CVPR 2026 Papers — Page 16
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional Evaluation
Rang Li (Peking University), Fuli Luo (Renmin University of China)
Large Language ModelVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper proposes the GroundingME benchmark to evaluate the real-world capabilities of multimodal large language models (MLLMs) in visual localization tasks from multiple dimensions.
GroundVTS: Visual Token Sampling in Multimodal Large Language Models for Video Temporal Grounding
Rong Fan (Newcapec AI Research), Zhao Yang (Newcapec AI Research)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: Propose a query-guided visual token sampling framework called GroundVTS within video large language models (Vid-LLM) to improve the accuracy of video temporal grounding (VTG).
Group Diffusion: Enhancing Image Generation by Unlocking Cross-Sample Collaboration
Sicheng Mo (University of California, Los Angeles), Yuheng Li (Adobe Research)
GenerationTransformerDiffusion modelImage
🎯 What it does: This paper proposes Group Diffusion, which achieves cross-sample collaborative denoising by allowing multiple images under the same condition to share attention during inference;
Group Editing: Edit Multiple Images in One Go
Yue Ma (Hong Kong University Of Science And Technology), Qifeng Chen (Hong Kong University Of Science And Technology)
Image TranslationGenerationTransformerVision Language ModelDiffusion modelImageTextMultimodalityBenchmark
🎯 What it does: Propose a framework named GroupEditing for achieving consistent and unified edits across a group of related images (e.g., different views or poses of the same object).
GROW: Watermark Generation with Progressive Guidance for Diffusion Models
Pengcheng Luo (Peking University), Jie Zhou (Tencent Inc.)
GenerationSafty and PrivacyDiffusion modelImage
🎯 What it does: Propose a training-agnostic digital watermarking method called GROW, which embeds watermarks in the generation process of diffusion models using progressive guidance.
GRPO-Guard: Mitigating Implicit Over-Optimization in Flow Matching via Regulated Clipping
Jing Wang (Shenzhen Campus of Sun Yat-Sen University), Xiaodan Liang (UCAS)
GenerationOptimizationDiffusion modelFlow-based ModelImageBenchmarkStochastic Differential Equation
🎯 What it does: This paper proposes an improved GRPO framework, GRPO-Guard, to address the implicit over-optimization (reward hacking) problem in flow matching models, while stabilizing training while maintaining or enhancing generation quality.
GS-ASM: 2DGS-Supervised Active Stereo Matching
Zhengling Wu (Hangzhou Dianzi University), Chenggang Yan (Hangzhou Dianzi University)
Depth EstimationConvolutional Neural NetworkTransformerGaussian SplattingOptical FlowImage
🎯 What it does: Propose the GS-ASM framework, which utilizes active 2D Gaussian Splatting to generate proxy labels, enabling supervised training of active stereo matching networks without real depth labels.
GS-CLIP: Zero-shot 3D Anomaly Detection by Geometry-Aware Prompt and Synergistic View Representation Learning
Zehao Deng (Soochow University), Yan Wang (Tsinghua University)
Anomaly DetectionConvolutional Neural NetworkTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: Designed the GS-CLIP framework, enhancing CLIP's zero-shot 3D anomaly detection capability through geometric-aware text prompts and synergistic perspective representation learning.
GS^2: Graph-based Spatial Distribution Optimization for Compact 3D Gaussian Splatting
Xianben Yang (Beijing Jiaotong University), Haibin Ling (Westlake University)
CompressionOptimizationGraph Neural NetworkGaussian SplattingPoint Cloud
🎯 What it does: Propose a graph-based spatial distribution optimization framework GSˆ2, which can significantly compress the number of 3D Gaussian points while improving the quality of novel view synthesis.
GSNR: Graph Smooth Null-Space Representation for Inverse Problems
Romario Gualdrón-Hurtado (Universidad Industrial de Santander), Henry Arguello (Universidad Industrial de Santander)
RestorationSuper ResolutionGraph Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes the Graph-Smooth Null-Space Representation (GSNR), which explicitly constrains the zero-space components invisible to sensors in inverse problems.
GSV2X: Geometry-Aware Uncertainty Modeling and Orthogonal Fusion for Robust Roadside Perception
Jianqiang Xu (Nanjing University of Science and Technology), Yazhou Yao (Nanjing University of Science and Technology)
Autonomous DrivingConvolutional Neural NetworkGaussian SplattingImageMultimodalityPoint Cloud
🎯 What it does: Propose the GSV2X framework to achieve roadside multi-perspective multi-modal (camera + LiDAR) perception, addressing spatial uncertainty and modality imbalance issues.
GT-SVJ: Generative-Transformer-Based Self-Supervised Video Judge For Efficient Video Reward Modeling
Shivanshu Shekhar (University of Illinois Urbana Champaign), Tong Zhang (University of Illinois Urbana Champaign)
Reinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningContrastive LearningVideoText
🎯 What it does: Propose GT-SVJ by converting video generation models into energy models for self-supervised contrastive learning, then fine-tuning to obtain an efficient temporal-aware video reward model.
