CVPR 2025 论文列表

Fine-grained open-set semi-supervised learning (OSSL) investigates a practical scenario where unlabeled data may contain fine-grained out-of-distribution (OOD) samples. Due to the subtle visual differences among in-distribution (ID) samples, as well as between ID and OOD samples, it is extremely challenging to separate the ID and OOD samples. Due to the subtle visual differences among in-distribution (ID) and OOD samples. Recent Vision-Language Models, such as CLIP, have shown excellent generalization capabilities. However, it tends to focus on general attributes, and thus is insufficient to distinguish the fine-grained details. To tackle the issues, in this paper, we propose a novel CLIP-driven coarse-to-fine semantic-guided framework, named CFSG-CLIP, to progressively focus on the distinctive fine-grained clues. Specifically, CFSG-CLIP comprises a coarse-guidance branch and a fine-guidance branch derived from the pre-trained CLIP model. In the coarse-guidance branch, we design a semantic filtering module to initially filter and highlight local visual features guided by cross-modality features. Then, in the fine-guidance branch, we further design a visual-semantic injection strategy, which embeds category-related visual cues into the visual encoder to further refine the local visual features. By the designed dual-guidance framework, local subtle cues are progressively discovered to distinct the subtle difference between ID and OOD samples. Extensive experiments demonstrate that CFSG-CLIP achieves competitive performance on multiple fine-grained datasets. The source code is available at https://github.com/LxxxxK/CFSG-CLIP.

Despite its prevalent use in image-text matching tasks in a zero-shot manner, CLIP has been shown to be highly vulnerable to adversarial perturbations added onto images. Recent studies propose to finetune the vision encoder of CLIP with adversarial samples generated on the fly, and show improved robustness against adversarial attacks on a spectrum of downstream datasets, a property termed as zero-shot robustness. In this paper, we show that malicious perturbations that seek to maximise the classification loss lead to `falsely stable' images, and propose to leverage the pre-trained vision encoder of CLIP to counterattack such adversarial images during inference to achieve robustness. Our paradigm is simple and training-free, providing the first method to defend CLIP from adversarial attacks at test time, which is orthogonal to existing methods aiming to boost zero-shot adversarial robustness of CLIP. We conduct experiments across 16 classification datasets, and demonstrate stable and consistent gains compared to test-time defence methods adapted from existing adversarial robustness studies that do not rely on external networks, without noticeably impairing performance on clean images. We also show that our paradigm can be employed on CLIP models that have been adversarially finetuned to further enhance their robustness at test time. Our code will be released.

Scene Text Retrieval (STR) seeks to identify all images containing a given query string. Existing methods typically rely on an explicit Optical Character Recognition (OCR) process of text spotting or localization, which is susceptible to complex pipelines and accumulated errors. To settle this, we resort to the Contrastive Language-Image Pre-training (CLIP) models, which have demonstrated the capacity to perceive and understand scene text, making it possible to achieve strictly OCR-free STR. From the perspective of parameter-efficient transfer learning, a lightweight visual position adapter is proposed to provide a positional information complement for the visual encoder. Besides, we introduce a visual context dropout technique to improve the alignment of local visual features. A novel, parameter-free cross-attention mechanism transfers the contrastive relationship between images and text to that between visual tokens and text, producing a rich cross-modal representation, which can be utilized for efficient reranking with a linear classifier. The resulting model, CAYN, which proves that CLIP is Almost all You Need for STR with no more than 0.50M additional parameters required, achieves new state-of-the-art performance on the STR task, with 92.46%/89.49%/85.98% mAP on the SVT/IIIT-STR/TTR datasets. Our findings demonstrate that CLIP can serve as a reliable and efficient solution for OCR-free STR.

Contrastive Language-Image Pre-training (CLIP) models excel in zero-shot classification, yet face challenges in complex multi-object scenarios. This study offers a comprehensive analysis of CLIP's limitations in these contexts using a specialized dataset, ComCO, designed to evaluate CLIP's encoders in diverse multi-object scenarios. Our findings reveal significant biases: the text encoder prioritizes first-mentioned objects, and the image encoder favors larger objects. Through retrieval and classification tasks, we quantify these biases across multiple CLIP variants and trace their origins to CLIP's training process, supported by analyses of the LAION dataset and training progression. Our image-text matching experiments show substantial performance drops when object size or token order changes, underscoring CLIP's instability with rephrased but semantically similar captions. Extending this to longer captions and text-to-image models like Stable Diffusion, we demonstrate how prompt order influences object prominence in generated images.

