CVPR 2026 Papers — Page 17
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
High-Fidelity Virtual Try-On beyond Paired Data Scarcity via Diffusion-based Cycle-Consistent Learning
Jia Wu (Alibaba International Digital Commerce Group), Xiaoyi Zeng (Harbin Institute of Technology)
Image TranslationGenerationTransformerVision Language ModelDiffusion modelImage
🎯 What it does: Implement virtual try-on using diffusion models, and train on massive unpaired human image data through cycle-consistent learning to complete the full closed-loop of garment removal and re-wearing.
High-Precision Dichotomous Image Segmentation via Depth Integrity-Prior and Fine-Grained Patch Strategy
Xianjie Liu (Sichuan University), Qijun Zhao (Sichuan University)
SegmentationTransformerImage
🎯 What it does: For high-precision binary image segmentation tasks, the paper proposes using pseudo depth as an auxiliary modality and designs the PDFNet network to achieve fine-grained segmentation.
High-Quality and Efficient Turbulence Mitigation with Events
Xiaoran Zhang (Huazhong University of Science and Technology), Luxin Yan (Sun Yat-sen University)
RestorationConvolutional Neural NetworkTransformerContrastive LearningOptical FlowImageVideo
🎯 What it does: Proposed an event camera-based turbulence elimination method called EHETM, which utilizes high temporal resolution events to extract fine-grained motion information within a short time, achieving high-quality image recovery with low frame rates.
Hilbert Curve-Based Attention Enabling Topology-Preserving Image Tensor Representation for Semantic Segmentation Network
Linkang Xu (Tongji University), Xiangxin Ji (Tongji University)
SegmentationTransformerImage
🎯 What it does: Designed and validated a lightweight semantic segmentation network called TPSegformer based on the Hilbert curve for precise segmentation of building defect images collected by drones.
Hilbert-Geo: Solving Solid Geometric Problems by Neural-Symbolic Reasoning
Ruoran Xu (Ricoh Software Research Center Beijing), Qiufeng Wang (Ricoh Software Research Center Beijing)
Large Language ModelPrompt EngineeringTextMultimodality
🎯 What it does: Proposed the Hilbert-Geo framework, integrating a multi-modal formal parser and a solid geometry reasoning engine to solve solid geometry problems;
HiLoRA: Hierarchical Low-Rank Adaptation for Personalized Federated Learning
Zihao Peng (Beijing Normal University), Tian Wang (Beijing Normal University)
Federated LearningImage
🎯 What it does: Propose HiLoRA, a three-layer low-rank adaptation framework in federated learning, which divides LoRA modules into root, cluster, and leaf layers to achieve knowledge sharing and personalization at the global, subgroup, and client levels, respectively.
Hint2Gen: Bridging Understanding and Generation via Code-structured Hints
Yuanpeng Tu (University of Hong Kong), Hengshuang Zhao (University of Hong Kong)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelAuto EncoderImageTextBenchmark
🎯 What it does: This paper proposes the idea of bridging understanding and generation through code-structured visual prompts (SVG/HTML), constructing the Hint2Gen model and creating the Reason2Gen benchmark;
HiSpatial: Taming Hierarchical 3D Spatial Understanding in Vision-Language Models
Huizhi Liang (Tsinghua University), Jiaolong Yang (Microsoft Research Asia)
Depth EstimationTransformerSupervised Fine-TuningVision Language ModelImageMultimodalityPoint CloudBenchmark
🎯 What it does: Proposed a four-tier 3D spatial understanding framework and constructed a spatial VQA dataset containing approximately 5M images, 45M objects, and 2 billion QA pairs through an automated data generation pipeline, achieving multi-level spatial reasoning capabilities via supervised fine-tuning of RGB-D VLM.
Hist2Style: Histogram-Guided Stylization with Bilateral Grids
Dekel Galor (Adobe University of California, Berkeley), Ilya Chugunov (Adobe University of California, Berkeley)
Image TranslationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkVision Language ModelImage
🎯 What it does: Proposed a lightweight rasterization network called Hist2Style for achieving high-resolution, real-time, interactive photorealistic style transfer.
History to Future: Evolving Agent with Experience and Thought for Zero-shot Vision-and-Language Navigation
Guangzhao Dai (Nanjing University of Science and Technology), Xiangbo Shu (Singapore Management University)
Autonomous DrivingLarge Language ModelAgentic AIVision Language ModelImageTextMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose a zero-shot continuous environment vision-language navigation framework named EvoNav, which enhances the decision-making quality of large language models (LLMs) by integrating future chain-of-thought (F-CoT) and historical chain-of-experience (H-CoE).