GThinker: Towards General Multimodal Reasoning via Cue-Guided Rethinking
Yufei Zhan (Chinese Academy of Sciences), Jinqiao Wang (Chinese Academy of Sciences)
Large Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose a novel multimodal large language model called GThinker to address visual inertia issues;
GTR-Turbo: Merged Checkpoint is Secretly a Free Teacher for Agentic VLM Training
Tong Wei (Tsinghua University), Deheng Ye (Tencent Hunyuan)
Knowledge DistillationSupervised Fine-TuningReinforcement LearningAgentic AIVision Language ModelMultimodality
🎯 What it does: This paper proposes a method called GTR-Turbo, which utilizes historical checkpoints fusion to generate a 'free' teacher model in multi-round vision-language model reinforcement learning;
Guardians of the Hair: Rescuing Soft Boundaries in Depth, Stereo, and Novel Views
Xiang Zhang (ETH Zurich), Studios blank
Image HarmonizationGenerationDepth EstimationConvolutional Neural NetworkTransformerNeural Radiance FieldAuto EncoderOptical FlowImageVideo
🎯 What it does: Propose the HairGuard framework, significantly enhancing the preservation and reconstruction of soft boundary details (e.g., fine hair strands) in depth estimation, stereo conversion, and novel view synthesis.
GuardTrace-VL: Detecting Unsafe Multimodel Reasoning via Iterative Safety Supervision
Yuxiao Xiang (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
Safty and PrivacyTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityBenchmark
🎯 What it does: Propose GuardTrace-VL, a security auditor specifically designed to monitor unsafe content throughout the entire process (question-thinking-answer) of multimodal reasoning models (MLRM).
GUI-CEval: A Hierarchical and Comprehensive Chinese Benchmark for Mobile GUI Agents
Yang Li (Xiaomi Corporation), Ying Huang (Xiaomi Corporation)
Prompt EngineeringVision Language ModelVision-Language-Action ModelMultimodalityBenchmark
🎯 What it does: Designed and released the GUI-CEval benchmark, covering 201 mainstream Chinese mobile applications, including 4,028 Agent tasks and 4,194 multimodal QA tasks, divided into two layers: basic and application.
GUI-SAGE: Enhancing GUI Automation with Self-Explanatory Learning
Fei Tang (Zhejiang University), Yueting Zhuang (Zhejiang University)
Explainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelMultimodalityBenchmark
🎯 What it does: Proposed the GUI-SAGE framework, which generates self-explaining, in-distribution reasoning trajectories aligned with target actions to address the zero-advantage trap encountered by traditional RL in GUI automation;
GUIDE: A Benchmark for Understanding and Assisting Users in Open-Ended GUI Tasks
Saelyne Yang (KAIST), Juho Kim (KAIST)
Large Language ModelPrompt EngineeringVision-Language-Action ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: Proposed the GUIDE benchmark, collected 67.5 hours of screen recordings and think-aloud audio from 120 beginners using 10 common software (Photoshop, Figma, PowerPoint, etc.), constructed three tasks: behavior state detection, intent prediction, and help prediction, and conducted zero-shot evaluation on eight multimodal large language models.
GuideFlow: Constraint-Guided Flow Matching for Planning in End-to-End Autonomous Driving
Lin Liu (Beijing Jiaotong University), Yadan Luo (Beijing Academy of Artificial Intelligence)
Autonomous DrivingFlow-based ModelRectified Flow
🎯 What it does: Proposes an end-to-end planning framework called GuideFlow based on flow matching, which can directly incorporate safety and physical constraints during the generation process, addressing the issues of multi-modal mode collapse and insufficient generation safety in traditional imitation learning.
Guiding a Diffusion Model by Swapping Its Tokens
Weijia Zhang (Shanghai Jiao Tong University), Chao Ma (Shanghai Jiao Tong University)
GenerationTransformerDiffusion modelImage
🎯 What it does: Proposed an unconditional guidance method called Self-Swap Guidance (SSG) during diffusion model inference, generating perturbations by swapping the least semantically similar tokens in the model to guide sampling and enhance image quality.
Guiding a Diffusion Transformer with the Internal Dynamics of Itself
Xingyu Zhou (University of Electronic Science and Technology of China), Shuhang Gu (University of Electronic Science and Technology of China)
GenerationComputational EfficiencyTransformerDiffusion modelImage
🎯 What it does: Proposes an Internal Guidance strategy, introducing intermediate layer supervision during the training of diffusion transformers and leveraging intermediate layer outputs during sampling to extrapolate the final output, thereby enhancing generation quality.
Guiding Diffusion Models with Fine-Grained Conditions and Semantics-Preserving Sampling for One-Shot Federated Learning
Xiaojun Deng (Sun Yat-sen University), Zibin Zheng (Sun Yat-sen University)
Data SynthesisFederated LearningDiffusion modelImage
🎯 What it does: This work proposes a framework named Espresso to enhance the semantic fidelity and diversity of synthetic data in one-shot federated learning (OSFL), thereby improving the performance of the global model;
Guiding Diffusion Models with Semantically Degraded Conditions
Shilong Han (National University of Defense Technology), Hongxia Wang (National University of Defense Technology)
GenerationTransformerDiffusion modelImageMultimodality
🎯 What it does: Propose Condition-Degradation Guidance (CDG), which improves CFG by replacing empty prompts with negative conditions based on semantic degradation.
Guiding Diffusion-based Reconstruction with Contrastive Signals for Balanced Visual Representation
Boyu Han (Institute of Computing Technology, Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
Representation LearningTransformerVision Language ModelDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: Propose the Diffusion Contrastive Reconstruction (DCR) framework, which integrates contrastive learning with diffusion reconstruction to enhance the discriminative ability (D-Ability) and detail perception ability (P-Ability) of the CLIP visual encoder.
Guiding Token-Sparse Diffusion Models
Felix Krause (LMU Munich), Björn Ommer (LMU Munich)
GenerationComputational EfficiencyTransformerDiffusion modelFlow-based ModelImage
🎯 What it does: Propose Sparse Guidance (SG), a method for sparse diffusion models that utilizes token sparsity during the inference phase as a guidance signal without requiring additional fine-tuning.
