CVPR 2025 Papers

In ordinal classification, misclassifying neighboring ranks is common, yet the consequences of these errors are not the same.For example, misclassifying benign tumor categories is less consequential, compared to an error at the pre-cancerous to cancerous threshold, which could profoundly influence treatment choices. Despite this, existing ordinal classification methods do not account for the varying importance of these margins, treating all neighboring classes as equally significant. To address this limitation, we propose CLOC, a new margin-based contrastive learning method for ordinal classification that learns an ordered representation based on the optimization of multiple margins with a novel multi-margin n-pair loss (MMNP).CLOC enables flexible decision boundaries across key adjacent categories, facilitating smooth transitions between classes and reducing the risk of overfitting to biases present in the training data.We provide empirical discussion regarding the properties of MMNP and show experimental results on five real-world image datasets (Adience, Historical Colour Image Dating, Knee Osteoarthritis, Indian Diabetic Retinopathy Image, and Breast Carcinoma Subtyping) and one synthetic dataset simulating clinical decision bias.Our results demonstrate that CLOC outperforms existing ordinal classification methods and show the interpretability and controllability of CLOC in learning meaningful, ordered representations that align with clinical and practical needs.

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.

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.

Multi-modal large language models (MLLMs) extend large language models (LLMs) to process multi-modal information, enabling them to generate responses to image-text inputs. MLLMs have been incorporated into diverse multi-modal applications, such as autonomous driving and medical diagnosis, via plug-and-play without fine-tuning. This deployment paradigm increases the vulnerability of MLLMs to backdoor attacks. However, existing backdoor attacks against MLLMs achieve limited effectiveness and stealthiness. In this work, we propose BadToken, the first token-level backdoor attack to MLLMs. BadToken introduces two novel backdoor behaviors: Token-substitution and Token-addition, which enable flexible and stealthy attacks by making token-level modifications to the original output for backdoored inputs. We formulate a general optimization problem that considers the two backdoor behaviors to maximize the attack effectiveness. We evaluate BadToken on two open-source MLLMs and various tasks. Our results show that our attack maintains the model's utility while achieving high attack success rates and stealthiness. We also show the real-world threats of BadToken in two scenarios, i.e., autonomous driving and medical diagnosis. Furthermore, we consider defenses including fine-tuning and input purification. Our results highlight the threat of our attack.

The success of existing deep multi-view graph clustering methods is based on the assumption that node attributes are fully available across all views. However, in practical scenarios, node attributes are frequently missing due to factors such as data privacy concerns or failures in data collection devices. Although some methods have been proposed to address the issue of missing node attributes, they come with the following limitations: i) Existing methods are often not tailored specifically for clustering tasks and struggle to address missing attributes effectively. ii) They tend to ignore the relational dependencies between nodes and their neighboring nodes. This oversight results in unreliable imputations, thereby degrading clustering performance. To address the above issues, we propose an Attribute-Missing Multi-view Graph Clustering (AMMGC). Specifically, we first impute missing node attributes by leveraging neighborhood information through an adjacency matrix. Then, to improve the consistency, we integrate a dual structure consistency module that aligns graph structures across multiple views, reducing redundancy and retaining key information. Furthermore, we introduce a high-confidence guidance module to improve the reliability of clustering. Extensive experiment results showcase the effectiveness and superiority of our proposed method on multiple benchmark datasets.

The capability of tracking objects in low-light environments like nighttime is crucial for numerous real-world applications such as crowd behavior analysis and traffic scene understanding. However, previous Multi-Camera Multi-Target(MCMT) tracking methods are primarily focused on tracking during daytime with favorable lighting, shying away from low-light environments. The main difficulty of tracking under low light condition is lack of detailed visible appearance features. To address this issue, we incorporate the infrared modality into MCMT tracking framework to provide more useful information. We constructed the first Multi-modality(RGBT) Multi-camera Multi-target tracking dataset named M3Track, which contains sequences captured in low-light environments, laying a solid foundation for all-day multi-camera tracking. Based on the proposed dataset, we propose All-Day Multi-Camera Multi-Target tracking network, termed as ADMCMT. Specifically, we propose an All-Day Mamba Fusion(ADMF) model to adaptively fuse information from different modalities. Within ADMF, the Lighting Guidance Model(IGM) extracts lighting relevant information to guide the fusion process. Furthermore, the Nearby Target Collection(NTC) strategy is designed to enhance tracking accuracy by leveraging information derived from surrounding objects of target. Experiments conducted on M3Track demonstrate that ADMCMT exhibits strong generalization across different lighting conditions. The code will be released soon at https://github.com/QTRACKY/ADMCMT.

