CVPR 2025 Papers

Addressing the retrieval of unsafe content from vision-language models such as CLIP is an important step towards real-world integration. Current efforts have relied on unlearning techniques that try to erase the model's knowledge of unsafe concepts. While effective in reducing unwanted outputs, unlearning limits the model's capacity to discern between safe and unsafe content. In this work, we introduce a novel approach that shifts from unlearning to an awareness paradigm by leveraging the inherent hierarchical properties of the hyperbolic space. We propose to encode safe and unsafe content as an entailment hierarchy, where both are placed in different regions of hyperbolic space. Our HySAC, Hyperbolic Safety-Aware CLIP, employs entailment loss functions to model the hierarchical and asymmetrical relations between safe and unsafe image-text pairs. This modelling - ineffective in standard vision-language models due to their reliance on Euclidean embeddings - endows the model with awareness of unsafe content, enabling it to serve as both a multimodal unsafe classifier and a flexible content retriever, with the option to dynamically redirect unsafe queries toward safer alternatives or retain the original output. Extensive experiments show that our approach not only enhances safety recognition, but also establishes a more adaptable and interpretable framework for content moderation in vision-language models. Our source code is available at: https://github.com/aimagelab/HySAC

Despite advancements in multimodal large language models (MLLMs), current approaches struggle in medium-to-long video understanding due to frame and context length limitations. As a result, these models often depend on frame sampling, which risks missing key information over time and lacks task-specific relevance. To address these challenges, we introduce **HierarQ**, a task-aware hierarchical Q-Former based framework that sequentially processes frames to bypass the need for frame sampling, while avoiding LLM's context length limitations. We introduce a lightweight two-stream language-guided feature modulator to incorporate task awareness in video understanding, with the entity stream capturing frame-level object information within a short context and the scene stream identifying their broader interactions over longer period of time. Each stream is supported by dedicated memory banks which enables our proposed **Hierar**chical **Q**uerying transformer (HierarQ) to effectively capture short and long-term context. Extensive evaluations on **10** video benchmarks across video understanding, question answering, and captioning tasks demonstrate HierarQ's state-of-the-art performance across most datasets, proving its robustness and efficiency for comprehensive video analysis. All code will be made available upon acceptance.

We present a novel, hardware-rasterized rendering approach for ray-based 3D Gaussian Splatting (RayGS), obtaining both fast and high-quality results for novel view synthesis. Our work contains a mathematically rigorous and geometrically intuitive derivation about how to efficiently estimate all relevant quantities for rendering RayGS models, structured with respect to standard hardware rasterization shaders. Our solution is the first enabling rendering RayGS models at sufficiently high frame rates to support quality-sensitive applications like Virtual and Mixed Reality. Our second contribution enables alias-free rendering for RayGS, by addressing MIP-related issues arising when rendering diverging scales during training and testing. We demonstrate significant performance gains, across different benchmark scenes, while retaining state-of-the-art appearance quality of RayGS.

The advancement of AI technology has significantly influenced production activities, increasing the focus on copyright protection for AI models. Model perceptual hashing offers an efficient solution for retrieving the pirated models. Existing methods, such as handcrafted feature-based and dual-branch network-based perceptual hashing, have proven effective in detecting pirated models. However, these approaches often struggle to differentiate non-pirated models, leading to frequent false positives in model authentication and protection. To address this challenge, this paper proposes a structurally-aware perceptual model hashing technique that achieved reduced false positives while maintaining high true positive rates. Specifically, we introduce a method for converting the diverse neural network structures into graph structures suitable for DNN processing, then utilize a graph neural network to learn their structural features representation. Our approach integrates perceptual parameter-based model hashing, achieving robust performance with higher detection accuracy and fewer false positives. Experimental results show that the proposed method has only 3% false positive rate when detecting the non-pirated model, and the detection accuracy of pirated model reaches more than 98%.