Multimodal large language models (MLLMs) have garnered widespread attention from researchers due to their remarkable understanding and generation capabilities in visual language tasks (e.g., visual question answering). However, the rapid pace of knowledge updates in the real world makes offline training of MLLMs costly, and when faced with non-stationary data streams, MLLMs suffer from catastrophic forgetting during learning. In this paper, we propose an MLLMs-based dual momentum Mixture-of-Experts (\texttt CL-MoE ) framework for continual visual question answering. We integrate MLLMs with continual learning to utilize the rich commonsense knowledge in LLMs.We introduce a Dual-Router MoE (RMoE) to select the global and local experts using task-level and instance-level routers, to robustly assign weights to the experts most appropriate for the task. Then, we design a dynamic Momentum MoE (MMoE) to update the parameters of experts dynamically based on the relationships between the experts and tasks, so that the model can absorb new knowledge while maintaining existing knowledge. The extensive experimental results indicate that our method achieves state-of-the-art performance on 10 VQA tasks, proving the effectiveness of our approach. The codes and weights will be released on GitHub.

Class-Incremental Learning (CIL) aims to learn new classes sequentially while retaining the knowledge of previously learned classes. Recently, pre-trained models (PTMs) combined with parameter-efficient fine-tuning (PEFT) have shown remarkable performance in rehearsal-free CIL without requiring exemplars from previous tasks. However, existing adapter-based methods, which incorporate lightweight learnable modules into PTMs for CIL, create new adapters for each new task, leading to both parameter redundancy and failure to leverage shared knowledge across tasks. In this work, we propose ContinuaL Low-Rank Adaptation (CL-LoRA), which introduces a novel dual-adapter architecture combining task-shared adapters to learn cross-task knowledge and task-specific adapters to capture unique features of each new task. Specifically, the shared adapters utilize random orthogonal matrices and leverage knowledge distillation with gradient reassignment to preserve essential shared knowledge. In addition, we introduce learnable block-wise weights for task-specific adapters, which mitigate inter-task interference while maintaining the model's plasticity. We demonstrate CL-LoRA consistently achieves promising performance under multiple benchmarks with reduced training and inference computation, establishing a more efficient and scalable paradigm for continual learning with pre-trained models.

Depth estimation is a fundamental task in 3D vision. An ideal depth estimation model is expected to embrace meticulous detail, temporal consistency, and high efficiency. Although existing foundation models can perform well in certain specific aspects, most of them fall short of fulfilling all the above requirements simultaneously. In this paper, we present CH_3Depth, an efficient and flexible model for depth estimation with flow matching to address this challenge. Specifically, 1) we reframe the optimization objective of flow matching as the Inversion by Direct Iteration (InDI) to improve accuracy. 2) To enhance efficiency, we propose non-uniform sampling to achieve better prediction with fewer sampling steps. 3) We design the Latent Temporal Stabilizer (LTS) to enhance temporal consistency by aggregating latent codes of adjacent frames, enabling our method to be lightweight and compatible for video depth estimation. CH_3Depth achieves state-of-the-art performance in zero-shot evaluations across multiple image and video datasets, excelling in prediction accuracy, efficiency, and temporal consistency, highlighting its potential as the next foundation model for depth estimation.