Incorporating transformer models into edge devices poses a significant challenge due to the computational demands of adapting these large models across diverse applications. Parameter-efficient tuning (PET) methods (e.g. LoRA, Adapter, Visual Prompt Tuning, etc.) allow for targeted adaptation by modifying only small parts of the transformer model. However, adapting to dynamic unlabeled target distributions at the test time remains complex. To address this, we introduce AdMiT: Adaptive Multi -Source Tuning in Dynamic Environments. AdMiT innovates by pre-training a set of PET modules, each optimized for different source distributions or tasks, and dynamically selecting and integrating a sparse subset of relevant modules when encountering a new, few-shot, unlabeled target distribution. This integration leverages Kernel Mean Embedding (KME)-based matching to align the target distribution with relevant source knowledge efficiently, without requiring additional routing networks or hyperparameter tuning. AdMiT achieves adaptation with a single inference step, making it particularly suitable for resource-constrained edge deployments. Furthermore, AdMiT preserves privacy by performing an adaptation locally on each edge device, without the need for data exchange. Our theoretical analysis establishes guarantees for AdMiT's generalization, while extensive benchmarks demonstrate that AdMiT consistently outperforms other PET methods across a range of tasks, achieving robust and efficient adaptation.

Image inpainting aims to fill the missing region of an image.Recently, there has been a surge of interest in foreground-conditioned background inpainting, a sub-task that fills the background of an image while the foreground subject and associated text prompt are provided.Existing background inpainting methods typically strictly preserve the subject's original position from the source image,resulting in inconsistencies between the subject and the generated background.To address this challenge, we propose a new task, the "Text-Guided Subject-Position Variable Background Inpainting", which aims to dynamically adjust the subject position to achieve a harmonious relationship between the subject andthe inpainted background, and propose the Adaptive Transformation Agent (A^\text T A) for this task.Firstly, we design a PosAgent Block that adaptively predicts an appropriate displacement based on given features to achieve variable subject-position.Secondly, we design the Reverse Displacement Transform (RDT) module, which arranges multiple PosAgent blocks in a reverse structure, to transform hierarchical feature maps from deep to shallow based on semantic information.Thirdly, we equip A^\text T A with a Position Switch Embedding to control whether the subject's position in the generated image is adaptively predicted or fixed.Extensive comparative experiments validate the effectiveness of our A^\text T A approach, which not only demonstrates superior inpainting capabilities in subject-position variable inpainting, but also ensures good performance on subject-position fixed inpainting.

Recent graph-based multi-view clustering (GMVC) methods typically encode view features into high-dimensional spaces and construct graphs based on distance similarity. However, the high dimensionality of the embeddings often leads to the hubness problem, where a few points repeatedly appear in the nearest neighbor lists of other points. We show that this negatively impacts the extracted graph structures and message passing, thus degrading clustering performance. To the best of our knowledge, we are the first to highlight the detrimental effect of hubness in GMVC methods and introduce the hubREP (hub-aware Representation Embedding and Pairing) framework. Specifically, we propose a simple yet effective encoder that reduces hubness while preserving neighborhood topology within each view. Additionally, we propose a hub-aware pairing module to maintain structure consistency across views, efficiently enhancing the view-specific representations. The proposed hubREP is lightweight compared to the conventional autoencoders used in state-of-the-art GMVC methods and can be integrated into existing GMVC methods that mostly focus on novel fusion mechanisms, further boosting their performance. Comprehensive experiments performed on eight benchmarks confirm the superiority of our method. The code is available at https://github.com/zmxu196/hubREP.

Despite advances in neural rendering, due to the scarcity of high-quality 3D datasets and the inherent limitations of multi-view diffusion models, view synthesis and 3D model generation are restricted to low resolutions with suboptimal multi-view consistency. In this study, we present a novel 3D enhancement pipeline, dubbed 3DEnhancer, which employs a multi-view latent diffusion model to enhance coarse 3D inputs while preserving multi-view consistency. Our method includes a pose-aware encoder and a diffusion-based denoiser to refine low-quality multi-view images, along with data augmentation and a multi-view attention module with epipolar aggregation to maintain consistent, high-quality 3D outputs across views. Unlike existing video-based approaches, our model supports seamless multi-view enhancement with improved coherence across diverse viewing angles. Extensive evaluations show that 3DEnhancer significantly outperforms existing methods, boosting both multi-view enhancement and per-instance 3D optimization tasks.

Current 3D Large Multimodal Models (3D LMMs) have shown tremendous potential in 3D-vision-based dialogue and reasoning. However, how to further enhance 3D LMMs to achieve fine-grained scene understanding and facilitate flexible human-agent interaction remains a challenging problem. In this work, we introduce 3D-LLaVA, a simple yet highly powerful 3D LMM designed to act as an intelligent assistant in comprehending, reasoning, and interacting with the 3D world. Unlike existing top-performing methods that rely on complicated pipelines--such as offline multi-view feature extraction or additional task-specific heads--3D-LLaVA adopts a minimalist design with integrated architecture and only takes point clouds as input. At the core of 3D-LLaVA is a new Omni Superpoint Transformer (OST), which integrates three functionalities: (1) a visual feature selector that converts and selects visual tokens, (2) a visual prompt encoder that embeds interactive visual prompts into the visual token space, and (3) a referring mask decoder that produces 3D masks based on text description. This versatile OST is empowered by the hybrid pretraining to obtain perception priors and leveraged as the visual connector that bridges the 3D data to the LLM. After performing unified instruction tuning, our 3D-LLaVA reports impressive results on various benchmarks. The code and model will be released at https://github.com/djiajunustc/3D-LLaVA.