CVPR 2025 论文列表

Autonomous driving demands safe motion planning, especially in critical "long-tail" scenarios. Recent end-to-end autonomous driving systems leverage large language models (LLMs) as planners to improve generalizability to rare events. However, using LLMs at test time introduces high computational costs. To address this, we propose DiMA, an end-to-end autonomous driving system that maintains the efficiency of an LLM-free (or vision-based) planner while leveraging the world knowledge of an LLM. DiMA distills the information from a multi-modal LLM to a vision-based end-to-end planner through a set of specially designed surrogate tasks. Under a joint training strategy, a scene encoder common to both networks produces structured representations that are semantically grounded as well as aligned to the final planning objective. Notably, the LLM is optional at inference, enabling robust planning without compromising on efficiency. Training with DiMA results in a 37% reduction in the L2 trajectory error and an 80% reduction in the collision rate of the vision-based planner, as well as a 44% trajectory error reduction in long-tail scenarios. \ours also achieves state-of-the-art performance on the nuScenes planning benchmark.

Depth completion involves predicting dense depth maps from sparse LiDAR inputs, a critical task for applications such as autonomous driving and robotics. However, sparse depth annotations from sensors limit the availability of dense supervision, which is necessary for learning detailed geometric features. To overcome this limitation, we propose a two-stage knowledge distillation framework that leverages powerful monocular foundation models to provide dense supervision for depth completion. In the first stage, we introduce a pre-training strategy that generates diverse training data from natural images to distill geometric knowledge to depth completion. Specifically, we simulate LiDAR scans by utilizing monocular depth and mesh reconstruction, thereby creating training data without requiring ground-truth depth. Nonetheless, monocular depth estimation suffers from inherent scale ambiguity in real-world settings. To address this, in the second stage, we employ a scale- and shift-invariant loss (SSI Loss) to learn real-world scales when fine-tuning on real-world datasets. Our two-stage distillation framework enables depth completion models to harness the strengths of monocular foundation models. Experimental results show that models trained with our two-stage distillation framework achieve top-ranked performance on the KITTI benchmark, demonstrating improvements in both quantitative and qualitative metrics.

Dataset distillation (DD) aims to synthesize a small information-rich dataset from a large dataset for efficient neural network training. However, existing dataset distillation methods struggle with long-tailed datasets, which are prevalent in real-world scenarios. By investigating the reasons behind this unexpected result, we identified two main causes: 1) The distillation process on imbalanced datasets develops biased gradients, leading to the synthesis of similarly imbalanced distilled datasets. 2) The experts trained on such datasets perform suboptimally on tail classes, resulting in misguided distillation supervision and poor-quality soft-label initialization. To address these issues, we first propose Distribution-agnostic Matching to avoid directly matching the biased expert trajectories. It reduces the distance between the student and the biased expert trajectories and prevents the tail class bias from being distilled to the synthetic dataset. Moreover, we improve the distillation guidance with Expert Decoupling, which jointly matches the decoupled backbone and classifier to improve the tail class performance and initialize reliable soft labels. This work pioneers the field of long-tailed dataset distillation (LTDD), marking the first effective effort to distill long-tailed datasets. Our code will be made public at https://github.com/ichbill/LTDD.

The integration of multimodal data including pathology images and gene profiles is widely applied in precise survival prediction. Despite recent advances in multimodal survival models, collecting complete modalities for multimodal fusion still poses a significant challenge, hindering their application in clinical settings. Current approaches tackling incomplete modalities often fall short, as they typically compensate for only a limited part of the knowledge of missing modalities. To address this issue, we propose a Distilled Prompt Learning framework (DisPro) to utilize the strong robustness of Large Language Models (LLMs) to missing modalities, which employs two-stage prompting for compensation of comprehensive information for missing modalities. In the first stage, Unimodal Prompting (UniPro) distills the knowledge distribution of each modality, preparing for supplementing modality-specific knowledge of the missing modality in the subsequent stage. In the second stage, Multimodal Prompting (MultiPro) leverages available modalities as prompts for LLMs to infer the missing modality, which provides modality-common information. Simultaneously, the unimodal knowledge acquired in the first stage is injected into multimodal inference to compensate for the modality-specific knowledge of the missing modality. Extensive experiments covering various missing scenarios demonstrated the superiority of the proposed method. The code is available at https://github.com/Innse/DisPro.

