CVPR 2025 论文列表

Recently, learned video compression (LVC) is undergoing a period of rapid development. However, due to absence of large and high-quality high dynamic range (HDR) video training data, LVC on HDR video is still unexplored. In this paper, we are the first to collect a large-scale HDR video benchmark dataset, named HDRVD2K, featuring huge quantity, diverse scenes and multiple motion types. HDRVD2K fills gaps of video training data and facilitate the development of LVC on HDR videos. Based on HDRVD2K, we further propose the first learned bit-depth scalable video compression (LBSVC) network for HDR videos by effectively exploiting bit-depth redundancy between videos of multiple dynamic ranges. To achieve this, we first propose a compression-friendly bit-depth enhancement module (BEM) to effectively predict original HDR videos based on compressed tone-mapped low dynamic range (LDR) videos and dynamic range prior, instead of reducing redundancy only through spatio-temporal predictions. Our method greatly improves the reconstruction quality and compression performance on HDR videos. Extensive experiments demonstrate the effectiveness of HDRVD2K on learned HDR video compression and great compression performance of our proposed LBSVC network.

Dataset distillation is an emerging dataset reduction method, which condenses large-scale datasets while maintaining task accuracy. Current parameterization methods achieve enhanced performance under extremely high compression ratio by optimizing determined synthetic dataset in informative feature domain. However, they limit themselves to a fixed optimization space for distillation, neglecting the diverse guidance across different informative latent spaces. To overcome this limitation, we propose a novel parameterization method dubbed Hierarchical Parameterization Distillation (H-PD), to systematically explore hierarchical feature within provided feature space (e.g., layers within pre-trained generative adversarial networks). We verify the correctness of our insights by applying the hierarchical optimization strategy on GAN-based parameterization method. In addition, we introduce a novel class-relevant feature distance metric to alleviate the computational burden associated with synthetic dataset evaluation, bridging the gap between synthetic and original datasets. Experimental results demonstrate that the proposed H-PD achieves a significant performance improvement under various settings with equivalent time consumption, and even surpasses current generative distillation using diffusion models under extreme compression ratios IPC=1 and IPC=10.

In mixed reality applications, a realistic acoustic experience in spatial environments is as crucial as the visual experience for achieving true immersion. Despite recent advances in neural approaches for Room Impulse Response (RIR) estimation, most existing methods are limited to the single environment on which they are trained, lacking the ability to generalize to new rooms with different geometries and surface materials. We aim to develop a unified model capable of reconstructing the spatial acoustic experience of any environment with minimum additional measurements. To this end, we present xRIR, a framework for cross-room RIR prediction. The core of our generalizable approach lies in combining a geometric feature extractor, which captures spatial context from panorama depth images, with a RIR encoder that extracts detailed acoustic features from only a few reference RIR samples. To evaluate our method, we introduce AcousticRooms, a new dataset featuring high-fidelity simulation of over 300,000 RIRs from 260 rooms. Experiments show that our method strongly outperforms a series of baselines. Furthermore, we successfully perform sim-to-real transfer by evaluating our model on four real-world environments, demonstrating the generalizability of our approach and the realism of our dataset.

Recent contrastive multimodal vision-language models like CLIP have demonstrated robust open-world semantic understanding, becoming the standard image backbones for vision-language applications. However, recent findings suggest high semantic similarity between well-trained unimodal encoders, which raises a key question: Are semantically similar embedding spaces separated only by simple projection transformations? To validate this, we propose a novel framework that aligns vision and language using frozen unimodal encoders. It involves selecting semantically similar encoders in the latent space, curating a concept-rich dataset of image-caption pairs, and training simple MLP projectors. We evaluated our approach on various tasks involving both strong unimodal vision (0-shot localization) and language encoders (multi-lingual, long context) and show that simple Projectors retain unimodal capabilities in joint embedding space. Furthermore, our best model, utilizing DINOv2 and All-Roberta-Large text encoder, achieves 76(%) accuracy on ImageNet with a 20-fold reduction in data and 65-fold reduction in compute requirements compared to multimodal alignment where models are trained from scratch. The proposed framework enhances the accessibility of multimodal model development while enabling flexible adaptation across diverse scenarios. Code and curated datasets are available at github.com/mayug/freeze-align.