GVIS: Generative Vector Image Steganography
Zihao Xu (Changchun University), Chuan Zhang (Beijing Institute of Technology)
GenerationSafty and PrivacyDiffusion modelImage
🎯 What it does: Proposed a new generative vector image steganography framework called GVIS, which utilizes diffusion models to generate raster images and vectorizes them into vector images to achieve high capacity and high accuracy information hiding.
Gyro-based Deep Video Deblurring
Jaesung Rim (POSTECH), Sunghyun Cho (POSTECH)
RestorationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: Proposed a gyroscope-based deep video deblurring framework called GyroDVD, which constructs a pixel-level blur kernel by combining rotational and translational motion and restores clear videos through a deep network.
H-Sets: Hessian-Guided Discovery of Set-Level Feature Interactions in Image Classifiers
Ayushi Mehrotra (California Institute of Technology), Nidhi Rastogi (Rochester Institute of Technology)
ClassificationExplainability and InterpretabilityImage
🎯 What it does: Propose a two-stage framework H-Sets for detecting and attributing high-order feature interactions in image classifiers
H^2A^2: Homogeneity-Aware and Heterogeneity-Aware Feature Perception for Unified Indoor 3D Object Detection
Tao Xie (Harbin Institute of Technology), Ruifeng Li (Harbin Institute of Technology)
Object DetectionConvolutional Neural NetworkPoint Cloud
🎯 What it does: Proposes a unified indoor 3D object detection framework HA, which can simultaneously capture homogeneous structural features across scenes and scene-specific heterogeneous features, and achieve more robust cross-scenario learning through structural-aware convolution kernel selection and gradient homogenization.
H2-Surv: Hierarchical Hyperbolic Multimodal Representation Learning for Survival Prediction
Jiaqi Yang (University of Nottingham), Xiaohan Xing (Stanford University)
ClassificationRepresentation LearningContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Propose a hierarchical hyperbolic multimodal learning framework, H2-Surv, for cancer survival prediction, integrating pathological images and genomic data.
HAD: Hallucination-Aware Diffusion Priors for 3D Reconstruction
Xi Liu (Amazon AWS), Laurent Guigues (Amazon AWS)
GenerationDiffusion modelScore-based ModelNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Proposed the HAD method, which combines diffusion priors with hallucination score prediction, using a multi-view encoder to evaluate hallucinations in enhanced views and thereby mask unreliable pixels in 3D Gaussian Splatting reconstruction, improving the quality of 3D reconstruction under sparse views.
HAD: Heterogeneity-Aware Distillation for Lifelong Heterogeneous Learning
Xuerui Zhang (Southern University of Science and Technology), Yu Zhang (Technical University of Munich)
SegmentationDepth EstimationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new scenario called Lifelong Heterogeneous Learning (LHL) and designs a method named Lifelong Heterogeneous Distillation (HAD) for dense prediction tasks, addressing the problem of catastrophic forgetting when continuously learning tasks with different output structures.
HalluGen: Synthesizing Realistic and Controllable Hallucinations for Evaluating Image Restoration
Seunghoi Kim (University College London), Daniel C. Alexander (University College London)
RestorationGenerationData SynthesisDiffusion modelBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose a controllable diffusion model framework called HalluGen for synthesizing realistic and locatable hallucinations in image restoration, and build the first large-scale hallucination dataset;
HamiPose: Hamiltonian Optimization for Unsupervised Domain Adaptive Pose Estimation
Jiawen Li (East China Normal University), Aimin Zhou (East China Normal University)
Pose EstimationDomain AdaptationOptimizationImageOrdinary Differential Equation
🎯 What it does: Propose an unsupervised domain adaptation pose estimation framework based on Hamiltonian dynamics called HamiPose, aiming to alleviate optimization instability caused by gradient conflicts between source supervision and target consistency.
HAMMER: Harnessing MLLMs via Cross-Modal Integration for Intention-Driven 3D Affordance Grounding
Lei Yao (Hong Kong Polytechnic University), Lap-Pui Chau (Hong Kong Polytechnic University)
Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImagePoint Cloud
🎯 What it does: This paper proposes the HAMMER framework, which leverages multimodal large language models (MLLM) to extract intent information from interactive images and combines 3D point clouds to achieve precise 3D affordance localization.
HandDreamer: Zero-Shot Text to 3D Hand Model Generation using Corrective Hand Shape Guidance
Green Rosh (Samsung R&D Institute), Pawan Prasad B H (Samsung R&D Institute)
GenerationData SynthesisTransformerDiffusion modelScore-based ModelNeural Radiance FieldGaussian SplattingTextMesh
🎯 What it does: Proposed the HandDreamer method, achieving zero-shot generation of high-fidelity, view-consistent 3D hand models from text prompts.
HandVQA: Diagnosing and Improving Fine-Grained Spatial Reasoning about Hands in Vision-Language Models
MD Khalequzzaman Chowdhury Sayem, Seungryul Baek (UNIST)
Pose EstimationSupervised Fine-TuningPrompt EngineeringVision Language ModelImageMultimodalityPoint CloudBenchmark
🎯 What it does: Designed and released the HandVQA large-scale diagnostic benchmark, generating 1.6M multiple-choice questions through automated 3D hand keypoint coordinates to evaluate VLM's performance in fine-grained hand spatial reasoning (angles, distances, relative positions), and validate the benchmark's training for zero-shot transfer effectiveness in gesture recognition and hand-object interaction tasks.
HandWorld: Hand-Centric Unified Video Action Generation
Zhihao Sun (Institute of Trustworthy Embodied AI, Fudan University), Zuxuan Wu (Institute of Trustworthy Embodied AI, Fudan University)
GenerationData SynthesisTransformerVision-Language-Action ModelDiffusion modelFlow-based ModelAuto EncoderVideo
🎯 What it does: Propose HandWorld, a unified generation framework for jointly generating egocentric videos and hand actions.