Active Real-time interaction with video LLMs introduces a new paradigm for human-computer interaction, where the model not only understands user intent but also responds while continuously processing streaming video on the fly. Unlike offline video LLMs, which analyze the entire video before answering questions, active real-time interaction requires three capabilities: 1) Perception: real-time video monitoring and interaction capturing. 2) Decision: raising proactive interaction in proper situations, 3) Reaction: continuous interaction with users. However, inherent conflicts exist among the desired capabilities. The Decision and Reaction require a contrary Perception scale and grain, and the autoregressive decoding blocks the real-time Perception and Decision during the Reaction. To unify the conflicted capabilities within a harmonious system, we present Dispider, a solution built on a Disentangled Perception, Decision, and Reaction framework. Dispider features a lightweight Proactive Streaming Video Processing module that tracks the video stream and identifies optimal moments for interaction. Once the interaction is triggered, an asynchronous Precise Interaction module provides detailed responses, while the processing module continues to monitor the video in the meantime. Our disentangled and asynchronous design ensures timely, contextually accurate, and computationally efficient responses, making Dispider ideal for active real-time interaction for long-duration video streams. Experiments prove that Dispider outperforms existing methods not only in its superior understanding of video content in conventional video QA settings, but also in proactive response capability and temporal awareness under the streaming setting.

DiskVPS is a novel and robust vanishing point (VP) detection scheme based on Hough Transform (HT) over an image-plane-mapped disk region. In the DiskVPS, the image plane is first mapped into a disk region, which is then partitioned into concentrical ring-shaped subspaces. Each subspace is further partitioned into approximately equal-probability Hough cells. By using individual edges rather than edge pairs as voters, DiskVPS can achieve high accuracy with extreme efficiency. Involving no calibration parameters, DiskVPS is particularly suitable for VP detection in uncalibrated images and thus can be applied in image calibration. A comparative experimental study demonstrates that the basic DiskVPS model without parameter optimization achieved significantly better performance over the SOTA in detection accuracy and processing speed with real-world images. The study also shows that DiskVPS is robust against parameter changes.

State-of-the-art machine learning systems are vulnerable to small perturbations to their input, where _small_ is defined according to a threat model that assigns a positive threat to each perturbation. Most prior works define a task-agnostic, isotropic, and global threat, like the l_p norm, where the magnitude of the perturbation fully determines the degree of the threat and neither the direction of the attack nor its position in space matter. However, common corruptions in computer vision, such as blur, compression, or occlusions, are not well captured by such treat models. This paper proposes a novel threat model called \texttt Projected Displacement (PD) to study robustness beyond existing isotropic and global threat models. The proposed threat model measures the threat of a perturbation via its alignment with _unsafe directions_, defined as directions in the input space along which a perturbation of sufficient magnitude changes the ground truth class label. Unsafe directions are identified locally for each input based on observed training data. In this way, the PD-threat model exhibits anisotropy and locality. The PD-threat model is computationally efficient and can be easily integrated into existing robustness pipelines. Experiments on Imagenet-1k data indicate that, for any input, the set of perturbations with small PD threat includes _safe_ perturbations of large l_p norm that preserve the true label, such as noise, blur and compression, while simultaneously excluding _unsafe_ perturbations that alter the true label. Unlike perceptual threat models based on embeddings of large-vision models, the PD-threat model can be readily computed for arbitrary classification tasks without pre-training or finetuning. Further additional task information such as sensitivity to image regions or concept hierarchies can be easily integrated into the assessment of threat and thus the PD threat model presents practitioners a flexible, task-driven threat specification that alleviates the limitations of l_p-threat models.

Human pose generation is a complex task due to the non-rigid and highly variable nature of human body structures and appearances. However, existing methods often overlook the fundamental differences between spatial transformations of poses and texture generation for appearance, which makes them prone to overfitting. To address this issue, we propose a multi-pose generation framework driven by disentangled pose and appearance guidance. Our approach includes a Global-aware Pose Generation module that iteratively generates pose embeddings, enabling effective control over non-rigid body deformations. Additionally, we introduce the Global-aware Transformer Decoder, which leverages similarity queries and attention mechanisms to achieve spatial transformations and enhance pose consistency through a Global-aware block. In the appearance generation phase, we condition a diffusion model on pose embeddings produced in the initial stage and introduce an Appearance Adapter that extracts high-level contextual semantic information from multi-scale features, enabling further refinement of pose appearance textures and providing appearance guidance. Extensive experiments on the UBC Fashion and TikTok datasets demonstrate that our framework achieves state-of-the-art results in both quality and fidelity, establishing it as a powerful approach for complex pose generation tasks.

Generating continuous sign language videos from discrete segments is challenging due to the need for smooth transitions that preserve natural flow and meaning. Traditional approaches that simply concatenate isolated signs often result in abrupt transitions, disrupting video coherence. To address this, we propose a novel framework, Sign-D2C, that employs a conditional diffusion model to synthesize contextually smooth transition frames, enabling the seamless construction of continuous sign language sequences. Our approach transforms the unsupervised problem of transition frame generation into a supervised training task by simulating the absence of transition frames through random masking of segments in long-duration sign videos. The model learns to predict these masked frames by denoising Gaussian noise, conditioned on the surrounding sign observations, allowing it to handle complex, unstructured transitions. During inference, we apply a linearly interpolating padding strategy that initializes missing frames through interpolation between boundary frames, providing a stable foundation for iterative refinement by the diffusion model. Extensive experiments on the PHOENIX14T, USTC-CSL100, and USTC-SLR500 datasets demonstrate the effectiveness of our method in producing continuous, natural sign language videos.