Modern vision-language models (VLMs) develop patch embedding and convolution backbone within vector space, especially Euclidean ones, at the very founding. When expanding VLMs to a galaxy-scale for understanding astronomical phenomena, the integration of spherical space for planetary orbits and hyperbolic spaces for black holes raises two formidable challenges. a) The current pre-training model is confined to Euclidean space rather than a comprehensive geometric embedding. b) The predominant architecture lacks suitable backbones for anisotropic physical geometries. In this paper, we introduced Galaxy-Walker, a geometry-aware VLM, for the universe-level vision understanding tasks. We proposed the geometry prompt that generates geometry tokens by random walks across diverse spaces on a multi-scale physical graph, along with a geometry adapter that compresses and reshapes the space anisotropy in a mixture-of-experts manner. Extensive experiments demonstrate the effectiveness of our approach, with Galaxy-Walker achieving state-of-the-art performance in both galaxy property estimation (R2 scores up to 0.91) and morphology classification tasks (up to +0.17 F1 improvement in challenging features), significantly outperforming both domain-specific models and general-purpose VLMs.

Recently, 3D multi-object tracking (MOT) has widely adopted the standard tracking-by-detection paradigm, which solves the association problem between detections and tracks. Many tracking-by-detection approaches establish constrained relationships between detections and tracks using a distance threshold to reduce confusion during association. However, this approach does not effectively and comprehensively utilize the information regarding objects due to the constraints of the distance threshold. In this paper, we propose GRAE-3DMOT, Geometry Relation-Aware Encoder 3D Multi-Object Tracking, which contains a geometric relation-aware encoder to produce informative features for association. The geometric relation-aware encoder consists of three components: a spatial relation-aware encoder, a spatiotemporal relation-aware encoder, and a distance-aware feature fusion layer. The spatial relation-aware encoder effectively aggregates detection features by comprehensively exploiting as many detections as possible. The spatiotemporal relation-aware encoder provides spatiotemporal relation-aware features by combing spatial and temporal relation features, where the spatiotemporal relation-aware features are transformed into association scores for MOT. The distance-aware feature fusion layer is integrated into both encoders to enhance the relation features of physically proximate objects. Experimental results demonstrate that the proposed GRAE-3DMOT outperforms the state-of-the-art on the nuScenes. Our approach achieves 73.7% and 70.2% AMOTA on the nuScenes validation and test sets using CenterPoint detections.

We propose a novel 3D gaze estimation approach that learns spatial relationships between the subject and objects in the scene, and outputs 3D gaze direction. Our method targets unconstrained settings, including cases where close-up views of the subject's eyes are unavailable, such as when the subject is distant or facing away. Previous approaches typically rely on either 2D appearance alone or incorporate limited spatial cues using depth maps in the non-learnable post-processing step. Estimating 3D gaze direction from 2D observations in these scenarios is challenging; variations in subject pose, scene layout, and gaze direction, combined with differing camera poses, yield diverse 2D appearances and 3D gaze directions even when targeting the same 3D scene. To address this issue, we propose GA3CE: Gaze-Aware 3D Context Encoding. Our method represents subject and scene using 3D poses and object positions, treating them as 3D context to learn spatial relationships in 3D space. Inspired by human vision, we align this context in an egocentric space, significantly reducing spatial complexity. Furthermore, we propose D^3 (direction-distance-decomposed) positional encoding to better capture the spatial relationship between 3D context and gaze direction in direction and distance space. Experiments demonstrate substantial improvements, reducing mean angle error by 13%-37% compared to leading baselines on benchmark datasets in single-frame settings.

Robotic manipulation in high-precision tasks is essential for numerous industrial and real-world applications where accuracy and speed are required. Yet current diffusion-based policy learning methods generally suffer from low computational efficiency due to the iterative denoising process during inference. Moreover, these methods do not fully explore the potential of generative models for enhancing information exploration in 3D environments. In response, we propose FlowRAM, a novel framework that leverages generative models to achieve region-aware perception, enabling efficient multimodal information processing. Specifically, we devise a Dynamic Radius Schedule, which enables adaptive perception, facilitating transitions from global scene comprehension to fine-grained geometric details. Furthermore, we integrate state space models to integrate multimodal information, while preserving linear computational complexity. In addition, we employ conditional flow matching to learn action poses by regressing deterministic vector fields, which simplifies the learning process while maintaining performance. We verify the effectiveness of the FlowRAM in the RLBench, an established manipulation benchmark, and achieve state-of-the-art performance. The results demonstrate that FlowRAM achieves a remarkable improvement, particularly in high-precision tasks, where it outperforms previous methods by 12.0% in average success rate. Additionally, FlowRAM is able to generate physically plausible actions for a variety of real-world tasks in less than 4 time steps, significantly increasing inference speed.