This paper introduces HarmonySet, a comprehensive dataset designed to advance video-music understanding. HarmonySet consists of 48,328 diverse video-music pairs, annotated with detailed information on rhythmic synchronization, emotional alignment, thematic coherence, and cultural relevance. We propose a multi-step human-machine collaborative framework for efficient annotation, combining human insights with machine-generated descriptions to identify key transitions and assess alignment across multiple dimensions. Additionally, we introduce a novel evaluation framework with tasks and metrics to assess the multi-dimensional alignment of video and music, including rhythm, emotion, theme, and cultural context. Our extensive experiments demonstrate that HarmonySet, along with the proposed evaluation framework, significantly improves the ability of multimodal models to capture and analyze the intricate relationships between video and music. Project page: https://harmonyset.github.io/.

Existing approaches of hand reconstruction predominantly adhere to a multi-stage framework, encompassing detection, left-right classification, and pose estimation. This paradigm induces redundant computation and cumulative errors. In this work, we propose HandOS, an end-to-end framework for 3D hand reconstruction. Our central motivation lies in leveraging a frozen detector as the foundation while incorporating auxiliary modules for 2D and 3D keypoint estimation. In this manner, we integrate the pose estimation capacity into the detection framework, while at the same time obviating the necessity of using the left-right category as a prerequisite. Specifically, we propose an interactive 2D-3D decoder, where 2D joint semantics is derived from detection cues while 3D representation is lifted from those of 2D joints. Furthermore, hierarchical attention is designed to enable the concurrent modeling of 2D joints, 3D vertices, and camera translation. Consequently, we achieve an end-to-end integration of hand detection, 2D pose estimation, and 3D mesh reconstruction within a one-stage framework, so that the above multi-stage drawbacks are overcome. Meanwhile, the HandOS reaches state-of-the-art performances on public benchmarks, e.g., 5.0 PA-MPJPE on FreiHand and 64.6% PCK@0.05 on HInt-Ego4D.

We present HOTFormerLoc, a novel and versatile Hierarchical Octree-based TransFormer, for large-scale 3D place recognition in both ground-to-ground and ground-to-aerial scenarios across urban and forest environments. We propose an octree-based multi-scale attention mechanism that captures spatial and semantic features across granularities. To address the variable density of point distributions from spinning lidar, we present cylindrical octree attention windows to reflect the underlying distribution during attention.We introduce relay tokens to enable efficient global-local interactions and multi-scale representation learning at reduced computational cost. Our pyramid attentional pooling then synthesises a robust global descriptor for end-to-end place recognition in challenging environments. In addition, we introduce CS-Wild-Places, a novel 3D cross-source dataset featuring point cloud data from aerial and ground lidar scans captured in dense forests. Point clouds in CS-Wild-Places contain representational gaps and distinctive attributes such as varying point densities and noise patterns, making it a challenging benchmark for cross-view localisation in the wild. HOTFormerLoc achieves a top-1 average recall improvement of 5.5% - 11.5% on the CS-Wild-Places benchmark. Furthermore, it consistently outperforms SOTA 3D place recognition methods, with an average performance gain of 4.9% on well-established urban and forest datasets. The code and CS-Wild-Places benchmark is available at https://csiro-robotics.github.io/HOTFormerLoc.

