CVPR 2025 Papers — Page 12
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
GIVEPose: Gradual Intra-class Variation Elimination for RGB-based Category-Level Object Pose Estimation
Ziqin Huang (Tsinghua University), Xiangyang Ji (Tsinghua University)
Object DetectionPose EstimationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: The GIVEPose framework is proposed, which achieves more accurate pose prediction by gradually eliminating intra-class variations in category-level object pose estimation.
GLane3D: Detecting Lanes with Graph of 3D Keypoints
Halil İbrahim Öztürk (Togg Trutek AI Team), Ozsel Kilinc (Togg Trutek AI Team)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: This paper proposes GLane3D, a keypoint-based 3D lane detection framework that utilizes a 3D keypoint graph with directed connections to achieve efficient lane extraction.
GLASS: Guided Latent Slot Diffusion for Object-Centric Learning
Krishnakant Singh (TU Darmstadt), Stefan Roth (hessian.AI)
Object DetectionSegmentationGenerationDiffusion modelImage
🎯 What it does: A slot attention-based object-centric learning framework called GLASS is designed and implemented, which learns slot representations in the generated image space and achieves precise decomposition and conditional generation of multiple objects in real scenes through semantic and instance guidance.
GliaNet: Adaptive Neural Network Structure Learning with Glia-Driven
Mengqiao Han (Northwest Agricultural and Forestry University), Xiabi Liu (Beijing Institute of Technology)
Object DetectionSegmentationRecurrent Neural NetworkTransformerImage
🎯 What it does: The Glia-Neuron (G-N) model and its GliaNet framework have been designed and implemented, simulating glial cells (Oligodendrocytes and Astrocytes) to dynamically select 'winning neurons' and adjust connections, achieving adaptive learning of network structure and parameters.
Global-Local Tree Search in VLMs for 3D Indoor Scene Generation
Wei Deng (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelTextPoint Cloud
🎯 What it does: A global-local tree search method is proposed, utilizing a large visual language model (VLM) to generate realistic 3D indoor scene layouts through hierarchical scene representation.
Glossy Object Reconstruction with Cost-effective Polarized Acquisition
Bojian Wu (Zhejiang University), Xiaowei Zhou (Zhejiang University)
RestorationGenerationNeural Radiance FieldImage
🎯 What it does: This paper proposes a low-cost, multi-view single-image polarization capture method combined with neural implicit surface rendering, achieving high-fidelity 3D reconstruction and reflection separation of smooth objects.
GLUS: Global-Local Reasoning Unified into A Single Large Language Model for Video Segmentation
Lang Lin (University of Illinois Urbana-Champaign), Yu-Xiong Wang (University of Illinois Urbana-Champaign)
SegmentationTransformerLarge Language ModelContrastive LearningVideoMultimodality
🎯 What it does: A unified global and local reasoning multimodal large language model (GLUS) is designed for reference video object segmentation.
GlyphMastero: A Glyph Encoder for High-Fidelity Scene Text Editing
Tong Wang (Meitu Inc.), Xiaolin Hu (Tsinghua University)
RecognitionGenerationDiffusion modelImageText
🎯 What it does: GlyphMastero is proposed, a trainable glyph encoder designed to enhance the glyph accuracy and style consistency in scene text editing.
GO-N3RDet: Geometry Optimized NeRF-enhanced 3D Object Detector
Zechuan Li (Hunan University), Naveed Akhtar (The University of Melbourne)
Object DetectionNeural Radiance FieldPoint Cloud
🎯 What it does: Proposes GO-N3RDet, a geometric optimization multi-view 3D object detection framework that integrates NeRF.
Go-with-the-Flow: Motion-Controllable Video Diffusion Models Using Real-Time Warped Noise
Ryan Burgert (Netflix Eyeline Studios), Ning Yu (Netflix)
GenerationData SynthesisDiffusion modelOptical FlowVideo
🎯 What it does: Introducing optical flow-driven noise warping in video diffusion models, structured noise is obtained by preprocessing training videos and directly used during the fine-tuning phase, enabling control over local objects, global camera motion, and motion transfer.
GOAL: Global-local Object Alignment Learning
Hyungyu Choi (Chung Ang University), Chanho Eom (Chung Ang University)
RetrievalTransformerSupervised Fine-TuningVision Language ModelContrastive LearningImageText
🎯 What it does: A fine-tuning framework called GOAL is proposed to refine CLIP, enhancing its local and global alignment capabilities in long text and image retrieval.
GoalFlow: Goal-Driven Flow Matching for Multimodal Trajectories Generation in End-to-End Autonomous Driving
Zebin Xing (University of Chinese Academy of Sciences), Wei Yin (Horizon Robotics)
GenerationAutonomous DrivingTransformerRectified FlowMultimodalityTime Series
🎯 What it does: This paper proposes GoalFlow, an end-to-end automatic driving multimodal trajectory generation method that generates high-quality trajectories using precise target point constraints.
Goku: Flow Based Video Generative Foundation Models
Shoufa Chen (University of Hong Kong), Xiaobing Liu (Bytedance)
GenerationData SynthesisTransformerFlow-based ModelRectified FlowImageVideoTextMultimodality
🎯 What it does: Goku is proposed—a Transformer architecture based on rectified flow that can simultaneously generate images and videos, supporting tasks such as text-to-image, text-to-video, and image-to-video.