HandX: Scaling Bimanual Motion and Interaction Generation
Zimu Zhang (University of Illinois Urbana Champaign), Liang-Yan Gui (University of Illinois Urbana Champaign)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelFlow-based ModelTextTime SeriesBenchmark
🎯 What it does: Constructed and released the HandX large-scale dataset of hand movements and interactions, and implemented a text-driven model for generating hand movements based on this dataset;
HanDyVQA: A Video QA Benchmark for Fine-Grained Hand-Object Interaction Dynamics
Masatoshi Tateno (Institute of Industrial Science, University of Tokyo), Takuma Yagi (National Institute of Advanced Industrial Science and Technology)
TransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Proposed the HanDyVQA video question-answering dataset to evaluate fine-grained spatiotemporal dynamics in hand-object interactions.
Haptic Neural Fields: Bringing Tactile Interactions to 3D Rendered Scenes
Antonio Luigi Stefani (University of Trento), Francesco De Natale (University of Trento)
ClassificationGenerationRobotic IntelligenceDiffusion modelNeural Radiance FieldContrastive LearningMultimodality
🎯 What it does: Train and apply a neural field model to synthesize corresponding tactile acceleration signals in real-time given a 3D reconstructed scene and user-specified touch trajectory (direction, velocity, normal force);
Harmonic Canvas: Inversion-Free Editing for Visually-Guided Music Style Transfer
Yue Lei (University of Electronic Science and Technology of China), Fan Zhou (Key Laboratory of Intelligent Digital Media Technology)
GenerationTransformerFlow-based ModelMultimodality
🎯 What it does: This paper proposes a multimodal music style transfer framework based on a non-invertible flow model, which can simultaneously utilize text and images as style instructions to reshape input music while maintaining melodic and rhythmic consistency.
Harmonious Parameter Adaptation in Continual Visual Instruction Tuning for Safety-Aligned MLLMs
Ziqi Wang (Hefei University of Technology), Meng Wang (Tsinghua University)
Safty and PrivacyTransformerLarge Language ModelVision Language ModelMultimodalityBenchmark
🎯 What it does: Propose a post-training framework HPA for continuous visual instruction fine-tuning (post-SA CVIT) on safe-aligned multi-modal large language models (MLLMs), minimizing task forgetting while ensuring safety.
Harmonized Feature Conditioning and Frequency-Prompt Personalization for Multi-Rater Medical Segmentation
Sanaz Karimijafarbigloo (University of Regensburg), Dorit Merhof (University of Regensburg)
SegmentationConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Propose a unified probabilistic multi-evaluation medical image segmentation framework that separates scan noise from rater variability and enables personalized prediction.
Harmony: Harmonizing Audio and Video Generation through Cross-Task Synergy
Teng Hu (Shanghai Jiao Tong University), Ran Yi (Shanghai Jiao Tong University)
GenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelVideoMultimodalityBenchmarkAudio
🎯 What it does: Propose a unified cross-modal generation framework called Harmony, capable of synchronously generating high-quality audio and video, compatible with three paradigms: audio-driven, video-driven, and joint generation.
Harnessing Chain-of-Thought Reasoning in Multimodal Large Language Models for Face Anti-Spoofing
Honglu Zhang (Didi Chuxing), Zhaofeng He (Beijing University of Posts and Telecommunications)
Anomaly DetectionSafty and PrivacyExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Constructed the FaceCoT large-scale chain-of-thought (CoT) visual question answering (VQA) dataset and proposed a CoT-Enhanced Progressive Learning (CEPL) training strategy to achieve high accuracy and interpretability in facial disguise detection.
Harnessing the Power of Foundation Models for Accurate Material Classification
Qingran Lin (Georgia Institute of Technology), Chaolun Zhu (Waseda University)
ClassificationGenerationData SynthesisRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: Constructed a material image synthesis dataset based on text generation and automatic annotation, and proposed a dual-stream framework combining visual foundation models and language foundation models for material classification.
HATS: Hardness-Aware Trajectory Synthesis for GUI Agents
Rui Shao (Harbin Institute of Technology), Gongwei Chen (Harbin Institute of Technology)
Data SynthesisReinforcement LearningVision Language ModelTextSequential
🎯 What it does: Propose a closed-loop trajectory synthesis framework called HATS, designed to generate high-quality, semantically aligned GUI trajectories, aiding in training more robust GUI agents.
HAVE-Bench: Hierarchical Audio-Visual Evaluation from Perception to Interaction
Muyan Zhong (Tsinghua University), Jifeng Dai (Tsinghua University)
Large Language ModelVision-Language-Action ModelMultimodalityBenchmark
🎯 What it does: Built a three-tier (perception, reasoning, interaction) audio-visual evaluation benchmark called HAVE-Bench to systematically assess the audio understanding and interaction capabilities of multi-modal large language models.
HAWK: Head Importance-Aware Visual Token Pruning in Multimodal Models
Qihui Zhu (University of Science and Technology of China), Yinfei Pan (Huawei Noah's Ark Lab)
Computational EfficiencyTransformerLarge Language ModelVision Language ModelImageVideoMultimodality
🎯 What it does: Propose a training-free head importance-aware visual token pruning method (HAWK), which efficiently trims visual tokens in multimodal large language models by leveraging the differential importance of attention heads and text-guided attention.