We introduce HOT3D, a publicly available dataset for egocentric hand and object tracking in 3D. The dataset offers over 833 minutes (3.7M+ images) of recordings that feature 19 subjects interacting with 33 diverse rigid objects. In addition to simple pick-up, observe, and put-down actions, the subjects perform actions typical for a kitchen, office, and living room environment. The recordings include multiple synchronized data streams containing egocentric multi-view RGB/monochrome images, eye gaze signal, scene point clouds, and 3D poses of cameras, hands, and objects. The dataset is recorded with two headsets from Meta: Project Aria, which is a research prototype of AI glasses, and Quest 3, a virtual-reality headset that has shipped millions of units. Ground-truth poses were obtained by a motion-capture system using small optical markers attached to hands and objects. Hand annotations are provided in the UmeTrack and MANO formats, and objects are represented by 3D meshes with PBR materials obtained by an in-house scanner. In our experiments, we demonstrate the effectiveness of multi-view egocentric data for three popular tasks: 3D hand tracking, model-based 6DoF object pose estimation, and 3D lifting of unknown in-hand objects. The evaluated multi-view methods, whose benchmarking is uniquely enabled by HOT3D, significantly outperform their single-view counterparts.

Despite progress in stereo depth estimation, omnidirectional imaging remains underexplored, mainly due to the lack of appropriate data. We introduce Helvipad, a real-world dataset for omnidirectional stereo depth estimation, featuring 40K video frames from video sequences across diverse environments, including crowded indoor and outdoor scenes with various lighting conditions. Collected using two 360deg cameras in a top-bottom setup and a LiDAR sensor, the dataset includes accurate depth and disparity labels by projecting 3D point clouds onto equirectangular images. Additionally, we provide an augmented training set with an increased label density by using depth completion. We benchmark leading stereo depth estimation models for both standard and omnidirectional images. The results show that while recent stereo methods perform decently, a challenge persists in accurately estimating depth in omnidirectional imaging. To address this, we introduce necessary adaptations to stereo models, leading to improved performance.

We present a validation dataset of newly-collected kitchen based egocentric videos, manually annotated with highly detailed and interconnected ground-truth labels covering: recipe steps, fine-grained actions, ingredients with nutritional values, moving objects, and audio annotations. Importantly, all annotations are grounded in 3D through digital twinning of the scene, fixtures, object locations, and primed with gaze. Footage is collected from unscripted recordings in diverse home environments, making HD-EPIC the first dataset collected in-the-wild but with detailed annotations matching those in controlled lab environments. We show the potential of our highly-detailed annotations through a challenging VQA benchmark of 26K questions assessing capability to recognise recipes, ingredients, nutrition, fine-grained actions, 3D perception, object motion, and gaze direction. The powerful long-context Gemini Pro only achieves 37.0% on this benchmark, showcasing its difficulty and highlighting shortcomings in current VLMs. We additionally assess action recognition, sound recognition, and long-term video-object segmentation on HD-EPIC. HD-EPIC is 41 hours of video in 9 kitchens with digital twins of 404 kitchen fixtures, capturing 69 recipes, 59K fine-grained actions, 51K audio events, 20K object movements and 37K object masks lifted to 3D. On average, we have 263 annotations per min of our unscripted videos.

This work presents a simple yet effective workflow for automatically scaling instruction-following data to elicit pixel-level grounding capabilities of VLMs under complex instructions. In particular, we address five critical real-world challenges in text-instruction-based grounding: hallucinated references, multi-object scenarios, reasoning, multi-granularity, and part-level references. By leveraging knowledge distillation from a pre-trained teacher model, our approach generates high-quality instruction-response pairs linked to existing pixel-level annotations, minimizing the need for costly human annotation. The resulting dataset, Ground-V, captures rich object localization knowledge and nuanced pixel-level referring expressions. Experiment results show that models trained on Ground-V exhibit substantial improvements across diverse grounding tasks. Specifically, incorporating \dataset during training directly achieve an average accuracy boost of 4.4% for LISA and a 7.9% for PSALM across six benchmarks on the gIoU metric. It also sets new state-of-the-art results on standard benchmarks such as RefCOCO/+/g. Notably, on gRefCOCO, we achieve an N-Acc of 83.3%, exceeding the previous state-of-the-art by more than 20%.

