantages of LLMs. Our approach introduces a Multimodal-LM architecture that jointly processes speech content, acoustics, and speaker emotion, capturing the full spectrum of communication cues. Additionally, we propose an identity disentanglement strategy using instance normalization and adaptive instance normalization in a VQ-VAE framework, enabling high-fidelity listening head synthesis with flexible identity control. Extensive experiments demonstrate that our method significantly outperforms existing approaches in terms of response naturalness and fidelity, while enabling effective identity control without retraining.

Current Large Language Vision Models (LLVMs) trained on web videos perform well in general video understanding but struggle with fine-grained details, complex human-object interactions (HOI), and view-invariant representation learning essential for Activities of Daily Living (ADL). This limitation stems from a lack of specialized ADL video instruction-tuning datasets and insufficient modality integration to capture discriminative action representations. To address this, we propose a semi-automated framework for curating ADL datasets, creating ADL-X, a multi-view, multi-modal RGBS instruction-tuning dataset. Additionally, we introduce LLAVIDAL, an LLVM integrating videos, 3D skeletons, and HOIs to model ADL's complex spatiotemporal relationships. For training LLAVIDAL, a simple joint alignment of all modalities yields suboptimal results; thus, we propose a Multimodal Progressive (MMPro) training strategy, incorporating modalities in stages following a curriculum. We also establish ADL MCQ and video description benchmarks to assess LLVM performance in ADL tasks. Trained on ADL-X, LLAVIDAL achieves state-of-the-art performance across ADL benchmarks. Code and data will be made publicly available at https://adl-x.github.io/.

First-person video assistants are highly anticipated to enhance our daily life through online video dialogue. However, existing online video assistants often sacrifice assistant efficacy for real-time efficiency by processing low-frame-rate videos with coarse-grained visual features. To overcome the trade-off between efficacy and efficiency, we propose "**F**ast & **S**low Video-Language Thinker" as on**LI**ne vide**O** assista**N**t, **LION-FS**, achieving real-time, proactive, temporally accurate, and contextually precise responses. LION-FS adopts a two-stage optimization strategy: **1) Fast Path: Routing-Based Response Determination** evaluates frame-by-frame whether a immediate response is necessary. To enhance responses determination accuracy and handle higher frame-rate inputs efficiently, we employ Token Aggregation Routing to dynamically fuse spatiotemporal features without increasing token numbers, while utilizing Token Dropping Routing to eliminate redundant features, and **2) Slow Path: Multi-granularity Keyframe Augmentation** optimizes keyframes during response generation. To provide comprehensive and detailed responses beyond atomic actions constrained by training data, fine-grained spatial features and human-environment interaction features are extracted through multi-granular pooling. They are further integrated into a meticulously designed multimodal Thinking Template to guide more precise response generation. Comprehensive evaluations on online video tasks demonstrate that LION-FS achieves state-of-the-art efficacy and efficiency. The codes will be released soon.

Reconstructing dynamic assets from video data is central to many in computer vision and graphics tasks. Existing 4D reconstruction approaches are limited by category-specific models or slow optimization-based methods. Inspired by the recent Large Reconstruction Model (LRM), we present the Large Interpolation Model (LIM), a transformer-based feed-forward solution, guided by a novel causal consistency loss, for interpolating implicit 3D representations across time. Given implicit 3D representations at times t_0 and t_1, LIM produces a deformed shape at any continuous time t\in[t_0,t_1] delivering high-quality interpolations in seconds (per frame).Furthermore, LIM allows explicit mesh tracking across time, producing a consistently uv-textured mesh sequence ready for integration into existing production pipelines. We also use LIM, in conjunction with a diffusion-based multiview generator, to produce dynamic 4D reconstructions from monocular videos. We evaluate LIM on various dynamic datasets, benchmarking against image-space interpolation methods (e.g., FiLM) and direct triplane linear interpolation, and demonstrate clear advantages. In summary, LIM is the first feed-forward model capable of high-speed tracked 4D asset reconstruction across diverse categories.

In this paper, we propose LC-Mamba, a Mamba-based model that captures fine-grained spatiotemporal information in video frames, addressing limitations in current interpolation methods and enhancing performance. The main contributions are as follows: First, we apply a shifted local window technique to reduce historical decay and enhance local spatial features, allowing multiscale capture of detailed motion between frames. Second, we introduce a Hilbert curve-based selective state scan to maintain continuity across window boundaries, preserving spatial correlations both within and between windows. Third, we extend the Hilbert curve to enable voxel-level scanning to effectively capture spatiotemporal characteristics between frames. The proposed LC-Mamba achieves competitive results, with a PSNR of 36.53 dB on Vimeo-90k, outperforming prior models by +0.03 dB. The code and models are publicly available at https://github.com/Miinuuu/LC-Mamba.git

Training-free Neural Architecture Search (NAS) efficiently identifies high-performing neural networks using zero-cost (ZC) proxies. Unlike multi-shot and one-shot NAS approaches, ZC-NAS is both (i) time-efficient, eliminating the need for model training, and (ii) interpretable, with proxy designs often theoretically grounded. Despite rapid developments in the field, current SOTA ZC proxies are typically constrained to well-established convolutional search spaces. With the rise of Large Language Models shaping the future of deep learning, this work extends ZC proxy applicability to Vision Transformers (ViTs). We present a new benchmark using the Autoformer search space evaluated on 6 distinct tasks, and propose Layer-Sample Wise Activation with Gradients information (L-SWAG), a novel, generalizable metric that characterises both convolutional and transformer architectures across 14 tasks. Additionally, previous works highlighted how different proxies contain complementary information, motivating the need for a ML model to identify useful combinations. To further enhance ZC-NAS, we therefore introduce LIBRA-NAS (Low Information gain and Bias Re-Alignment), a method that strategically combines proxies to best represent a specific benchmark. Integrated into the NAS search, LIBRA-NAS outperforms evolution and gradient-based NAS techniques by identifying an architecture with a 17.0% test error on ImageNet1k in just 0.1 GPU days.