Golden Cudgel Network for Real-Time Semantic Segmentation
Guoyu Yang (Shenzhen University), Yanzhong Wang (Shenzhen University)
SegmentationAutonomous DrivingConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: GCNet is proposed, which achieves self-expansion during the training phase using vertical multi-convolutions and horizontal multi-path structures, and re-parameterizes it into a single 3×3 convolution during the inference phase to achieve self-contraction, thus balancing the speed and accuracy of real-time semantic segmentation.
GoLF-NRT: Integrating Global Context and Local Geometry for Few-Shot View Synthesis
You Wang (Communication University of China), Zhan Ma (Nanjing University)
GenerationData SynthesisTransformerNeural Radiance FieldImage
🎯 What it does: We propose GoLF-NRT, a general neural rendering Transformer that integrates global context and local geometry to address the decline in rendering quality of NeRF under a limited number of viewpoints.
Good, Cheap, and Fast: Overfitted Image Compression with Wasserstein Distortion
Jona Ballé (New York University), Matthias Bauer (Google)
CompressionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Using C3 code streams for single image compression, the objective is changed to Wasserstein Distortion (WD), while incorporating shared randomness to enhance texture quality.
GPAvatar: High-fidelity Head Avatars by Learning Efficient Gaussian Projections
Wei-Qi Feng (Beihang University), Miao Wang (Beihang University)
RestorationGenerationGaussian SplattingVideo
🎯 What it does: A high-quality dynamic head avatar reconstruction method based on 3D Gaussian scattering, supporting real-time rendering and facial animation.
GPS as a Control Signal for Image Generation
Chao Feng (University of Michigan), Andrew Owens
GenerationData SynthesisPose EstimationDiffusion modelScore-based ModelNeural Radiance FieldImage
🎯 What it does: Trained a diffusion model based on GPS coordinates and text prompts to achieve image generation with geographic location awareness, and utilized this model to perform 3D reconstruction without pose markers through GPS-guided angle information.
GPVK-VL: Geometry-Preserving Virtual Keyframes for Visual Localization under Large Viewpoint Changes
Yunxuan Li (Northwestern University), Ying Wu (Northwestern University)
Pose EstimationRetrievalGaussian SplattingSimultaneous Localization and MappingMesh
🎯 What it does: This paper proposes GPVK-VL, which utilizes Geometrically Preserved Virtual Keyframes (GPVK) to cover the view space under large viewpoint changes, enhancing the robustness of structured visual localization.
Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning
Hasin Us Sami (University of California), Basak Guler (University of California)
Federated LearningAdversarial AttackTransformerImage
🎯 What it does: In federated learning, the authors propose PEFTLeak, which maliciously designs pre-trained models and adapter modules to recover user local fine-tuning data using only lightweight adapter gradients.
Gradient-Guided Annealing for Domain Generalization
Aristotelis Ballas (Harokopio University of Athens), Christos Diou (Harokopio University of Athens)
Domain AdaptationImage
🎯 What it does: This paper proposes an early parameter annealing strategy based on gradient consistency (Gradient-Guided Annealing, GGA), which enhances the model's generalization ability to unseen domains by searching for parameter points with similar gradients in the early stages of training.
GRAE-3DMOT: Geometry Relation-Aware Encoder for Online 3D Multi-Object Tracking
Hyunseop Kim (Chungnam National University), Yeong Jun Koh (Chungnam National University)
Object TrackingAutonomous DrivingGraph Neural NetworkPoint Cloud
🎯 What it does: This paper proposes an online 3D multi-object tracking method based on geometric relationships, GRAE-3DMOT, which aggregates detections using a relationship graph without distance thresholds and completes trajectory association.
Graph Neural Network Combining Event Stream and Periodic Aggregation for Low-Latency Event-based Vision
Manon Dampfhoffer (Univ. Grenoble Alpes), Christoph Posch (Prophesee)
Autonomous DrivingGraph Neural NetworkOptical FlowImageVideo
🎯 What it does: This paper proposes a new architecture that combines the HUGNet2 asynchronous event graph neural network with a Periodic Aggregator for real-time low-latency optical flow estimation from event cameras.
Graph-Embedded Structure-Aware Perceptual Hashing for Neural Network Protection and Piracy Detection
Ruiheng Liu (University of Science and Technology of China), Weiming Zhang (University of Science and Technology of China)
Graph Neural NetworkImageGraph
🎯 What it does: By constructing a directed acyclic graph of the model structure and combining parameter features, a structure-aware perceptual hashing method is proposed for copyright protection and piracy detection in neural networks.
GRAPHGPT-O: Synergistic Multimodal Comprehension and Generation on Graphs
Yi Fang (New York University), Jiawei Han (University of Illinois)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelMultimodalityGraph
🎯 What it does: Proposes the GRAPHGPT-O method, which simultaneously generates text and images on multimodal attribute graphs (MMAG).
GraphI2P: Image-to-Point Cloud Registration with Exploring Pattern of Correspondence via Graph Learning
Lin Bie (Tsinghua University), Yue Gao (Tsinghua University)
Pose EstimationDepth EstimationAutonomous DrivingGraph Neural NetworkImagePoint Cloud
🎯 What it does: A graph learning-based image-point cloud registration method called GraphI2P is proposed, which utilizes virtual spherical representation, distribution adaptive sampling, and graph correspondence selection to address the image-point cloud registration problem in asynchronous frames.