Understanding bimanual human hand activities is a critical problem in AI and robotics. We cannot build large models of bimanual activities because existing datasets lack the scale, coverage of diverse hand activities, and detailed annotations. We introduce GigaHands, a massive annotated dataset capturing 34 hours of bimanual hand activities from 56 subjects and 417 objects, totaling 14k motion clips derived from 183 million frames paired with 84k text annotations. Our markerless capture setup and data acquisition protocol enable fully automatic 3D hand and object estimation while minimizing the effort required for text annotation. The scale and diversity of GigaHands enable broad applications, including text-driven action synthesis, hand motion captioning, and dynamic radiance field reconstruction.

We present Geometry in Style, a new method for identity-preserving mesh stylization. Existing techniques either adhere to the original shape through overly restrictive deformations such as bump maps or significantly modify the input shape using expressive deformations that may introduce artifacts or alter the identity of the source shape. In contrast, we represent a deformation of a triangle mesh as a target normal vector for each vertex neighborhood. The deformations we recover from target normals are expressive enough to enable detailed stylizations yet restrictive enough to preserve the shape's identity. We achieve such deformations using our novel differentiable As-Rigid-As-Possible (dARAP) layer, a neural-network-ready adaptation of the classical ARAP algorithm which we use to solve for per-vertex rotations and deformed vertices. As a differentiable layer, dARAP is paired with a visual loss from a text-to-image model to drive deformations toward style prompts, altogether giving us Geometry in Style.

Thanks to the introduction of large-scale datasets, deep-learning has become the mainstream approach for appearance-based gaze estimation problems. However, current large-scale datasets contain annotation errors and provide only a single vector for gaze annotation, lacking key information such as 3D eyeball structures. Limitations in annotation accuracy and variety have constrained the progress in research and development of deep-learning methods for appearance-based gaze-related tasks. In this paper, we present GazeGene, a new large-scale synthetic gaze dataset with photo-realistic samples. More importantly, GazeGene not only provides accurate gaze annotations, but also offers 3D annotations of vital eye structures such as the pupil, iris, eyeball, optical and visual axes for the first time. Experiments show that GazeGene achieves comparable quality and generalization ability with real-world datasets, even outperforms most existing datasets on high-resolution images. Furthermore, its 3D eyeball annotations expand the application of deep-learning methods on various gaze-related tasks, offering new insights into this field. The dataset is available at: https://phiai.buaa.edu.cn/GazeGene/.

3D Gaussian Splatting (3DGS) has emerged as a mainstream for novel view synthesis, leveraging continuous aggregations of Gaussian functions to model scene geometry. However, 3DGS suffers from substantial memory requirements to store the large amount of Gaussians, hindering its efficiency and practicality. To address this challenge, we introduce GaussianSpa, an optimization-based simplification framework for compact and high-quality 3DGS. Specifically, we formulate the simplification objective as a constrained optimization problem associated with the 3DGS training. Correspondingly, we propose an efficient "optimizing-sparsifying" solution for the formulated problem, alternately solving two independent sub-problems and gradually imposing substantial sparsity onto the Gaussians in the 3DGS training process. We conduct quantitative and qualitative evaluations on various datasets, demonstrating the superiority of GaussianSpa over existing state-of-the-art approaches. Notably, GaussianSpa achieves an average PSNR improvement of 0.9 dB on the real-world Deep Blending dataset with 10X fewer Gaussians compared to the vanilla 3DGS.