Semi-supervised learning (SSL) has garnered significant attention due to its ability to leverage limited labeled data and a large amount of unlabeled data to improve model generalization performance. Recent approaches achieve impressive successes by combining ideas from both consistency regularization and pseudo-labeling. However, these methods tend to underperform in the more realistic situations with relatively scarce labeled data. We argue that this issue arises because existing methods rely solely on the model's confidence, making them challenging to accurately assess the model's state and identify unlabeled examples contributing to the training phase when supervision information is limited, especially during the early stages of model training. In this paper, we propose a novel SSL model called CGMatch, which, for the first time, incorporates a new metric known as Count-Gap (CG). We demonstrate that CG is effective in discovering unlabeled examples beneficial for model training. Along with confidence, a commonly used metric in SSL, we propose a fine-grained dynamic selection (FDS) strategy. This strategy dynamically divides the unlabeled dataset into three subsets with different characteristics: easy-to-learn set, ambiguous set, and hard-to-learn set. By selective filtering subsets, and applying corresponding regularization with selected subsets, we mitigate the negative impact of incorrect pseudo-labels on model optimization and generalization. Extensive experimental results on several common SSL benchmarks indicate the effectiveness of CGMatch especially when the labeled data are particularly limited. Source code is available at https://github.com/BoCheng-96/CGMatch.

Diffusion Models (DMs) benefit from large and diverse datasets for their training. Since this data is often scraped from the Internet without permission from the data owners, this raises concerns about copyright and intellectual property protections. While (illicit) use of data is easily detected for training samples perfectly re-created by a DM at inference time, it is much harder for data owners to verify if their data was used for training when the outputs from the suspect DM are not close replicas. Conceptually, membership inference attacks (MIAs), which detect if a given data point was used during training, present themselves as a suitable tool to address this challenge. However, we demonstrate that existing MIAs are not strong enough to reliably determine the membership of individual images in large, state-of-the-art DMs. To overcome this limitation, we propose Copyrighted Data Identification (CDI), a framework for data owners to identify whether their dataset was used to train a given DM. CDI relies on dataset inference techniques, i.e., instead of using the membership signal from a single data point, CDI leverages the fact that most data owners, such as providers of stock photography, visual media companies, or even individual artists, own datasets with multiple publicly exposed data points which might all be included in the training of a given DM. By selectively aggregating signals from existing MIAs and using new handcrafted methods to extract features from these datasets, feeding them to a scoring model, and applying rigorous statistical testing, CDI allows data owners with as little as 70 data points to identify with a confidence of more than 99% whether their data was used to train a given DM. Thereby, CDI represents a valuable tool for data owners to claim illegitimate use of their copyrighted data.

Composed Image Retrieval (CIR) is a multi-modal task that seeks to retrieve target images by harmonizing a reference image with a modified instruction. A key challenge in CIR lies in compositional conflicts between the reference image (e.g., blue, long sleeve) and the modified instruction (e.g., grey, short sleeve). Previous works attempt to mitigate such conflicts through feature-level manipulation, commonly employing learnable masks to obscure conflicting features within the reference image. However, the inherent complexity of feature spaces poses significant challenges in precise conflict neutralization, thereby leading to uncontrollable results. To this end, this paper proposes the Compositional Conflict Identification and Neutralization (CCIN) framework, which sequentially identifies and neutralizes compositional conflicts for effective CIR. Specifically, CCIN comprises two core modules: 1) Compositional Conflict Identification module, which utilizes LLM-based analysis to identify specific conflicting attributes, and 2) Compositional Conflict Neutralization module, which first generates a kept instruction to preserve non-conflicting attributes, then neutralizes conflicts under collaborative guidance of both the kept and modified instructions. Extensive experiments demonstrate the superiority of CCIN over the state-of-the-arts. Code repository: https://github.com/LikaiTian/CCIN.

Recent advances in audio-visual learning have shown promising results in learning representations across modalities. However, most approaches rely on global audio representations that fail to capture fine-grained temporal correspondences with visual frames.Additionally, existing methods often struggle with conflicting optimization objectives when trying to jointly learn reconstruction and cross-modal alignment. In this work, we propose CAV-MAE Sync as a simple yet effective extension of the original CAV-MAE framework for self-supervised audio-visual learning. We address three key challenges: First, we tackle the granularity mismatch between modalities by treating audio as a temporal sequence aligned with video frames, rather than using global representations. Second, we resolve conflicting optimization goals by separating contrastive and reconstruction objectives through dedicated global tokens. Third, we improve spatial localization by introducing learnable register tokens that reduce semantic load on patch tokens. We evaluate the proposed approach on AudioSet, VGG Sound, and the ADE20K Sound dataset on zero-shot retrieval, classification and localization tasks demonstrating state-of-the-art performance and outperforming more complex architectures. Code available at https://github.com/edsonroteia/cav-mae-sync.