GraphMimic: Graph-to-Graphs Generative Modeling from Videos for Policy Learning
Guangyan Chen (Beijing Institute of Technology), Yufeng Yue (Beijing Institute of Technology)
GenerationRobotic IntelligenceRecurrent Neural NetworkGraph Neural NetworkTransformerVision Language ModelVideoGraph
🎯 What it does: A GraphMimic framework is designed, which abstracts video frames into objects and visual action graphs and pre-trains a graph-to-graph generative model to generate future graphs as policy guidance, enabling learning of robot manipulation from a small amount of labeled action data.
GREAT: Geometry-Intention Collaborative Inference for Open-Vocabulary 3D Object Affordance Grounding
Yawen Shao (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
Object DetectionRobotic IntelligenceConvolutional Neural NetworkLarge Language ModelSupervised Fine-TuningPrompt EngineeringImageMultimodalityPoint CloudChain-of-Thought
🎯 What it does: This paper proposes the GREAT framework, addressing the problem of 3D object affordance localization with open vocabulary. It utilizes geometric-intent collaborative reasoning and chain-of-thought (CoT) to extract implicit geometric properties and potential interaction intentions from interactive images. By combining point clouds and cross-modal image fusion, it achieves accurate localization of 3D object affordance regions.
Gromov-Wasserstein Problem with Cyclic Symmetry
Shoichiro Takeda (NTT Corporation), Yasunori Akagi (NTT Corporation)
OptimizationComputational EfficiencyPoint Cloud
🎯 What it does: Two gradient methods (C-CG, C-PG) are proposed to accelerate Gromov-Wasserstein computation using cyclic symmetry, and their effectiveness is validated on multi-scale data.
GroomLight: Hybrid Inverse Rendering for Relightable Human Hair Appearance Modeling
Yang Zheng (Stanford University), Thabo Beeler (Google)
Image
🎯 What it does: This paper proposes a hybrid inverse rendering method called GroomLight, which can reconstruct the re-lightable appearance of human hair from multi-view single-light (O)LAT images and achieve high-fidelity rendering under new lighting and viewpoints.
Ground-V: Teaching VLMs to Ground Complex Instructions in Pixels
Yongshuo Zong (University of Edinburgh), Onkar Dabeer (Amazon Web Services)
Object DetectionSegmentationKnowledge DistillationTransformerLarge Language ModelVision Language ModelImageText
🎯 What it does: This paper expands the instruction tracking and pixel-level localization capabilities of visual language models (VLMs) by constructing the Ground-V dataset (approximately 500K image-instruction-segmentation triplets) and an automatic generation process based on knowledge distillation, systematically addressing five major real-world challenges (multi-target, hierarchical granularity, part reference, reasoning, and false positive suppression).
Grounding 3D Object Affordance with Language Instructions, Visual Observations and Interactions
He Zhu (Zhejiang University), Yue Wang (Zhejiang University)
Object DetectionRobotic IntelligenceConvolutional Neural NetworkLarge Language ModelVision Language ModelImageTextMultimodalityPoint Cloud
🎯 What it does: This paper proposes a 3D object operability localization task based on three modalities: language instructions, images, and point clouds. It constructs the AGPIL multi-view dataset and designs an end-to-end LMAffordance3D model to achieve operability reasoning.
GroundingFace: Fine-grained Face Understanding via Pixel Grounding Multimodal Large Language Model
Yue Han (Zhejiang University), Yong Liu (Zhejiang University)
RecognitionSegmentationTransformerLarge Language ModelMixture of ExpertsVision Language ModelImageTextMultimodality
🎯 What it does: This paper presents the FacePlayGround-240K dataset aimed at fine-grained facial understanding, and develops the GroundingFace framework based on it, achieving pixel-level facial description, question answering, and segmentation.
GroupMamba: Efficient Group-Based Visual State Space Model
Abdelrahman Shaker (Mohamed Bin Zayed University of Artificial Intelligence), Fahad Shahbaz Khan (Mohamed Bin Zayed University of Artificial Intelligence)
ClassificationObject DetectionSegmentationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed the Modulated Group Mamba layer and a distillation-based training objective to build a stable and efficient visual SSM network.
GROVE: A Generalized Reward for Learning Open-Vocabulary Physical Skill
Jieming Cui (Peking University), Siyuan Huang
Robotic IntelligenceTransformerLarge Language ModelReinforcement LearningVision Language ModelVideoMultimodality
🎯 What it does: Proposes the GROVE framework, which combines LLM-generated physical constraints with VLM evaluation to achieve open vocabulary physical skill learning without manual rewards or task-specific demonstrations.
GS-2DGS: Geometrically Supervised 2DGS for Reflective Object Reconstruction
Jinguang Tong (Australian National University), Hongdong Li (Australian National University)
RestorationDepth EstimationGaussian SplattingPoint Cloud
🎯 What it does: A reflection object reconstruction method based on 2D Gaussian Splatting (GS-2DGS) is proposed, which combines single-view depth/normal supervision provided by a foundational model and deferred shading techniques to achieve high-quality geometric and material recovery.
GS-DiT: Advancing Video Generation with Dynamic 3D Gaussian Fields through Efficient Dense 3D Point Tracking
Weikang Bian (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelGaussian SplattingVideo
🎯 What it does: GS-DiT achieves 4D control of videos (multi-camera shooting, zooming, object editing, etc.) by combining dynamic 3D Gaussian fields with video diffusion transformers.