Recent approaches, such as data augmentation, adversarial training, and transfer learning, have shown potential in addressing the issue of performance degradation caused by distributional shifts. However, they typically demand careful design in terms of data or models and lack awareness of the impact of distributional shifts. In this paper, we observe that classification errors arising from distribution shifts tend to cluster near the true values, suggesting that misclassifications commonly occur in semantically similar, neighboring categories. Furthermore, robust advanced vision foundation models maintain larger inter-class distances while preserving semantic consistency, making them less vulnerable to such shifts. Building on these findings, we propose a new method called GFN (Gain From Neighbors), which uses gradient priors from neighboring classes to perturb input images and incorporates an inter-class distance-weighted loss to improve class separation. This approach encourages the model to learn more resilient features from data prone to errors, enhancing its robustness against shifts in diverse settings. In extensive experiments across various model architectures and benchmark datasets, GFN consistently demonstrated superior performance. For instance, compared to the current state-of-the-art TAPADL method, our approach achieved a higher corruption robustness of 41.4% on ImageNet-C (+2.3%), without requiring additional parameters and using only minimal data.

Garments are common in daily life and are important for embodied intelligence community. Current category-level garments pose tracking works focus on predicting point-wise canonical correspondence and learning a shape deformation in point cloud sequences. In this paper, motivated by the 2D warping space and shape prior, we propose GaPT-DAR, a novel category-level Garments Pose Tracking framework with integrated 2D Deformation And 3D Reconstruction function, which fully utilize 3D-2D projection and 2D-3D reconstruction to transform the 3D point-wise learning into 2D warping deformation learning. Specifically, GaPT-DAR firstly builds a Voting-based Project module that learns the optimal 3D-2D projection plane for maintaining the maximum orthogonal entropy during point projection. Next, a Garments Deformation module is designed in 2D space to explicitly model the garments warping procedure with deformation parameters. Finally, we build a Depth Reconstruction module to recover the 2D images into 3D warp field. We provide extensive experiments on VR-Folding dataset to evaluate our GaPT-DAR and the results show obvious improvements on most of the metrics compared to state-of-the-arts ( GarmentNets and GarmentTracking). More details are available at https://sites.google.com/view/gapt-dar.

Vision-language models like CLIP have shown impressive capabilities in aligning images and text, but they often struggle with lengthy and detailed text descriptions because of their training focus on short and concise captions. We present GOAL (Global-local Object Alignment Learning), a novel fine-tuning method that enhances CLIP's ability to handle lengthy text by leveraging both global and local semantic alignments between image and lengthy text. Our approach consists of two key components: Local Image-Sentence Matching (LISM), which identifies corresponding pairs between image segments and descriptive sentences, and Token Similarity-based Learning (TSL), which efficiently propagates local element attention through these matched pairs. Evaluating GOAL on three new benchmarks for image-lengthy text retrieval, we demonstrate significant improvements over baseline CLIP fine-tuning, establishing a simple yet effective approach for adapting CLIP to detailed textual descriptions. Through extensive experiments, we show that our method's focus on local semantic alignment alongside global context leads to more nuanced and representative embeddings, particularly beneficial for tasks requiring fine-grained understanding of lengthy text descriptions.

Accurate and efficient lane detection in 3D space is essential for autonomous driving systems, where robust generalization is the foremost requirement for 3D lane detection algorithms. Considering the extensive variation in lane structures worldwide, achieving high generalization capacity is particularly challenging, as algorithms must accurately identify a wide variety of lane patterns worldwide. Traditional top-down approaches rely heavily on learning lane characteristics from training datasets, often struggling with lanes exhibiting previously unseen attributes. To address this generalization limitation, we propose a method that detects keypoints of lanes and subsequently predicts sequential connections between them to construct complete 3D lanes. Each key point is essential for maintaining lane continuity, and we predict multiple proposals per keypoint by allowing adjacent grids to predict the same keypoint using an offset mechanism. PointNMS is employed to eliminate overlapping proposal keypoints, reducing redundancy in the estimated BEV graph and minimizing computational overhead from connection estimations. Our model surpasses previous state-of-the-art methods on both the Apollo and OpenLane datasets, demonstrating superior F1 scores and a strong generalization capacity when models trained on OpenLane are evaluated on the Apollo dataset, compared to prior approaches.