HOG-Layout: Hierarchical 3D Scene Generation, Optimization and Editing via Vision-Language Models
Haiyan Jiang (Hong Kong Polytechnic University), Henry Been-Lirn Duh (Hong Kong Polytechnic University)
GenerationOptimizationTransformerLarge Language ModelVision Language ModelTextMultimodalityBenchmarkRetrieval-Augmented GenerationOrdinary Differential Equation
🎯 What it does: Proposes a 3D scene generation and editing framework called HOG-Layout based on hierarchical representation, which can create high-quality, physically feasible, and interactive indoor scenes through text instructions.
Hoi! - A Multimodal Dataset for Force-Grounded, Cross-View Articulated Manipulation
Tim Engelbracht (ETH Zurich), Zuria Bauer (ETH Zurich)
Robotic IntelligenceSimultaneous Localization and MappingVideoMultimodalityTime Series
🎯 What it does: Constructed the Hoi! dataset, providing cross-perspective, cross-posture multimodal (RGB, depth, force/torque, tactile, pose) human-robot interaction data, and simultaneously recorded human hand, wrist-mounted camera, handheld gripper, and self-developed Hoi! gripper operations on the same object.
HOLO: Homography-Guided Pose Estimator Network for Fine-Grained Visual Localization on SD Maps
Xuchang Zhong (Beijing Institute of Technology), Hao Fang (Beijing Institute of Technology)
Pose EstimationAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: Propose a Homography-guided multi-camera visual localization network named HOLO, which achieves fine-grained localization by leveraging geometric constraints between BEV semantics and standard definition maps (SD maps).
HoloCine: Holistic Generation of Cinematic Multi-Shot Long Video Narratives
Yihao Meng (Hong Kong University of Science and Technology), Huamin Qu (Hong Kong University of Science and Technology)
GenerationTransformerLarge Language ModelPrompt EngineeringDiffusion modelVideoText
🎯 What it does: Proposed a holistic generation framework called HoloCine for multi-shot long video narratives, capable of generating a complete video sequence from the first shot to the Nth shot in one go.
Homaloidal parametrization for detecting critical two-view configurations
Rakshith Madhavan (Politecnico di Milano), Luca Magri (Politecnico di Milano)
Data SynthesisPose EstimationOptimizationImage
🎯 What it does: Investigated the detection of critical configurations (i.e., scenarios causing ambiguity or instability in fundamental matrix estimation) using eight point correspondences across two views, and proposed a novel discriminative method based on homotopy projective transformation.
HoneyBee: Data Recipes for Vision-Language Reasoners
Hritik Bansal (FAIR at Meta), Ramakanth Pasunuru (FAIR at Meta)
Computational EfficiencyData-Centric LearningLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: This paper constructs a large-scale high-quality chain-of-thought reasoning dataset called HONEYBEE through a systematic multi-stage data curation process (context selection, data intervention, and scale expansion), and uses this dataset to perform supervised fine-tuning on multiple sizes of visual-language models, significantly improving the models' performance on mathematical and graphical reasoning tasks.
HOPS: Hierarchical Open-vocabulary Part Segmentation with Attention-Aware Filtering and Affinity-Guided Enhancement
Xinlong Li (Tianjin University), Wei Feng (Tianjin University)
SegmentationTransformerVision Language ModelContrastive LearningImageMultimodalityBenchmark
🎯 What it does: Proposes HOPS, a hierarchical open-vocabulary part segmentation framework, combining the Attention-Aware Filtering Module (AFM) and Affinity-Guided Enhancement Module (AEM) to address object over-segmentation and part under-segmentation issues.
HorizonForge: Driving Scene Editing with Any Trajectories and Any Vehicles
Yifan Wang (NEC Labs America), Ziyu Jiang (NEC Labs America)
Data SynthesisAutonomous DrivingLarge Language ModelDiffusion modelGaussian SplattingVideoTextMeshBenchmark
🎯 What it does: Propose a unified framework called HorizonForge, which can edit driving scenarios based on arbitrary trajectories and natural language descriptions, insert or move vehicles, and generate high-quality, temporally consistent videos.
How Far Can We Go With Synthetic Data for Audio-Visual Sound Source Localization?