HBridge: H-Shape Bridging of Heterogeneous Experts for Unified Multimodal Understanding and Generation
Xiang Wang (Huazhong University of Science and Technology), Nong Sang (Huazhong University of Science and Technology)
GenerationTransformerLarge Language ModelMixture of ExpertsDiffusion modelFlow-based ModelImageTextMultimodality
🎯 What it does: Constructed an H-shaped multimodal unified model called HBridge, which integrates heterogeneous experts (LLM and Diffusion generators) to simultaneously achieve visual understanding and image generation/editing.
HCL-FF: Hierarchical and Contrastive Learning for Forward-Forward Algorithm
Jie-En Yao (University of Southern California), C.-C. Jay Kuo (University of Southern California)
ClassificationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Propose a new hierarchical and contrastive learning Forward-Forward (FF) framework called HCL-FF, improving the hierarchical collaboration and feature semantic preservation issues in traditional FF frameworks.
HDR-VLM: HDR-Domain Adaptation of VLMs and Preference-Aligned Quality Assessment for HDR Video Color Grading
Hao Yuan (University of Chinese Academy of Sciences), Jing Li (HUJING Digital Media & Entertainment Group)
Domain AdaptationExplainability and InterpretabilityTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelVideoText
🎯 What it does: Proposes HDR-VLM, a two-stage Vision-Language Model (VLM) adaptation method for evaluating color grading quality in HDR videos. The first stage bridges the domain gap between SDR and HDR by unifying HLG encoding and progressive fine-tuning; the second stage aligns model outputs with subjective preferences through Group Relative Policy Optimization (GRPO)-based reinforcement learning, providing interpretable quality justifications.
HDW-SR: High-Frequency Guided Diffusion Model based on Wavelet Decomposition for Image Super-Resolution
Chao Yang (Xidian University), Guang Jiang (Xidian University)
Super ResolutionTransformerDiffusion modelImage
🎯 What it does: Proposed a high-frequency guided diffusion network based on waveform decomposition (HDW-SR) for single-image super-resolution, utilizing high-frequency information from pre-super-resolved images to guide the diffusion process, thereby significantly improving detail reconstruction quality while maintaining structural consistency.
Head-wise Adaptive Rotary Positional Encoding for Fine-Grained Image Generation
Jiaye Li (Fudan University), Siyu Zhu (Fudan University)
GenerationTransformerDiffusion modelImage
🎯 What it does: Propose a Head-Adaptive Rotating Positional Encoding (HARoPE), introducing a learnable linear transformation into multi-dimensional Transformer positional encoding to enhance fine-grained image generation and understanding.
Hear What Matters! Text-conditioned Selective Video-to-Audio Generation
Junwon Lee (KAIST), Jiyoung Lee (Ewha Womans University)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelFlow-based ModelVideoTextMultimodalityBenchmarkAudio
🎯 What it does: Propose a text-conditioned selective video-to-audio generation framework called SELVA, which can synthesize only the target audio source in multi-object videos based on user-provided text prompts.
Hear What You See: Video-to-Audio Generation with Diffusion Transformer and Semantic-Temporal Alignment-Ranked Direct Preference Optimization
Kai Wang (University Of Toronto), Yuewen Cao (Shanghai Ai Lab)
GenerationTransformerReinforcement LearningDiffusion modelFlow-based ModelVideoTextMultimodalityAudio
🎯 What it does: Propose VisioSonic, an end-to-end video-to-audio generation framework based on flow-matching diffusion transformers, capable of achieving high-fidelity synchronization in both semantics and temporal alignment;
Hear you are: Teaching LLMs Spatial Reasoning with Vision and Spatial Sound
Hyeonggon Ryu (Hankuk University of Foreign Studies), David Harwath (University of Texas at Austin)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityAudio
🎯 What it does: Designed a new audio-visual spatial reasoning task, constructed the Hear You Are QA dataset with 1 million question-answer pairs, and proposed a multi-modal LLM framework called Hear You Are LLM to answer these questions.
Hearing the Room Through the Shape of the Drum: Modal-Guided Sound Recovery from Multi-Point Surface Vibrations
Shai Bagon (Weizmann Institute of Science), Mark Sheinin (Weizmann Institute of Science)
RestorationTime SeriesPhysics Related
🎯 What it does: Restored on-site sound through multi-point laser speckle vibration measurement combined with a physical model of elastic vibration modes;
HERBench: A Benchmark for Multi-Evidence Integration in Video Question Answering
Dan Ben Ami (Ben-Gurion University of Negev), Chaim Baskin (Ben-Gurion University of Negev)
TransformerVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Propose HERBench benchmark, specifically designed for multi-evidence (at least three different time visual clues) indispensable question answering tasks, and introduce MRFS (Minimum Required Frame Set) metric to measure the model's requirements for multi-frame fusion.
Hermite Radial Basis Function for Surface Reconstruction via Differentiable Rendering
Hugo Blanc (PSL University), Alexis Paljic (PSL University)
GenerationNeural Radiance FieldPoint CloudMesh
🎯 What it does: This paper proposes a global implicit surface representation by combining Hermite Radial Basis Functions (HRBF) with differentiable rendering, and achieves end-to-end multi-view surface reconstruction through BVH-accelerated ray marching.
HERO: Hierarchical Embedding-Refinement for Open-Vocabulary Temporal Sentence Grounding in Videos
Tingting Han (Hangzhou Dianzi University), Zhou Yu (Hangzhou Dianzi University)
RetrievalRepresentation LearningTransformerLarge Language ModelVision Language ModelContrastive LearningVideoTextMultimodalityBenchmark
🎯 What it does: This paper proposes the open-vocabulary video sentence grounding (OV-TSGV) task and designs the HERO framework to address this problem; meanwhile, it constructs the first open-vocabulary benchmarks Charades-OV and ActivityNet-OV.