Recent efforts recognize the power of scale in 3D learning (e.g. PTv3) and attention mechanisms (e.g. FlashAttention).However, current point cloud backbones fail to holistically unify geometric locality, attention mechanisms, and GPU architectures in one view.In this paper, we introduce Flash3D Transformer, which aligns geometric locality and GPU tiling through a principled locality mechanism based on Perfect Spatial Hashing (PSH).The common alignment with GPU tiling naturally fuses our PSH locality mechanism with FlashAttention at negligible extra cost.This mechanism affords flexible design choices throughout the backbone that result in superior downstream task results.Flash3D outperforms state-of-the-art PTv3 results on benchmark datasets, delivering a 2.25x speed increase and 2.4x memory efficiency boost. This efficiency enables scaling to wider attention scopes and larger models without additional overhead.Such scaling allows Flash3D to achieve even higher task accuracies than PTv3 under the same compute budget.

Server aggregation conflict is a key challenge in personalized federated learning (PFL). While existing PFL methods have achieved significant progress with shallow base models (e.g., four-layer CNNs), they often overlook the negative impacts of deeper base models on personalization mechanisms. In this paper, we identify the phenomenon of deep model degradation in PFL, where as base model depth increases, the model becomes more sensitive to local client data distributions, thereby exacerbating server aggregation conflicts and ultimately reducing overall model performance. Moreover, we show that these conflicts manifest in insufficient global average updates and mutual constraints between clients. Motivated by our analysis, we proposed a two-stage conflict-aware layer-wise mitigation algorithm (FedCALM), which first constructs a conflict-free global update to alleviate negative conflicts, and then maximizes the benefits of all clients through a conflict-aware strategy. Notably, our method naturally leads to a selective mechanism that balances the tradeoff between clients involved in aggregation and the tolerance for conflicts. Consequently, it can boost the positive contribution to the clients even with the greatest conflicts with the global update. Extensive experiments across multiple datasets and deeper base models demonstrate that FedCALM outperforms four state-of-the-art (SOTA) methods by up to 9.88% and seamlessly integrates into existing PFL methods with performance improvements of up to 9.01%.

Image exposure correction enhances images captured under diverse real-world conditions by addressing issues of under- and over-exposure, which can result in the loss of critical details and hinder content recognition. While significant advancements have been made, current methods often fail to achieve optimal feature learning for effective correction.To overcome these challenges, we propose Exposure-slot, a novel framework that integrates a prompt-based slot-in-slot attention mechanism to cluster exposed feature regions and learn exposure-centric features for each cluster. By extending the Slot Attention algorithm with a hierarchical structure, our approach progressively clusters features, enabling precise and region-aware correction. In particular, learnable prompts tailored to exposure characteristics of slots further enhance feature quality, adapting dynamically to varying conditions. Our method delivers superior performance on benchmark datasets, surpassing the current state-of-the-art with a PSNR improvement of over 1.85 dB on the SICE dataset and 0.4 dB on the LCDP dataset, thereby establishing a new benchmark for multi-exposure correction. The source code can be found at:https://github.com/kdhRick2222/Exposure-slot.

Point tracking in videos is a fundamental task with applications in robotics, video editing, and more. While many vision tasks benefit from pre-trained feature backbones to improve generalizability, point tracking has primarily relied on simpler backbones trained from scratch on synthetic data, which may limit robustness in real-world scenarios. Additionally, point tracking requires temporal awareness to ensure coherence across frames, but using temporally-aware features is still underexplored. Most current methods often employ a two-stage process: an initial coarse prediction followed by a refinement stage to inject temporal information and correct errors from the coarse stage. These approach, however, is computationally expensive and potentially redundant if the feature backbone itself captures sufficient temporal information. In this work, we introduce Chrono, a feature backbone specifically designed for point tracking with built-in temporal awareness. Leveraging pre-trained representations from self-supervised learner DINOv2 and enhanced with a temporal adapter, Chrono effectively captures long-term temporal context, enabling precise prediction even without the refinement stage. Experimental results demonstrate that Chrono achieves state-of-the-art performance in a refiner-free setting on the TAP-Vid-DAVIS and TAP-Vid-Kinetics datasets, among common feature backbones used in point tracking as well as DINOv2, with exceptional efficiency.