Transformer-based methods have demonstrated impressive performance in low-level visual tasks such as Image Super-Resolution (SR). However, its computational complexity grows quadratically with the spatial resolution. A series of works attempt to alleviate this problem by dividing Low-Resolution images into local windows, axial stripes, or dilated windows. SR typically leverages the redundancy of images for reconstruction, and this redundancy appears not only in local regions but also in long-range regions. However, these methods limit attention computation to content-agnostic local regions, limiting directly the ability of attention to capture long-range dependency. To address these issues, we propose a lightweight Content-Aware Token Aggregation Network (CATANet). Specifically, we propose an efficient Content-Aware Token Aggregation module for aggregate long-range content-similar tokens, which shares token centers across all image tokens and updates them only during the training phase. Then we utilize intra-group self-attention to enable long-range information interaction. Moreover, we design an inter-group cross-attention to further enhance global information interaction. The experimental results show that, compared with the state-of-the-art cluster-based method SPIN, our method achieves superior performance, with a maximum PSNR improvement of 0.33dB and nearly double the inference speed.

We present CAT4D, a method for creating 4D (dynamic 3D) scenes from monocular video. CAT4D leverages a multi-view video diffusion model trained on a diverse combination of datasets to enable novel view synthesis at any specified camera poses and timestamps. Combined with a novel sampling approach, this model can transform a single monocular video into a multi-view video, enabling robust 4D reconstruction via optimization of a deformable 3D Gaussian representation. We demonstrate competitive performance on novel view synthesis and dynamic scene reconstruction benchmarks, and highlight the creative capabilities for 4D scene generation from real or generated videos. See our project page for results and interactive demos: cat-4d.github.io.

Omnidirectional videos (ODVs) present distinct challenges for accurate audio-visual saliency prediction due to their immersive nature, which combines spatial audio with panoramic visuals to enhance the user experience. While auditory cues are crucial for guiding visual attention across the panoramic scene, the interaction between audio and visual stimuli in ODVs remains underexplored. Existing models primarily focus on spatiotemporal visual cues and treat audio signals separately from their spatial and temporal contexts, often leading to misalignments between audio and visual content and undermining temporal consistency across frames. To bridge these gaps, we propose a novel audio-induced saliency prediction model for ODVs that holistically integrates audio and visual inputs through a multi-modal encoder, an audio-visual interaction module, and an audio-visual transformer. Unlike conventional methods that isolate audio cue locations and attributes, our model employs a query-based framework, where learnable audio queries capture comprehensive audio-visual dependencies, thus enhancing saliency prediction by dynamically aligning with audio cues. Besides, we introduce a novel consistency loss to enforce temporal coherence in saliency regions across frames. Extensive experiments demonstrate that our model outperforms state-of-the-art methods in predicting audio-visual salient regions in ODVs, establishing its robustness and superior performance.

In this work, we propose an extreme compression technique for Large Multimodal Models (LMMs). While previous studies have explored quantization as an efficient post-training compression method for Large Language Models (LLMs), low-bit compression for multimodal models remains under-explored. The redundant nature of inputs in multimodal models results in a highly sparse attention matrix. We theoretically and experimentally demonstrate that the attention matrix's sparsity bounds the compression error of the Query and Key weight matrices. Based on this, we introduce CASP, a model compression technique for LMMs. Our approach performs a data-aware low-rank decomposition on the Query and Key weight matrix, followed by quantization across all layers based on an optimal bit allocation process. CASP is compatible with any quantization technique and enhances state-of-the-art 2-bit quantization methods (AQLM and QuIP#) by an average of 21% on image- and video-language benchmarks. The code is provided in the supplementary materials.