HeroGS: Hierarchical Guidance for Robust 3D Gaussian Splatting under Sparse Views
Jiashu Li (University of Chinese Academy of Sciences), Jianbin Jiao (University of Chinese Academy of Sciences)
Gaussian SplattingImage
🎯 What it does: This paper proposes the HeroGS framework, which improves the reconstruction quality of 3D Gaussian Splatting under sparse viewpoint conditions through multi-layer guidance.
HeSS: Head Sensitivity Score for Sparsity Redistribution in VGGT
Yongsung Kim (Seoul National University), Sungroh Yoon (Seoul National University)
Pose EstimationDepth EstimationComputational EfficiencyTransformerImagePoint Cloud
🎯 What it does: Sparse the global attention layers of VGGT by introducing a two-phase sparsification process: first, use Head Sensitivity Score (HeSS) to evaluate the sparsity sensitivity of each attention head, then redistribute the attention budget based on HeSS during inference.
Heterogeneous Decentralized Diffusion Models
Zhiying Jiang (Bagel Labs), Bidhan Roy (Bagel Labs)
GenerationComputational EfficiencyTransformerMixture of ExpertsDiffusion modelFlow-based ModelImage
🎯 What it does: Propose a diffusion model framework that can train different objectives (DDPM and Flow Matching) in a decentralized environment, supporting single GPU training;
Heuristic Self-Paced Learning for Domain Adaptive Semantic Segmentation under Adverse Conditions
Shiqin Wang (Wuhan University), Kaiyan Zhao (Wuhan University)
SegmentationDomain AdaptationReinforcement LearningAuto EncoderImage
🎯 What it does: Propose a reinforcement learning-based adaptive curriculum learning framework (HeuSCM) for unsupervised domain adaptation semantic segmentation under extreme weather conditions.
Heuristic-inspired Reasoning Priors Facilitate Data-Efficient Referring Object Detection
Xu Zhang (University of Sydney), Dacheng Tao (Nanyang Technological University)
Object DetectionTransformerVision Language ModelImageTextMultimodality
🎯 What it does: Propose HeROD, which improves referring object detection under data-scarce conditions by injecting spatial and semantic reasoning priors.
HFedATM: Hierarchical Federated Domain Generalization via Optimal Transport and Regularized Mean Aggregation
Thinh Nguyen (VinUniversity), Kok-Seng Wong (VinUniversity)
Domain AdaptationFederated LearningConvolutional Neural NetworkImageText
🎯 What it does: Propose the HFedDG scenario and design HFedATM as a data-agnostic hierarchical aggregation method, significantly enhancing cross-domain generalization capabilities
HFR and HDR Video from Multi-Attenuated Spikes Using a Rapidly Rotating SpokeND Filter
Yakun Chang (Beijing Jiaotong University), Boxin Shi (Peking University)
RestorationConvolutional Neural NetworkVideoTime Series
🎯 What it does: This paper utilizes a rapidly rotating SpokeND filter and a spiking camera to capture multi-attenuated spikes, and reconstructs them into high frame rate (up to 2000 FPS) and high dynamic range (HDR) videos through a two-stage network ReST-Net (ReGain+ReFine);
Hg-I2P: Bridging Modalities for Generalizable Image-to-Point-Cloud Registration via Heterogeneous Graphs
Pei An (Huazhong University of Science and Technology), Liangliang Nan (Huazhong University of Science and Technology)
Pose EstimationGraph Neural NetworkImageMultimodalityPoint Cloud
🎯 What it does: Propose a heterogeneous graph-based image-to-point cloud registration framework named Hg-I2P.
HG-Lane: High-Fidelity Generation of Lane Scenes under Adverse Weather and Lighting Conditions without Re-annotation
Daichao Zhao (Shanghai Jiao Tong University), Qiankun Li (Shanghai Jiao Tong University)
GenerationData SynthesisAutonomous DrivingDiffusion modelImage
🎯 What it does: This paper proposes HG-Lane, which synthesizes high-fidelity lane images under adverse weather and lighting conditions using a two-stage ControlNet diffusion framework without re-annotation, and constructs a lane dataset containing 30,000 images across six weather/lighting categories.
Hi-Lo Prune: Look at What You'll Lose before Pruning with Hierarchical Token Selection
Zixun Sun, Yi Yang (State Key Lab Of Brain Machine Intelligence Zhejiang University)
Computational EfficiencyTransformerVision Language ModelImageMultimodality
🎯 What it does: Propose a training-agnostic visual token pruning method called Hi-Lo Prune, which enables high-ratio pruning in early layers without fine-tuning the model
HiCoGen: Hierarchical Compositional Text-to-Image Generation in Diffusion Models via Reinforcement Learning
Hongji Yang (University of Macau), Jianbing Shen (Chongqing Afari Intelligent Drive)
GenerationTransformerReinforcement LearningPrompt EngineeringVision Language ModelDiffusion modelImageTextBenchmarkChain-of-Thought
🎯 What it does: Propose the HiCoGen framework, combining Chain of Synthesis (CoS) to decompose complex text into minimal semantic units, utilizing LLM for parsing and rewriting, progressively generating and overlaying images; simultaneously introducing hierarchical reward-based RL optimization and decaying stochasticity scheduling; also constructing a new benchmark HiCoPrompt;
HiconAgent: History Context-aware Policy Optimization for GUI Agents
Xurui Zhou (Harbin Institute of Technology), Rui Shao (Shenzhen Loop Area Institute)
TransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelMultimodality
🎯 What it does: Propose HiconAgent, which improves the utilization of historical information in GUI agents by using HCPO with dynamic context sampling and Anchor-guided historical compression.