GuardSplat: Efficient and Robust Watermarking for 3D Gaussian Splatting
Zixuan Chen (Sun Yat-sen University), Xiaohua Xie (Sun Yat-sen University)
CompressionOptimizationGaussian SplattingPoint CloudMesh
🎯 What it does: This paper presents GuardSplat, a visual watermarking framework for 3D Gaussian Splatting (3DGS) assets to protect copyright.
GUI-Xplore: Empowering Generalizable GUI Agents with One Exploration
Yuchen Sun (Shanghai Jiao Tong University), Chongyang Zhang (Shanghai Jiao Tong University)
Large Language ModelAgentic AIVideoText
🎯 What it does: This paper presents the GUI-Xplore dataset and the Xplore-Agent framework, aimed at enhancing the GUI agent's capabilities across applications and tasks.
Guiding Human-Object Interactions with Rich Geometry and Relations
Mengqing Xue (South China University of Technology), Changxing Ding (South China University of Technology)
GenerationRobotic IntelligenceTransformerDiffusion modelVideo
🎯 What it does: A framework called ROG based on diffusion models is proposed, which generates character-object interactions that are consistent with text and geometrically more accurate by constructing an Interactive Distance Field (IDF) and a relational model.
Gyro-based Neural Single Image Deblurring
Heemin Yang (POSTECH), Sunghyun Cho (POSTECH)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes GyroDeblurNet, a method for single-frame image deblurring using smartphone gyroscope data.
h-Edit: Effective and Flexible Diffusion-Based Editing via Doob's h-Transform
Toan Nguyen (Deakin University), Thin Nguyen (Deakin University)
GenerationDiffusion modelImageBenchmark
🎯 What it does: A reverse bridge model based on Doob h-transform is proposed for training-free diffusion model image editing, named h-Edit;
H-MoRe: Learning Human-centric Motion Representation for Action Analysis
Zhanbo Huang (Michigan State University), Yu Kong (Michigan State University)
Representation LearningContrastive LearningOptical FlowVideo
🎯 What it does: This paper proposes H-MoRe, a fine-grained human motion representation method based on self-supervised learning, capable of directly extracting motion information from real videos.
H2ST: Hierarchical Two-Sample Tests for Continual Out-of-Distribution Detection
Yuhang Liu (University of Electronic Science and Technology of China), Yunhui Guo (University of Texas at Dallas)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: For open-world task incremental learning (TIL), a hierarchical two-sample test (H2ST) method is proposed that does not require a threshold, aimed at continuously detecting and identifying out-of-distribution (OOD) samples and their task IDs, thereby enabling automatic learning of new tasks.
Hallo3: Highly Dynamic and Realistic Portrait Image Animation with Video Diffusion Transformer
Jiahao Cui (Fudan University), Siyu Zhu (Fudan University)
GenerationTransformerDiffusion modelImageVideoTextMultimodalityAudio
🎯 What it does: The pre-trained DiT video generation model takes a single portrait, audio, and text input to generate high-quality portrait animation videos in multiple perspectives and dynamic backgrounds.
HalLoc: Token-level Localization of Hallucinations for Vision Language Models
Eunkyu Park (Seoul National University), Gunhee Kim (Seoul National University)
ClassificationObject DetectionGenerationTransformerVision Language ModelTextMultimodality
🎯 What it does: This paper presents a large-scale, fine-grained, word-level hallucination annotation dataset named HalLoc, consisting of 155K samples, and builds a lightweight HalLocalizer model based on this dataset, achieving parallel and probabilistic hallucination detection during the generation of visual language models.
Hand-held Object Reconstruction from RGB Video with Dynamic Interaction
Shijian Jiang, Jiming Chen
GenerationPose EstimationOptimizationVideo
🎯 What it does: This paper proposes a monocular RGB video handheld object dynamic reconstruction method based on a pre-trained 3D generative model prior and semantic consistency constraints.
Handling Spatial-Temporal Data Heterogeneity for Federated Continual Learning via Tail Anchor
Hao Yu (Southwestern University of Finance and Economics), Qiang Yang (Hong Kong University of Science and Technology)
Federated LearningTransformerContrastive LearningImage
🎯 What it does: The FedTA method is proposed, utilizing a frozen pre-trained ViT and incorporating learnable input augmentation and Tail Anchor to address catastrophic forgetting caused by spatial-temporal heterogeneity in federated continual learning.
HandOS: 3D Hand Reconstruction in One Stage
Xingyu Chen (Peking University), Lei Zhang (International Digital Economy Academy)
Object DetectionPose EstimationTransformerMesh
🎯 What it does: This paper proposes a one-stage end-to-end hand 3D mesh reconstruction framework called HandOS, which completes hand detection, pose estimation, and 3D mesh generation directly on a frozen detector, eliminating the need for a multi-stage process.
Hardware-Rasterized Ray-Based Gaussian Splatting
Samuel Rota Bulò (Meta Reality Labs), Peter Kontschieder (Meta Reality Labs)
Gaussian SplattingPoint CloudBenchmark
🎯 What it does: A hardware rasterization-based Ray-based 3D Gaussian splatting renderer is proposed, significantly improving the real-time rendering speed and quality of VR/MR scenes.
HarmonySet: A Comprehensive Dataset for Understanding Video-Music Semantic Alignment and Temporal Synchronization
Zitang Zhou (Tencent Inc), Fengyun Rao (Tencent Inc)
TransformerLarge Language ModelSupervised Fine-TuningVideoMultimodalityBenchmark
🎯 What it does: This paper proposes and releases the HarmonySet dataset, which contains 48,328 pairs of video-music matches, providing multi-dimensional annotations for each pair (rhythm synchronization, emotional consistency, thematic coherence, and cultural relevance). It also designs a human-machine collaborative annotation process, an LLM-based automatic description completion, and specialized evaluation benchmarks (HarmonySet-OE and HarmonySet-MC).