We present a novel approach for indoor scene synthesis, which learns to arrange decomposed cuboid primitives to represent 3D objects within a scene. Unlike conventional methods that use bounding boxes to determine the placement and scale of 3D objects, our approach leverages cuboids as a straightforward yet highly effective alternative for modeling objects. This allows for compact scene generation while minimizing object intersections. Our approach, coined CASAGPT for Cuboid Arrangement and Scene Assembly, employs an autoregressive model to sequentially arrange cuboids, producing physically plausible scenes. By applying rejection sampling during the fine-tuning stage to filter out scenes with object collisions, our model further reduces intersections and enhances scene quality. Additionally, we introduce a refined dataset, 3DFRONT-NC, which eliminates significant noise presented in the original dataset, 3D-FRONT. Extensive experiments on the 3D-FRONT dataset as well as our dataset demonstrate that our approach consistently outperforms the state-of-the-art methods, enhancing the realism of generated scenes, and providing a promising direction for 3D scene synthesis.

Image registration estimates spatial correspondences between image pairs. These estimates are typically obtained via numerical optimization or regression by a deep network. A desirable property is that a correspondence estimate (e.g., the true oracle correspondence) for an image pair is maintained under deformations of the input images. Formally, the estimator should be equivariant to a desired class of image transformations. In this work, we present careful analyses of equivariance properties in the context of multi-step deep registration networks. Based on these analyses we 1) introduce the notions of [U,U] equivariance (network equivariance to the same deformations of the input images) and [W,U] equivariance (where input images can undergo different deformations); we 2) show that in a suitable multi-step registration setup it is sufficient for overall [W,U] equivariance if the first step has [W,U] equivariance and all others have [U,U] equivariance; we 3) show that common displacement-predicting networks only exhibit [U,U] equivariance to translations instead of the more powerful [W,U] equivariance; and we 4) show how to achieve multi-step [W,U] equivariance via a coordinate-attention mechanism combined with displacement-predicting networks. Our approach obtains excellent practical performance for 3D abdomen, lung, and brain medical image registration. We match or outperform state-of-the-art (SOTA) registration approaches on all the datasets with a particularly strong performance for the challening abdomen registration.

Recently, large efforts have been made to design efficient linear-complexity visual Transformers. However, current linear attention models are generally unsuitable to be deployed in resource-constrained mobile devices, due to suffering from either few efficiency gains or significant accuracy drops. In this paper, we propose a new deCoupled duAl-interactive lineaR attEntion (CARE) mechanism, revealing that features' decoupling and interaction can fully unleash the power of linear attention. We first propose an asymmetrical feature decoupling strategy that asymmetrically decouples the learning process for local inductive bias and long-range dependencies, thereby preserving sufficient local and global information while effectively enhancing the efficiency of models. Then, a dynamic memory unit is employed to maintain critical information along the network pipeline. Moreover, we design a dual interaction module to effectively facilitate interaction between local inductive bias and long-range information as well as among features at different layers. By adopting a decoupled learning way and fully exploiting complementarity across features, our method can achieve both high efficiency and accuracy. Extensive experiments on ImageNet-1K, COCO, and ADE20K datasets demonstrate the effectiveness of our approach, e.g., achieving 78.4/82.1% top-1 accuracy on ImagegNet-1K at the cost of only 0.7/1.9 GMACs. Codes will be released on github.

Reconstructing photorealistic and dynamic portrait avatars from images is essential to many applications including advertising, visual effects, and virtual reality. Depending on the application, avatar reconstruction involves different capture setups and constraints -- for example, visual effects studios use camera arrays to capture hundreds of reference images, while content creators may seek to animate a single portrait image downloaded from the internet. As such, there is a large and heterogeneous ecosystem of methods for avatar reconstruction. Techniques based on multi-view stereo or neural rendering achieve the highest quality results, but require hundreds of reference images. Recent generative models produce convincing avatars from a single reference image, but visual fidelity lags behind multi-view techniques. Here, we present CAP4D: an approach that uses a morphable multi-view diffusion model to reconstruct photoreal 4D (dynamic 3D) portrait avatars from any number of reference images (i.e., one to 100) and animate and render them in real time. Our approach demonstrates state-of-the-art performance for single-, few-, and multi-image 4D portrait avatar reconstruction, and takes steps to bridge the gap in visual fidelity between single-image and multi-view reconstruction techniques.