Infants develop complex visual understanding rapidly, even preceding of the acquisition of linguistic skills. As computer vision seeks to replicate the human vision system, understanding infant visual development may offer valuable insights. In this paper, we present an interdisciplinary study exploring this question: can a computational model that imitates the infant learning process develop broader visual concepts that extend beyond the vocabulary it has heard, similar to how infants naturally learn? To investigate this, we analyze a recently published model in Science by Vong et al., which is trained on longitudinal, egocentric images of a single child paired with transcribed parental speech. We perform neuron labeling to identify visual concept neurons hidden in the model's internal representations. We then demonstrate that these neurons can recognize objects beyond the model's original vocabulary. Furthermore, we compare the differences in representation between infant models and those in modern computer vision models, such as CLIP and ImageNet pre-trained model. Ultimately, our work bridges cognitive science and computer vision by analyzing the internal representations of a computational model trained on an infant visual and linguistic inputs. Our code is available at https://github.com/Kexueyi/discover_infant_vis.

Concept Bottleneck Models (CBMs) try to make the decision-making process transparent by exploring an intermediate concept space between the input image and the output prediction. Existing CBMs just learn coarse-grained relations between the whole image and the concepts, less considering local image information, leading to two main drawbacks: i) they often produce spurious visual-concept relations, hence decreasing model reliability; and ii) though CBMs could explain the importance of every concept to the final prediction, it is still challenging to tell which visual region produces the prediction. To solve these problems, this paper proposes a Disentangled Optimal Transport CBM (DOT-CBM) framework to explore fine-grained visual-concept relations between local image patches and concepts. Specifically, we model the concept prediction process as a transportation problem between the patches and concepts, thereby achieving explicit fine-grained feature alignment. We also incorporate orthogonal projection losses within the modality to enhance local feature disentanglement. To further address the shortcut issues caused by statistical biases in the data, we utilize the visual saliency map and concept label statistics as transportation priors. Thus, DOT-CBM can visualize inversion heatmaps, provide more reliable concept predictions, and produce more accurate class predictions. Comprehensive experiments demonstrate that our proposed DOT-CBM achieves SOTA performance on several tasks, including image classification, local part detection and out-of-distribution generalization.

The parameter-efficient adaptation of the image-text pretraining model CLIP for video-text retrieval is a prominent area of research. While CLIP is focused on image-level vision-language matching, video-text retrieval demands comprehensive understanding at the video level. Three key discrepancies emerge in the transfer from image-level to video-level: vision, language, and alignment. However, existing methods mainly focus on vision while neglecting language and alignment. In this paper, we propose Discrepancy Reduction in Vision, Language, and Alignment (DiscoVLA), which simultaneously mitigates all three discrepancies. Specifically, we introduce Image-Video Features Fusion to integrate image-level and video-level features, effectively tackling both vision and language discrepancies. Additionally, we generate pseudo image captions to learn fine-grained image-level alignment. To mitigate alignment discrepancies, we propose Image-to-Video Alignment Distillation, which leverages image-level alignment knowledge to enhance video-level alignment. Extensive experiments demonstrate the superiority of our DiscoVLA. In particular, on MSRVTT with CLIP (ViT-B/16), DiscoVLA outperforms previous methods by 2.2% R@1 and 7.5% R@sum. The code is available at https://github.com/LunarShen/DsicoVLA.

We present Disco4D, a novel Gaussian Splatting framework for 4D human generation and animation from a single image. Different from existing methods, Disco4D distinctively disentangles clothings (with Gaussian models) from the human body (with SMPL-X model), significantly enhancing the generation details and flexibility. It has the following technical innovations. (1) Disco4D learns to efficiently fit the clothing Gaussians over the SMPL-X Gaussians. (2) It adopts diffusion models to enhance the 3D generation process, e.g. modeling occluded parts not visible in the input image. (3) It learns an identity encoding for each clothing Gaussian to facilitate the separation and extraction of clothing assets. Furthermore, Disco4D naturally supports 4D human animation with vivid dynamics. Extensive experiments demonstrate the superiority of Disco4D on 4D human generation and animation tasks.