Harnessing Frequency Spectrum Insights for Image Copyright Protection Against Diffusion Models
Zhenguang Liu (Zhejiang University), Kui Ren (Zhejiang University)
GenerationData SynthesisSafty and PrivacyDiffusion modelAuto EncoderImage
🎯 What it does: CoprGuard is proposed, a watermarking framework based on spectral features, designed to detect and prevent image copyright infringement during the training and fine-tuning of diffusion models.
Harnessing Frozen Unimodal Encoders for Flexible Multimodal Alignment
Mayug Maniparambil (Dublin City University), Noel E. O'Connor (Dublin City University)
RetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A framework is proposed that only trains the projection layer, utilizing frozen unimodal visual and language encoders to achieve multimodal alignment.
Harnessing Global-Local Collaborative Adversarial Perturbation for Anti-Customization
Long Xu (Beihang University), Xianglong Liu (Beihang University)
GenerationAdversarial AttackDiffusion modelImage
🎯 What it does: The GoodAC framework is proposed, which generates strong robust adversarial perturbations by utilizing global feature correlation disruption and local facial attribute distortion to prevent Latent Diffusion Models (such as Stable Diffusion) from personalized customization.
Hash3D: Training-free Acceleration for 3D Generation
Xingyi Yang (National University of Singapore), Xinchao Wang (National University of Singapore)
GenerationComputational EfficiencyKnowledge DistillationDiffusion modelScore-based ModelGaussian SplattingImagePoint Cloud
🎯 What it does: This paper presents Hash3D, a hash-based feature reuse technique designed to accelerate Score Distillation Sampling (SDS) in 3D generation, significantly improving generation speed and viewpoint consistency without additional training.
HaWoR: World-Space Hand Motion Reconstruction from Egocentric Videos
Jinglei Zhang (Shanghai Jiao Tong University), Rolandos Alexandros Potamias (Imperial College London)
Pose EstimationTransformerSimultaneous Localization and MappingVideo
🎯 What it does: A method called HaWoR is proposed to recover the 3D motion of the hand in the world coordinate system from a single first-person view video.
Hazy Low-Quality Satellite Video Restoration Via Learning Optimal Joint Degradation Patterns and Continuous-Scale Super-Resolution Reconstruction
Ning Ni (Beijing Normal University), Libao Zhang (Beijing Normal University)
RestorationSuper ResolutionTransformerVideo
🎯 What it does: An end-to-end ODPNET is proposed for dehazing and achieving continuous scale low-quality satellite video super-resolution recovery.
HD-EPIC: A Highly-Detailed Egocentric Video Dataset
Toby Perrett (University of Bristol), Dima Damen (University of Bristol)
RecognitionObject TrackingSegmentationVision Language ModelSimultaneous Localization and MappingVideoMultimodalityBenchmarkAudio
🎯 What it does: A 41-hour first-person kitchen video dataset was constructed, providing fine-grained manual annotations, including recipe steps, ingredients and nutrition, fine-grained actions, 3D digital twins, object trajectories, and gaze, among other multimodal information;
Hearing Anywhere in Any Environment
Xiulong Liu (University of Washington), Ruohan Gao (University of Maryland)
TransformerAudio
🎯 What it does: A cross-room Room Impulse Response (RIR) prediction framework called XRIR is proposed, along with the construction of a large-scale high-fidelity dataset ACOUSTICROOMS, which supports accurate predictions in new rooms using only a minimal number of reference RIRs.
Hearing Hands: Generating Sounds from Physical Interactions in 3D Scenes
Yiming Dou (University of Michigan), Andrew Owens (University of Michigan)
GenerationData SynthesisRectified FlowGaussian SplattingVideoAudio
🎯 What it does: The study investigates how to utilize 3D hand trajectories and scene visual information to generate audio produced during the interaction of hands with 3D scenes.
HeatFormer: A Neural Optimizer for Multiview Human Mesh Recovery
Yuto Matsubara (Kyoto University), Ko Nishino (Kyoto University)
Pose EstimationOptimizationTransformerImageMesh
🎯 What it does: This paper proposes HeatFormer, a neural optimizer that utilizes multi-view static images for SMPL shape and pose recovery.
HEIE: MLLM-Based Hierarchical Explainable AIGC Image Implausibility Evaluator
Fan Yang (Tsinghua University), Guiguang Ding (Tsinghua University)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: This paper proposes the HEIE (Hierarchical Explainable Image Implausibility Evaluator) based on a multimodal large language model, which generates fine-grained heatmaps through an adaptive hierarchical implausibility mapper and outputs heatmaps, scores, and explanations simultaneously using a CoT-driven explainable triplet evaluator; it also constructs the Expl-AIGI-Eval dataset.
HELVIPAD: A Real-World Dataset for Omnidirectional Stereo Depth Estimation
Mehdi Zayene (École Polytechnique Fédérale de Lausanne), Alexandre Alahi (École Polytechnique Fédérale de Lausanne)
Depth EstimationImage
🎯 What it does: A real-world 360° binocular stereo depth estimation dataset, HELVIPAD, has been constructed, providing techniques for depth completion and model adaptation.