Arda Senocak (Ulsan National Institute of Science and Technology), Joon Son Chung (Korea Advanced Institute of Science and Technology)
Data SynthesisLarge Language ModelVision Language ModelDiffusion modelContrastive LearningImageMultimodalityAudio
🎯 What it does: This paper proposes a scalable voice source localization training framework based on text-to-X generation models and constructs synthetic clone data of VGGSound;
How Much 3D Do Video Foundation Models Encode?
Zixuan Huang (University of Illinois at Urbana Champaign), James M. Rehg (University of Illinois at Urbana Champaign)
Pose EstimationDepth EstimationTransformerContrastive LearningVideoPoint CloudBenchmark
🎯 What it does: Propose a model-agnostic shallow Transformer detector for evaluating and detecting 3D awareness in video foundation models (VidFMs), which directly predicts 3D points, depth, and camera pose from frozen VidFM features, and systematically benchmarks on the CO3Dv2 and DL3DV datasets.
How to Take a Memorable Picture? Empowering Users with Actionable Feedback
Francesco Laiti (University of Trento), Elisa Ricci (University of Trento)
Explainability and InterpretabilityKnowledge DistillationRepresentation LearningTransformerLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Propose the 'Memorability Feedback (MemFeed)' task, construct the MemBench benchmark, and design an activation-oriented, training-free method called MemCoach, providing users with actionable natural language guidance to enhance photo memorability.
HP-Edit: A Human-Preference Post-Training Framework for Image Editing
Fan Li (Huawei Noah's Ark Lab), Wangmeng Zuo (Harbin Institute of Technology)
GenerationTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelFlow-based ModelImageMultimodality
🎯 What it does: This paper proposes the HP-Edit framework, which post-trains pre-trained image editing models to make the generated editing results more aligned with human preferences and realism.
HQC-NBV: A Hybrid Quantum-Classical View Planning Approach
Xiaotong Yu (Hong Kong Polytechnic University), Chang Wen Chen (Hong Kong Polytechnic University)
OptimizationRobotic Intelligence
🎯 What it does: Proposed a hybrid quantum-classical framework named HQC-NBV for solving the next best view (NBV) planning problem.
HSI-GPT2: A Dual-Granularity Large Motion Reasoning Model with Diffusion Refinement for Human-Scene Interaction
Yuan Wang (Tsinghua University), Shengjin Wang (Tsinghua University)
GenerationData SynthesisTransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelDiffusion modelContrastive LearningImageVideoTextMultimodalityPoint CloudChain-of-Thought
🎯 What it does: Proposed HSI-GPT2, a unified large-scale human-robot interaction (HSI) model capable of simultaneously generating, understanding, and completing actions in 3D scenes, and introduced multi-modal chain-of-thought (MoCoT) and reinforcement learning (GRPO) to enhance semantic reasoning and physical consistency.
HTNav: A Hybrid Navigation Framework with Tiered Structure for Urban Aerial Vision-and-Language Navigation
Chengjie Fan (Nanjing University of Aeronautics and Astronautics), Jie Qin (Nanjing University of Aeronautics and Astronautics)
Autonomous DrivingConvolutional Neural NetworkReinforcement LearningVision Language ModelVision-Language-Action ModelImageTextMultimodality
🎯 What it does: Designed and implemented HTNav, a hierarchical navigation framework combining imitation learning and reinforcement learning for urban aerial vision-language navigation tasks.
HTTM: Head-wise Temporal Token Merging for Faster VGGT
Weitian Wang (Robert Bosch GmbH), Akash Kumar (Robert Bosch GmbH)
Computational EfficiencyTransformerImagePoint Cloud
🎯 What it does: Proposed Head-wise Temporal Token Merging (HTTM), a training-agnostic token merging method that significantly accelerates the global attention layer of VGGT;
Hugging Visual Prompt and Segmentation Tokens: Consistency Learning for Fine-Grained Visual Understanding in MLLMs
Jing Yang (Tongji University), Hanli Wang (Tongji University)
RecognitionSegmentationLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Propose FCLM, a multimodal large language model that unifies region-level captioning and pixel-level grounding through consistency learning;
HulluEdit: Single-Pass Evidence-Consistent Subspace Editing for Mitigating Hallucinations in Large Vision-Language Models
Yangguang Lin (Beijing University of Posts and Telecommunications), Jitao Sang (Beijing University of Posts and Telecommunications)
Explainability and InterpretabilityComputational EfficiencyVision Language ModelMultimodality
🎯 What it does: Propose a single-channel, reference-free orthogonal subspace editing method called HulluEdit to address the problem of object hallucination in large vision-language models.