Blurry video frame interpolation (BVFI), which aims to generate high-frame-rate clear videos from low-frame-rate blurry inputs, is a challenging yet significant task in computer vision. Current state-of-the-art approaches typically rely on linear or quadratic models to estimate intermediate motion. However, these methods often overlook depth variations that occur during fast object motion, leading to changes in object size and hindering interpolation performance. This paper proposes the Differential Curves-guided Blurry Video Frame Interpolation (DC-BVFI) framework, which leverages the differential curves theory to analyze and mitigate the effects of depth variations caused by object motion. Specifically, DC-BVFI consists of UBNet and MPNet. Unlike prior approaches that rely on optical flow for frame interpolation, MPNet is designed to estimate the 3D scene flow, which facilitates a more precise awareness of depth and velocity variations. Since scene flow cannot be directly inferred in the 2D frame space, UBNet is introduced to transform them into 3D point maps. Extensive experiments demonstrate that the proposed DC-BVFI framework surpasses state-of-the-art performance in simulated and real-world datasets.

Estimating full-body motion using the tracking signals of head and hands from VR devices holds great potential for various applications. However, the sparsity and unique distribution of observations present a significant challenge, resulting in an ill-posed problem with multiple feasible solutions (i.e., hypotheses). This amplifies uncertainty and ambiguity in full-body motion estimation, especially for the lower-body joints. Therefore, we propose a new method, EnvPoser, that employs a two-stage framework to perform full-body motion estimation using sparse tracking signals and pre-scanned environment from VR devices. EnvPoser models the multi-hypothesis nature of human motion through an uncertainty-aware estimation module in the first stage. In the second stage, we refine these multi-hypothesis estimates by integrating semantic and geometric environmental constraints, ensuring that the final motion estimation aligns realistically with both the environmental context and physical interactions.Qualitative and quantitative experiments on two public datasets demonstrate that our method achieves state-of-the-art performance, highlighting significant improvements in human motion estimation within motion-environment interaction scenarios. Project page: https://xspc.github.io/EnvPoser/.

Given an isolated garment image in a canonical product view and a separate image of a person, the virtual try-on task aims to generate a new image of the person wearing the target garment.Prior virtual try-on works face two major challenges in achieving this goal: a) the paired (human, garment) training data has limited availability; b) generating textures on the human that perfectly match that of the prompted garment is difficult, often resulting in distorted text and faded textures. Our work explores ways to tackle these issues through both synthetic data as well as model refinement. We introduce a garment extraction model that generates (human, synthetic garment) pairs from a single image of a clothed individual. The synthetic pairs can then be used to augment the training of virtual try-on. We also propose an Error-Aware Refinement-based Schrodinger Bridge (EARSB) that surgically targets localized generation errors for correcting the output of a base virtual try-on model. To identify likely errors, we propose a weakly-supervised error classifier that localizes regions for refinement, subsequently augmenting the Schrodinger Bridge's noise schedule with its confidence heatmap. Experiments on VITON-HD and DressCode-Upper demonstrate that our synthetic data augmentation enhances the performance of prior work, while EARSB improves the overall image quality. In user studies, our model is preferred by the users in an average of 59% of cases.

Most current video MLLMs rely on uniform frame sampling and image-level encoders, resulting in inefficient data processing and limited motion awareness. To address these challenges, we introduce EMA, an Efficient Motion-Aware video MLLM that utilizes compressed video structures as inputs. We propose a motion-aware GOP (Group of Pictures) encoder that fuses spatial and motion information within a GOP unit in the compressed video stream, generating compact, informative visual tokens. By integrating fewer but denser RGB frames with more but sparser motion vectors in this native slow-fast input architecture, our approach reduces redundancy and enhances motion representation. Additionally, we introduce MotionBench, a benchmark for evaluating motion understanding across four motion types: linear, curved, rotational, and contact-based. Experimental results show that EMA achieves state-of-the-art performance on both MotionBench and popular video question answering benchmarks, while reducing inference costs. Moreover, EMA demonstrates strong scalability, as evidenced by its competitive performance on long video understanding benchmarks.