HeMoRa: Unsupervised Heuristic Consensus Sampling for Robust Point Cloud Registration
Shaocheng Yan (Wuhan University), Jiayuan Li (Wuhan University)
OptimizationReinforcement LearningPoint Cloud
🎯 What it does: An unsupervised HeMoRa framework is proposed, utilizing HeGen to generate sampling probabilities and features for corresponding point clouds, enhancing the quality of Consensus-set sampling, thereby achieving more robust corresponding point cloud registration.
HERA: Hybrid Explicit Representation for Ultra-Realistic Head Avatars
Hongrui Cai (University of Science and Technology of China), Juyong Zhang (University of Science and Technology of China)
GenerationData SynthesisComputational EfficiencyGaussian SplattingVideoPoint CloudMesh
🎯 What it does: This paper proposes a HERA hybrid explicit representation method for efficiently constructing and real-time rendering of ultra-realistic head avatars.
Heterogeneous Skeleton-Based Action Representation Learning
Hongsong Wang (Southeast University), Jie Gui (Southeast University)
RecognitionRetrievalRepresentation LearningTransformerPrompt EngineeringContrastive LearningVideo
🎯 What it does: A unified framework is proposed for learning skeletal action representations from different sources, dimensions, and topological structures. It first maps 2D skeletons to 3D, generates a unified skeleton using learnable skeleton-specific prompts, and incorporates semantic motion encoding. Subsequently, it utilizes a Transformer for self-supervised unified representation learning.
Hiding Images in Diffusion Models by Editing Learned Score Functions
Haoyu Chen (City University of Hong Kong), Kede Ma (City University of Hong Kong)
GenerationData SynthesisDiffusion modelScore-based ModelImage
🎯 What it does: A method for hiding images in diffusion models is proposed, which achieves implicit embedding and extraction of secret images by editing the learned scoring function at specific time steps during the reverse diffusion process.
Hierarchical Adaptive Filtering Network for Text Image Specular Highlight Removal
Zhi Jiang (Wuhan University), Chunxia Xiao (Wuhan University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Hierarchical Adaptive Filtering Network (HAFNet) for removing large areas of specular highlights from text images.
Hierarchical Compact Clustering Attention (COCA) for Unsupervised Object-Centric Learning
Can Kucuksozen (Koc University), Yucel Yemez (Koc University)
Object DetectionSegmentationTransformerImage
🎯 What it does: Proposes COCA layer and COCA-Net, achieving unsupervised object perception and segmentation through hierarchical clustering attention.
Hierarchical Features Matter: A Deep Exploration of Progressive Parameterization Method for Dataset Distillation
Xinhao Zhong (Harbin Institute of Technology), Shu-Tao Xia (Tsinghua University)
Data SynthesisKnowledge DistillationGenerative Adversarial NetworkImage
🎯 What it does: A hierarchical parameterized dataset distillation method H-PD is proposed, which improves data distillation performance at extreme compression rates by optimizing the generation of synthetic datasets layer by layer in the multi-layer feature space of GAN.
Hierarchical Flow Diffusion for Efficient Frame Interpolation
Yang Hai (Insta360 Research), Yinlin Hu (MagicLeap)
GenerationData SynthesisComputational EfficiencyDiffusion modelOptical FlowVideo
🎯 What it does: A hierarchical optical flow diffusion model is proposed for video frame interpolation, explicitly modeling optical flow and progressively denoising at multiple scales, achieving significant improvements in both efficiency and accuracy.
Hierarchical Gaussian Mixture Model Splatting for Efficient and Part Controllable 3D Generation
Qitong Yang (Xidian University), Ajmal Mian (University of Western Australia)
GenerationData SynthesisDiffusion modelGaussian SplattingPoint CloudMesh
🎯 What it does: A 3D Gaussian splatting framework based on Hierarchical Gaussian Mixture Model (HGMM) and tree topology diffusion Mamba is proposed, achieving efficient and controllable 3D generation from text/images.
Hierarchical Knowledge Prompt Tuning for Multi-task Test-Time Adaptation
Qiang Zhang (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
Domain AdaptationKnowledge DistillationTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: A multi-task testing adaptation framework named HKPT is designed, enabling CLIP to adapt simultaneously across multiple target domains.
HierarQ: Task-Aware Hierarchical Q-Former for Enhanced Video Understanding
Shehreen Azad (Center for Research in Computer Vision University of Central Florida), Yogesh Singh Rawat (Microsoft Research)
RecognitionGenerationTransformerLarge Language ModelVideoText
🎯 What it does: The HierarQ framework is designed to process videos frame by frame in an autoregressive manner, and it achieves task-aware hierarchical Q-Transformer through a dual-stream (entity stream and scene stream) feature modulator, completing long video understanding, question answering, and subtitle tasks.
HiFi-Portrait: Zero-shot Identity-preserved Portrait Generation with High-fidelity Multi-face Fusion
Yifang Xu (Nanjing University), Sidan Du (Nanjing University)
GenerationTransformerDiffusion modelImage
🎯 What it does: The HiFi-Portrait method is proposed to achieve zero-shot, high-fidelity identity-preserving portrait generation.
High Dynamic Range Video Compression: A Large-Scale Benchmark Dataset and A Learned Bit-depth Scalable Compression Algorithm
Zhaoyi Tian (Shanghai University), Liquan Shen (Shanghai University)
CompressionOptical FlowVideoBenchmark
🎯 What it does: The first large-scale HDR video dataset HDRVD2K has been constructed, and a learning-based bit-depth scalable video compression network LBSVC has been proposed.