Visual Emotion Recognition (VER) is a critical yet challenging task aimed at inferring emotional states of individuals based on visual cues. However, existing works focus on single domains, e.g., realistic images or stickers, limiting VER models' cross-domain generalizability. To fill this gap, we introduce an Unsupervised Cross-Domain Visual Emotion Recognition (UCDVER) task, which aims to generalize visual emotion recognition from the source domain (e.g., realistic images) to the low-resource target domain (e.g., stickers) in an unsupervised manner. Compared to the conventional unsupervised domain adaptation problems, UCDVER presents two key challenges: a significant emotional expression variability and an affective distribution shift. To mitigate these issues, we propose the Knowledge-aligned Counterfactual-enhancement Diffusion Perception (KCDP) framework. Specifically, KCDP leverages a VLM to align emotional representations in a shared knowledge space and guides diffusion models for improved visual affective perception. Furthermore, a Counterfactual-Enhanced Language-image Emotional Alignment (CLIEA) method generates high-quality pseudo-labels for the target domain. Extensive experiments demonstrate that our model surpasses SOTA models in both perceptibility and generalization, e.g., gaining 12% improvements over SOTA VER model TGCA-PVT.

Video inpainting, which aims to fill missing regions with visually coherent content, has emerged as a crucial technique for creative applications such as editing. While existing approaches achieve visual consistency or text-guided generation, they often struggle to balance coherence and creative diversity. In this work, we introduce VideoRepainter, a two-stage framework that allows users to inpaint a keyframe using established image-level techniques, then propagate the changes to other frames. Our approach can leverage state-of-the-art image models for keyframe manipulation, thereby easing the burden of the video-inpainting process. To this end, we integrate an image-to-video model with a symmetric condition mechanism to address ambiguity caused by direct mask downsampling. We further explore efficient strategies for mask synthesis and parameter tuning to reduce costs in data processing and model training. Evaluations demonstrate our method achieves superior results in both visual fidelity and content diversity compared to existing approaches, providing a practical solution for creative video manipulation.

The rapid advancement of visual generative models necessitates efficient and reliable evaluation methods. Arena platform, which gathers user votes on model comparisons, can rank models with human preferences. However, traditional Arena methods, while established, require an excessive number of comparisons for ranking to converge and are vulnerable to preference noise in voting, suggesting the need for better approaches tailored to contemporary evaluation challenges. In this paper, we introduce K-Sort Arena, an efficient and reliable platform based on a key insight: images and videos possess higher perceptual intuitiveness than texts, enabling rapid evaluation of multiple samples simultaneously. Consequently, K-Sort Arena employs K-wise comparisons, allowing K models to engage in free-for-all competitions, which yield much richer information than pairwise comparisons. To enhance the robustness of the system, we leverage probabilistic modeling and Bayesian updating techniques. We propose an exploration-exploitation-based matchmaking strategy to facilitate more informative comparisons. In our experiments, K-Sort Arena exhibits 16.3xfaster convergence compared to the widely used ELO algorithm. To further validate the superiority and obtain a comprehensive leaderboard, we collect human feedback via crowdsourced evaluations of numerous cutting-edge text-to-image and text-to-video models. Thanks to its high efficiency, K-Sort Arena can continuously incorporate emerging models and update the leaderboard with minimal votes. Our project has undergone several months of internal testing and is now available online.

Weakly-supervised methods for video anomaly detection (VAD) are conventionally based merely on RGB spatio-temporal features, which continues to limit their reliability in real-world scenarios. This is due to the fact that RGB-features are not sufficiently distinctive in setting apart categories such as shoplifting from visually similar events. Therefore, towards robust complex real-world VAD, it is essential to augment RGB spatio-temporal features by additional modalities. Motivated by this, we introduce the Poly-modal Induced framework for VAD: PI-VAD (or \pi-VAD), a novel approach that augments RGB representations by five additional modalities. Specifically, the modalities include sensitivity to fine-grained motion (Pose), three dimensional scene and entity representation (Depth), surrounding objects (Panoptic masks), global motion (optical flow), as well as language cues (VLM). Each modality represents an axis of a polygon, streamlined to add salient cues to RGB. \pi-VAD includes two plug-in modules, namely Pseudo-modality Generation module and Cross Modal Induction module, which generate modality-specific prototypical representation and, thereby, induce multi-modal information into RGB cues. These modules operate by performing anomaly-aware auxiliary tasks and necessitate five modality backbones -- only during training. Notably, \pi-VAD achieves state-of-the-art accuracy on three prominent VAD datasets encompassing real-world scenarios, without requiring the computational overhead of five modality backbones at inference.