Building reconstruction is a challenging problem at the intersection of computer vision, photogrammetry and computer graphics. 3D wireframe presents a compelling representation for building modeling through its compact structure. Existing wireframe reconstruction methods employing vertex detection and edge regression have achieved promising results. In this paper, we develop an Edge-aware Diffusion network, dubbed EdgeDiff. As a novel paradigm for wireframe reconstruction, the EdgeDiff generates wireframe models from noise using a conditional diffusion model. During the training process, the ground truth wireframes firstly are formulated as a set of parameterized edges and then diffused into a random noise distribution. EdgeDiff learns both the noise reversal process and the network structure simultaneously. During inference, EdgeDiff iteratively refines the generated edge distribution using the denoising diffusion implicit model, enabling flexible single- or multi-step denoising and dynamic adaptation to buildings of varying complexity. Additionally, given the unique structure of wireframes, we introduce an edge attention module to extract point-wise attention from point features, using it as auxiliary information to facilitate learning of edge cues and guide the network toward improved edge awareness. Extensive experiments on the real-world Building3D dataset demonstrate that our approach achieves state-of-the-art performance.

Echocardiography is crucial for cardiovascular disease detection but relies heavily on experienced sonographers. Echocardiography probe guidance systems, which provide real-time movement instructions for acquiring standard plane images, offer a promising solution for AI-assisted or fully autonomous scanning. However, developing effective machine learning models for this task remains challenging, as they must grasp heart anatomy and the intricate interplay between probe motion and visual signals. To address this, we present EchoWorld, a motion-aware world modeling framework for probe guidance that encodes anatomical knowledge and motion-induced visual dynamics, while effectively leveraging past visual-motion sequences to enhance guidance precision. EchoWorld employs a pre-training strategy inspired by world modeling principles, where the model predicts masked anatomical regions and simulates the visual outcomes of probe adjustments. Built upon this pre-trained model, we introduce a motion-aware attention mechanism in the fine-tuning stage that effectively integrates historical visual-motion data, enabling precise and adaptive probe guidance. Trained on more than one million ultrasound images from over 200 routine scans, EchoWorld effectively captures key echocardiographic knowledge, as validated by qualitative analysis. Moreover, our method significantly reduces guidance errors compared to existing visual backbones and guidance frameworks, excelling in both single-frame and sequential evaluation protocols. Code is available at https://github.com/LeapLabTHU/EchoWorld.

Training diffusion models (DMs) requires substantial computational resources due to multiple forward and backward passes across numerous timesteps, motivating research into efficient training techniques. In this paper, we propose EB-Diff-Train, a new efficient DM training approach that is orthogonal to other methods of accelerating DM training, by investigating and leveraging Early-Bird (EB) tickets--sparse subnetworks that manifest early in the training process and maintain high generation quality. We first investigate the existence of traditional EB tickets in DMs, enabling competitive generation quality without fully training a dense model. Then, we delve into the concept of diffusion-dedicated EB tickets, drawing on insights from varying importance of different timestep regions. These tickets adapt their sparsity levels according to the importance of corresponding timestep regions, allowing for aggressive sparsity during non-critical regions while conserving computational resources for crucial timestep regions. Building on this, we develop an efficient DM training technique that derives timestep-aware EB tickets, trains them in parallel, and combines them during inference for image generation. Extensive experiments validate the existence of both traditional and timestep-aware EB tickets, as well as the effectiveness of our proposed EB-Diff-Train method. This approach can significantly reduce training time both spatially and temporally--achieving 2.9x 5.8x speedups over training unpruned dense models, and up to 10.3x faster training compared to standard train-prune-finetune pipelines--without compromising generative quality. Our code is available at https://github.com/GATECH-EIC/Early-Bird-Diffusion.