In image classification, the label quality of training data critically influences model generalization, especially for deep neural networks (DNNs). Traditionally, learning from noisy labels (LNL) can improve the generalization of DNNs through complex architectures or a series of robust techniques, but its performance improvement is limited by the discriminative paradigm. Unlike traditional ways, we resolve the LNL problems from the perspective of robust label generation, based on diffusion models within the generative paradigm. To expand the diffusion model into a robust classifier that explicitly accommodates more noise knowledge, we propose a Directional Label Diffusion (DLD) model. It disentangles the diffusion process into two paths, i.e., directional diffusion and random diffusion. Specifically, directional diffusion simulates the corruption of true labels into a directed noise distribution, prioritizing the removal of likely noise, whereas random diffusion introduces inherent randomness to support label recovery. This architecture enable DLD to gradually infer labels from an initial random state, interpretably diverging from the specified noise distribution. To adapt the model to diverse noisy environments, we design a low-cost label pre-correction method that automatically supplies more accurate label information to the diffusion model, without requiring manual intervention or additional iterations. Our approach outperforms state-of-the-art methods on both simulated and real-world noisy datasets. Code is available at https://github.com/SenyuHou/DLD.

We present DirectTriGS, a novel framework designed for 3D object generation with Gaussian Splatting (GS). GS-based rendering for 3D content has gained considerable attention recently. However, there has been limited exploration in directly generating 3D Gaussians compared to traditional generative modeling approaches. The main challenge lies in the complex data structure of GS represented by discrete point clouds with multiple channels.To overcome this challenge, we propose employing the triplane representation, which allows us to represent Gaussian Splatting as an image-like continuous field. This representation effectively encodes both the geometry and texture information, enabling smooth transformation back to Gaussian point clouds and rendering into images by a TriRenderer, with only 2D supervisions. The proposed TriRenderer is fully differentiable, so that the rendering loss can supervise both texture and geometry encoding. Furthermore, the triplane representation can be compressed using a Variational Autoencoder (VAE), which can subsequently be utilized in latent diffusion to generate 3D objects.The experiments demonstrate that the proposed generation framework can produce high-quality 3D object geometry and rendering results.

Recent studies highlighted a practical setting of unsupervised anomaly detection (UAD) that builds a unified model for multi-class images. Despite various advancements addressing this challenging task, the detection performance under the multi-class setting still lags far behind state-of-the-art class-separated models. Our research aims to bridge this substantial performance gap. In this paper, we present Dinomaly, a minimalist reconstruction-based anomaly detection framework that harnesses pure Transformer architectures without relying on complex designs, additional modules, or specialized tricks. Given this powerful framework consisting of only Attentions and MLPs, we found four simple components that are essential to multi-class anomaly detection: (1) Scalable foundation Transformers that extracts universal and discriminative features, (2) Noisy Bottleneck where pre-existing Dropouts do all the noise injection tricks, (3) Linear Attention that naturally cannot focus, and (4) Loose Reconstruction that does not force layer-to-layer and point-by-point reconstruction. Extensive experiments are conducted across popular anomaly detection benchmarks including MVTec-AD, VisA, and Real-IAD. Our proposed Dinomaly achieves impressive image-level AUROC of 99.6%, 98.7%, and 89.3% on the three datasets respectively, which is not only superior to state-of-the-art multi-class UAD methods, but also achieves the most advanced class-separated UAD records.

Current Structure-from-Motion (SfM) methods typically follow a two-stage pipeline, combining learned or geometric pairwise reasoning with a subsequent global optimization step. In contrast, we propose a data-driven multi-view reasoning approach that directly infers 3D scene geometry and camera poses from multi-view images. Our framework, DiffusionSfM, parameterizes scene geometry and cameras as pixel-wise ray origins and endpoints in a global frame and employs a transformer-based denoising diffusion model to predict them from multi-view inputs. To address practical challenges in training diffusion models with missing data and unbounded scene coordinates, we introduce specialized mechanisms that ensure robust learning. We empirically validate DiffusionSfM on both synthetic and real datasets, demonstrating that it outperforms classical and learning-based approaches while naturally modeling uncertainty.

While event-based deblurring have demonstrated impressive results, they are impractical for consumer photos captured by cell phones and digital cameras that are not equipped with the event sensor. To address this problem, we in this paper propose a novel deblurring framework called Event Generation Deblurring (EGDeblurring), which allows to effectively deblur an image by generating event guidance describing the motion information using a diffusion model. Specifically, we design a motion prior generation diffusion model and a feature extractor to produce prior information beneficial for deblurring, rather than generating the raw event representation. In order to achieve effective fusion of motion prior information with blurry images and produce high-quality results, we develop a regression deblurring network embedded with a dual-attention channel fusion block. Experiments on multiple datasets demonstrate that our method outperforms state-of-the-art image deblurring methods. Our code is available at https://github.com/XinanXie/EGDeblurring https://github.com/XinanXie/EGDeblurring.