High Temporal Consistency through Semantic Similarity Propagation in Semi-Supervised Video Semantic Segmentation for Autonomous Flight
Cédric Vincent (Institut Polytechnique de Paris), Henri Meeß (Fraunhofer IVI)
SegmentationAutonomous DrivingKnowledge DistillationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: A lightweight video semantic segmentation framework SSP (Semantic Similarity Propagation) is proposed, which enhances the temporal consistency of drone videos by performing linear interpolation between the current image model prediction and the prediction aligned with the previous frame for each frame.
High-fidelity 3D Object Generation from Single Image with RGBN-Volume Gaussian Reconstruction Model
Yiyang Shen (Zhejiang University), Tianjia Shao (Zhejiang University)
GenerationData SynthesisDiffusion modelImagePoint Cloud
🎯 What it does: A GS-RGBN model based on RGBN voxel-Gaussian reconstruction is proposed, which quickly generates high-fidelity 3D objects from a single view image.
High-Fidelity Lightweight Mesh Reconstruction from Point Clouds
Chen Zhang (Huazhong University of Science and Technology), Wenbing Tao (Huazhong University of Science and Technology)
GenerationOptimizationComputational EfficiencyPoint CloudMesh
🎯 What it does: A curvature-adaptive mesh reconstruction method is proposed, capable of extracting high-fidelity, lightweight meshes from implicit surfaces learned from point clouds.
High-Fidelity Relightable Monocular Portrait Animation with Lighting-Controllable Video Diffusion Model
Mingtao Guo (Sichuan University), Yanli Liu (Sichuan University)
GenerationData SynthesisDiffusion modelImageVideo
🎯 What it does: A controllable lighting video diffusion model LCVD is proposed, achieving high-fidelity re-lighting portrait animation from a single reference portrait under given lighting and target video.
High-quality Point Cloud Oriented Normal Estimation via Hybrid Angular and Euclidean Distance Encoding
Yuanqi Li (Nanjing University), Yanwen Guo (Nanjing University)
Graph Neural NetworkPoint Cloud
🎯 What it does: A point cloud normal estimation framework based on hybrid angle and Euclidean distance encoding (HAE) is proposed.
Higher-Order Ratio Cycles for Fast and Globally Optimal Shape Matching
Paul Roetzer (University of Bonn), Florian Bernard (MCML)
OptimizationPoint CloudMesh
🎯 What it does: This paper proposes the use of minimum ratio cycles to solve shape matching problems on product graphs, including 2D-3D, 3D-3D, and shape-graph matching, overcoming the issue of short path bias in minimum cost cycles.
HIIF: Hierarchical Encoding based Implicit Image Function for Continuous Super-resolution
Yuxuan Jiang (University of Bristol), David Bull (University of Bristol)
RestorationSuper ResolutionImage
🎯 What it does: A new Hierarchical Implicit Image Function (HIIF) based on hierarchical encoding is proposed for continuous image super-resolution, capable of capturing details at multiple scales.
HiLoTs: High-Low Temporal Sensitive Representation Learning for Semi-Supervised LiDAR Segmentation in Autonomous Driving
R.D. Lin (Xi'an Jiaotong University), Fei Wang (Zhejiang University)
SegmentationAutonomous DrivingRepresentation LearningTransformerPoint Cloud
🎯 What it does: What was done: A semi-supervised learning-based LiDAR point cloud semantic segmentation framework called HiLoTs was proposed, which significantly improves segmentation performance by utilizing high and low time-sensitive flows along with the Mean Teacher mechanism.
HiMoR: Monocular Deformable Gaussian Reconstruction with Hierarchical Motion Representation
Yiming Liang (Waseda University), Yuta Kikuchi (Preferred Networks, Inc.)
RestorationObject DetectionGaussian SplattingVideo
🎯 What it does: A hierarchical motion representation HiMoR is designed, and dynamic 3D reconstruction from monocular video is achieved using 3D Gaussian Splatting.
HiPART: Hierarchical Pose AutoRegressive Transformer for Occluded 3D Human Pose Estimation
Hongwei Zheng (Shanghai Jiao Tong University), Hongkai Xiong (Shanghai Jiao Tong University)
Pose EstimationTransformerImageVideo
🎯 What it does: This paper proposes the HiPART model, which generates hierarchical dense 2D poses in a self-regressive manner for 2D-to-3D human pose estimation, thereby addressing the occlusion problem.
HiRes-LLaVA: Restoring Fragmentation Input in High-Resolution Large Vision-Language Models
Runhui Huang (Shenzhen campus of Sun Yat-sen University), Xiaodan Liang (Shenzhen campus of Sun Yat-sen University)
RestorationCompressionTransformerVision Language ModelImageTextMultimodality
🎯 What it does: A new framework named HiRes-LLaVA is designed to address the fragmentation issue of high-resolution visual input slices, achieving efficient information recovery and visual token compression through the SliceRestore Adapter and Self-Mining Sampler.
HistoFS: Non-IID Histopathologic Whole Slide Image Classification via Federated Style Transfer with RoI-Preserving
Farchan Hakim Raswa (National Central University), Jia-Ching Wang (National Central University)
ClassificationFederated LearningImageBiomedical Data
🎯 What it does: A framework called HistoFS is proposed for classifying whole slide images (WSI) in a federated learning environment, addressing non-independent and identically distributed data by utilizing pseudo-bag style transfer and a realism module to achieve multi-modal style enhancement and ROI preservation.