Hidden Dangers of Compositional Generation: Diagnosing Semantic Safety Failures in Text-to-Image Models
Haoming Yang (University Of Chinese Academy Of Sciences), Qingming Huang (University Of Chinese Academy Of Sciences)
GenerationAdversarial AttackLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelImageText
🎯 What it does: Propose the CoRA attack framework, which bypasses the security filters of text-to-image (T2I) models by performing fine-grained semantic decomposition and recombination in the black-box text space to generate high-risk images.
Hidden Monotonicity: Explaining Deep Neural Networks via their DC Decomposition
Jakob Paul Zimmermann (Technical University Berlin), Georg Loho (Freie Universitat Berlin)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: Numerically stable DC decomposition of pre-trained ReLU networks yields two monotonic convex subnetworks, leading to the proposal of novel interpretable methods such as SplitCAM, SplitLRP, and SplitGrad.
HiDRA: Hierarchical Degradation Representation and Adaptation with Generative Priors for Enhancing Infrared Vision
Zihang Chen (Dalian University of Technology), Risheng Liu (Dalian University of Technology)
RestorationTransformerDiffusion modelContrastive LearningImage
🎯 What it does: Propose the HiDRA framework, which divides thermal image enhancement into degradation representation learning and adaptive fine-tuning based on pre-trained diffusion models;
Hier-COS: Making Deep Features Hierarchy-aware via Composition of Orthogonal Subspaces
Depanshu Sani (Indraprastha Institute of Information Technology), Saket Anand (Indraprastha Institute of Information Technology)
ClassificationRepresentation LearningImage
🎯 What it does: Propose a framework based on Hierarchical Combinatorial Orthogonal Subspaces (Hier-COS) to construct a hierarchical-aware feature space, jointly achieving fine-grained classification and hierarchical multi-level classification;
HieraMamba: Video Temporal Grounding via Hierarchical Anchor-Mamba Pooling
Joungbin An (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)
RetrievalComputational EfficiencyRepresentation LearningVision Language ModelContrastive LearningVideoText
🎯 What it does: Propose a hierarchical time grounding model called HieraMamba for long videos, which can precisely locate the time segments corresponding to natural language queries without compromising temporal resolution.
HierAmp: Coarse-to-Fine Autoregressive Amplification for Generative Dataset Distillation
Lin Zhao (Northeastern University), Jianyang Gu (Ohio State University)
Knowledge DistillationTransformerAuto EncoderImage
🎯 What it does: This paper proposes a method called HIERAMP, which utilizes a visual autoregressive model (VAR) to inject learnable class tokens at multiple scales and enhances attention at each scale, thereby guiding the model to focus on semantically important regions during the dataset distillation process and generating more discriminative synthetic samples.
Hierarchical Action Learning for Weakly-Supervised Action Segmentation
Junxian Huang (Guangdong University of Technology), Shenghua Gao (University of Hong Kong)
SegmentationTransformerVideoBenchmark
🎯 What it does: Proposed the Hierarchical Action Learning (HAL) model, which models weakly supervised action segmentation using hierarchical causal generative processes and pyramid Transformers, addressing issues of over-segmentation and boundary noise in traditional methods.
Hierarchical Attacks for Multi-Modal Multi-Agent Reasoning
Hao Zhou (JD.com), Ai Han (JD.com)
Adversarial AttackLarge Language ModelAgentic AIMultimodality
🎯 What it does: This paper proposes a hierarchical attack framework called HAM³ to systematically evaluate the vulnerability of multi-modal multi-agent systems in three layers: perception, communication, and reasoning, and experiments are conducted on the GQA visual question answering task.
Hierarchical Codec Diffusion for Video-to-Speech Generation
Jiaxin Ye (Fudan University), Hongming Shan (Harbin Institute of Technology)
GenerationData SynthesisTransformerDiffusion modelVideoMultimodalityAudio
🎯 What it does: Propose the HiCoDiT model to generate high-fidelity speech from silent videos;
Hierarchical Concept Embedding & Pursuit for Interpretable Image Classification
Nghia Nguyen (University of Pennsylvania), René Vidal (University of Pennsylvania)
ClassificationExplainability and InterpretabilityTransformerVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes the Hierarchical Concept Embedding & Pursuit (HCEP) framework, which constructs concept embeddings in the latent space of vision-language models by leveraging semantic hierarchical structures, and recovers the conceptual paths contained in images through hierarchical sparse coding (Hierarchical Beam OMP), thereby achieving interpretable image classification.
Hierarchical Enhancement of Semantic Priors for Disentangled Text-Driven Motion Generation
Wenhan Lv (Xiamen University), Qingqiang Wu (Xiamen University)
GenerationDiffusion modelAuto EncoderTextSequential
🎯 What it does: Propose a unified text-driven 3D human motion generation framework HESP, enabling semantic disentanglement and controllable generation;
Hierarchical Long Video Understanding with Audiovisual Entity Cohesion and Agentic Search
Xinlei Yin (University of Science and Technology of China), Yan Lu (Microsoft Research Asia)
RetrievalRepresentation LearningTransformerLarge Language ModelAgentic AIVision Language ModelVideoTextBenchmarkRetrieval-Augmented GenerationChain-of-ThoughtAudio
🎯 What it does: Propose a unified framework HAVEN that combines hierarchical video indexing (global, scene, clip, entity) with cross-modal entity consistency, augmented by intelligent agent search, to achieve semantic consistency, entity tracking, and multi-level reasoning in long videos.
Hierarchical Point-Patch Fusion with Adaptive Patch Codebook for 3D Shape Anomaly Detection
Xueyang Kang (University of Melbourne), Liangliang Nan (Delft University of Technology)
Anomaly DetectionTransformerContrastive LearningPoint CloudMesh
🎯 What it does: Propose a multi-scale point-surface fusion network that achieves 3D shape anomaly detection and localization using self-supervised patching and an adaptive patch codebook.