HMAR: Efficient Hierarchical Masked Auto-Regressive Image Generation
Hermann Kumbong (Stanford University), David W. Romero (NVIDIA)
GenerationAuto EncoderImage
🎯 What it does: The Hierarchical Masked Auto-Regressive (HMAR) model is proposed to generate high-quality images at multiple scales, balancing speed and quality.
HoGS: Unified Near and Far Object Reconstruction via Homogeneous Gaussian Splatting
Xinpeng Liu (University of Osaka), Yasuyuki Matsushita (Microsoft Research Asia)
RestorationGenerationGaussian SplattingPoint Cloud
🎯 What it does: A 3D Gaussian rendering method based on homogeneous coordinates, HoGS, is proposed, achieving unified reconstruction and real-time rendering of both near and far scenes.
HOIGen-1M: A Large-scale Dataset for Human-Object Interaction Video Generation
Kun Liu (JD Explore Academy), Wu Liu (University of Science and Technology of China)
GenerationData SynthesisLarge Language ModelMixture of ExpertsDiffusion modelOptical FlowVideoTextMultimodality
🎯 What it does: A large-scale human-object interaction video generation dataset called HOIGen-1M is proposed, and a multi-modal expert mixture strategy called MoME is designed to generate high-quality subtitles. Furthermore, two evaluation metrics, CoarseHOIScore and FineHOIScore, are introduced.
HOIGPT: Learning Long-Sequence Hand-Object Interaction with Language Models
Mingzhen Huang (Meta), Hao Tang (Meta)
GenerationData SynthesisRobotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningTextMultimodalitySequential
🎯 What it does: A unified HOIGPT model has been developed, capable of generating 3D hand-object interaction sequences based on text, as well as converting hand-object interaction sequences into natural language descriptions, achieving bidirectional generation and understanding.
Holmes-VAU: Towards Long-term Video Anomaly Understanding at Any Granularity
Huaxin Zhang (Huazhong University of Science and Technology), Nong Sang (Huazhong University of Science and Technology)
Anomaly DetectionTransformerLarge Language ModelVision Language ModelVideoMultimodality
🎯 What it does: A multi-granularity video anomaly understanding framework called Holmes-VAU is proposed, aiming to capture both short-term and long-term anomalies.
HomoGen: Enhanced Video Inpainting via Homography Propagation and Diffusion
Ding Ding (Wuhan University), Zhenzhong Chen (Wuhan University)
RestorationGenerationDiffusion modelVideo
🎯 What it does: We propose HomoGen, a two-stage video inpainting framework: first, it propagates pixels from adjacent frames as priors through homography transformation, and then fills in the missing areas using a video diffusion model.
Homogeneous Dynamics Space for Heterogeneous Humans
Xinpeng Liu (Shanghai Jiao Tong University), Yong-Lu Li (Shanghai Jiao Tong University)
Pose EstimationTransformerReinforcement LearningAuto EncoderContrastive LearningMultimodality
🎯 What it does: A unified latent space HDyS has been constructed to map multi-source kinematic and dynamic data to the same representation, thereby learning human dynamics.
HOP: Heterogeneous Topology-based Multimodal Entanglement for Co-Speech Gesture Generation
Hongye Cheng (Northwest A&F University), Yanwei Fu (Fudan University)
GenerationData SynthesisGraph Neural NetworkLarge Language ModelGenerative Adversarial NetworkTextMultimodalityAudio
🎯 What it does: A topology-based heterogeneous multimodal coupling model HOP is proposed to model the coupling relationships among text, audio, and gestures, thereby generating gestures that are synchronized and coherent with speech rhythm and semantics.
Horizon-GS: Unified 3D Gaussian Splatting for Large-Scale Aerial-to-Ground Scenes
Lihan Jiang (University of Science and Technology of China), Bo Dai (University of Hong Kong)
RestorationGenerationGaussian SplattingImage
🎯 What it does: This paper presents Horizon-GS, a unified three-dimensional Gaussian expansion method for simultaneously reconstructing and rendering large-scale scenes from aerial and street view perspectives.
HORP: Human-Object Relation Priors Guided HOI Detection
Pei Geng (Nanjing University of Science and Technology), Shanshan Zhang (Nanjing University of Science and Technology)
RecognitionObject DetectionTransformerVision Language ModelImageMultimodality
🎯 What it does: A human-object interaction (HOI) detection method based on human-object relationship priors (HORP) is proposed, which enhances the recognition capability of the 'no interaction' category by integrating 3D position priors and human gaze area priors into multimodal queries.
HOT: Hadamard-based Optimized Training
Seonggon Kim (POSTECH), Eunhyeok Park (POSTECH)
ClassificationObject DetectionSegmentationCompressionOptimizationTransformerSupervised Fine-TuningImageText
🎯 What it does: To optimize the matrix multiplication in backpropagation during deep learning training, a Hadamard-based Optimized Training (HOT) scheme is proposed.
HOT3D: Hand and Object Tracking in 3D from Egocentric Multi-View Videos
Prithviraj Banerjee (Meta Reality Labs), Tomas Hodan (Meta Reality Labs)
Object TrackingPose EstimationOptical FlowVideoPoint Cloud
🎯 What it does: The authors constructed the HOT3D dataset, collecting 833 minutes and 1.5 million frames of multi-view first-person videos from 19 subjects using the Aria and Quest 3 head-mounted devices, and provided annotations for hand and object 3D poses, point clouds, gaze, and more.