Hierarchical Process Reward Models are Symbolic Vision Learners
Shan Zhang (Adelaide AIML), Anton van den Hengel (Adelaide AIML)
Representation LearningLarge Language ModelSupervised Fine-TuningReinforcement LearningAuto EncoderContrastive LearningImageTextMultimodality
🎯 What it does: Propose a self-supervised symbolic autoencoder that can parse geometric diagrams into structured logical forms and regenerate the original diagram
Hierarchical Visual Relocalization with Nearest View Synthesis from Feature Gaussian Splatting
Huaqi Tao (Southern University of Science and Technology), Hong Zhang (Southern University of Science and Technology)
Pose EstimationRetrievalGaussian SplattingSimultaneous Localization and MappingImage
🎯 What it does: Propose a hierarchical visual localization framework named SplatHLoc based on Feature Gaussian Splatting (FGS), addressing the limitations of traditional point-based methods in sparse observations and feature matching.
Hierarchically Robust Zero-shot Vision-language Models
Junhao Dong (Nanyang Technological University), Piotr Koniusz (University of New South Wales)
Representation LearningAdversarial AttackVision Language ModelImageTextMultimodalityBiomedical Data
🎯 What it does: Propose an adversarial fine-tuning framework based on hierarchical hyperplane embedding, aiming to enhance the zero-shot robustness of Vision-Language models on leaf categories and their parent categories (superclasses).
HierEdit: Region-Aware Hierarchical Diffusion for Efficient High-Resolution Editing
Yuyao Zhang (Dartmouth College), Yu-Wing Tai (Dartmouth College)
Image TranslationComputational EfficiencyTransformerDiffusion modelFlow-based ModelImageBenchmark
🎯 What it does: This paper proposes the HierEdit framework, which achieves efficient and accurate local image editing at ultra-high resolution (4K) by utilizing low-resolution proxies and region-aware hierarchical diffusion models;
HierUQ: Hierarchical Uncertainty Quantification with Adaptive Granularity Reconciliation for Degraded Image Classification
Yang Chu (Zhejiang University), Yuntao Qian (Zhejiang University)
ClassificationTransformerImage
🎯 What it does: Propose the HierUQ framework, achieving hierarchical classification of degraded images through Hierarchical Uncertainty Quantization (HUQ), Confidence-Aware Path Adjustment (CAPA), and Multi-Layer Joint Optimization (MLJO);
HiF-VLA: Hindsight, Insight and Foresight through Motion Representation for Vision-Language-Action Models
Minghui Lin, Donglin Wang (Westlake University)
Robotic IntelligenceTransformerLarge Language ModelMixture of ExpertsVision-Language-Action ModelOptical FlowVideoText
🎯 What it does: Propose HiF-VLA, a vision-language-action model that achieves bidirectional temporal reasoning through motion vectors, significantly improving the continuity and consistency of long-horizon manipulation tasks.
HiFi-BRep: High-Fidelity Latent Representation for Robust B-Rep Generation
Junhao Hou (State Key Lab of CAD&CG, Zhejiang University), Kun Zhou (State Key Lab of CAD&CG, Zhejiang University)
GenerationTransformerDiffusion modelAuto EncoderMesh
🎯 What it does: Propose HiFi-BRep, a single-stage method generating high-fidelity, structurally valid B-Rep models that balance geometry and topology;
HiFi-Inpaint: Towards High-Fidelity Reference-Based Inpainting for Generating Detail-Preserving Human-Product Images
Yichen Liu (University of Chinese Academy of Sciences), Pheng-Ann Heng (Chinese University of Hong Kong)
GenerationTransformerDiffusion modelImage
🎯 What it does: Proposes the HiFi-Inpaint framework, which uses reference images to perform high-fidelity inpainting on human and product images, generating synthetic images that preserve fine details.
HiFICL: High-Fidelity In-Context Learning for Multimodal Tasks
Xiaoyu Li (University of Electronic Science and Technology of China), Zihan Xiong (University of Electronic Science and Technology of China)
Meta LearningTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Propose HiFICL, achieving In-Context Learning (ICL) mechanism for high-fidelity multimodal models through learnable virtual key-value pairs and low-rank decomposition, and present it as a parameter-efficient fine-tuning method.
High Resolution Neural Video Coding with Bi-directional Confidence-Guided Reference Information Modeling
Feng Ye (Wangxuan Institute of Computer Technology, Peking University), Chuanmin Jia (Wangxuan Institute of Computer Technology, Peking University)
CompressionConvolutional Neural NetworkOptical FlowVideoBenchmark
🎯 What it does: Proposes a neural B-frame compression method for high-resolution videos, HR-NVC, which achieves more reliable motion estimation and reference fusion through bidirectional confidence-guided reference information modeling.
High-Fidelity Diffusion Face Swapping with ID-Constrained Facial Conditioning
Dailan He (Cuhk Mmlab), Hongsheng Li (Cuhk Mmlab)
Image TranslationGenerationRetrievalDiffusion modelGenerative Adversarial NetworkImageText
🎯 What it does: Developed a high-fidelity face swapping method based on diffusion models, proposing a multi-stage training framework with identity constraints and decoupled identity/attribute conditional injection technology.
High-Fidelity Mobile Avatars with Pruned Local Blendshapes
Youyi Zhan (Zhejiang University), Kun Zhou (Zhejiang University)
GenerationGaussian SplattingVideo
🎯 What it does: Reconstruct a high-fidelity animated 3D Gaussian Spline avatar of the full human body from multi-view videos, which can be rendered at high frame rates on mobile devices.