Human Geometry Distribution for 3D Animation Generation
Xiangjun Tang, Peter Wonka
GenerationConvolutional Neural NetworkDiffusion modelFlow-based ModelContrastive LearningGaussian SplattingMesh
🎯 What it does: Built a two-stage generative framework for generating high-quality human geometry animations rich in clothing dynamic details from noise.
Human Interaction-Aware 3D Reconstruction from a Single Image
Gwanghyun Kim (Seoul National University), Se Young Chun (Seoul National University)
GenerationDiffusion modelImageMesh
🎯 What it does: Reconstruct high-fidelity 3D models of multiple humans and generate complete textures from a single RGB image.
Human-Centric Multi-Exposure Fusion: Benchmark and Bi-level Cognition Distillation Framework
Jingjie Shang (Peking University), Xiaochen Bo (Peking University)
OptimizationKnowledge DistillationTransformerImageBiomedical DataBenchmark
🎯 What it does: Constructed the human EEG cognitive dataset Cog-Expo and proposed a bi-level optimized teacher-student framework, achieving multi-exposure fusion with EEG-guided teachers for EEG-free inference.
Human-like Abstract Visual Reasoning via Understanding and Solving Reasoning Loop
Xinwang Chen (Beijing Institute of Technology), Xia Wu (Beijing Institute of Technology)
Explainability and InterpretabilityRepresentation LearningTransformerContrastive LearningImageBenchmark
🎯 What it does: This paper proposes the Understanding-Solving Loop (USRL) framework to simulate human iterative understanding and solving processes in abstract visual reasoning tasks with extremely few samples, enhancing the model's ability to capture and apply implicit transformation rules.
HumanBA: Human-Aware Bundle Adjustment via Global Human-Camera Decoupling
Fengyuan Yang (National University of Singapore), Angela Yao (National University of Singapore)
Pose EstimationOptimizationSimultaneous Localization and MappingVideoMesh
🎯 What it does: This paper proposes HumanBA, a method that decouples dynamic human motion into camera motion and human body motion, and incorporates pseudo-static human joints as constraints into Bundle Adjustment;
HumanNOVA: Photorealistic, Universal and Rapid 3D Human Avatar Modeling from a Single Image
Hezhen Hu (University of Texas at Austin), Georgios Pavlakos (University of Texas at Austin)
GenerationData SynthesisTransformerNeural Radiance FieldGaussian SplattingImageMesh
🎯 What it does: Propose HumanNOVA, a model that can generate high-quality 3D human avatars in real-time from a single image.
Humanoid Generative Pre-Training for Zero-Shot Motion Tracking
Zekun Qi (Tsinghua University), Li Yi (Tsinghua University)
Object TrackingKnowledge DistillationRobotic IntelligenceTransformerLarge Language ModelReinforcement LearningVideo
🎯 What it does: This paper proposes Humanoid-GPT, a GPT-style transformer model, which achieves zero-shot tracking of arbitrary dynamic motions on full-body robots without any fine-tuning, through pre-training on 2 billion frames of relocalization motion corpus.
HumanVBench: Probing Human-Centric Video Understanding in MLLMs with Automatically Synthesized Benchmarks
Ting Zhou (Sun Yat-Sen University), Ying Shen (Sun Yat-Sen University)
Data SynthesisTransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmarkAudio
🎯 What it does: Propose the HUMANVBENCH video benchmark, which includes 16 fine-grained human-centric tasks, and develop an automated annotation and multiple-choice question generation pipeline.
HUMAPS-4D: A Multimodal Dataset for HUman Motion Analysis with Physiological and Semantic informations
Matthieu Dabrowski (IMT Nord Europe), Benjamin Allaert (IMT Nord Europe)
ClassificationPose EstimationRepresentation LearningVision-Language-Action ModelImageVideoMultimodalityTime SeriesBiomedical DataBenchmark
🎯 What it does: Created a large-scale multimodal human motion dataset HUMAPS-4D, containing synchronized motion capture, multi-view videos, IMU, foot pressure, sEMG, and semantic annotations;
HUMORCHAIN: Theory-Guided Multi-Stage Reasoning for Interpretable Multimodal Humor Generation
Jiajun Zhang (Peking University), Qi Su (Peking University)
GenerationLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Proposes the HUMORCHAIN framework, integrating multi-stage humorous reasoning to generate humorous image captions.
Hunting Normality from Query Sample via Residual Learning for Generalist Anomaly Detection
Xiaolei Wang (Xi'an Jiaotong-Liverpool University), Jimin Xiao (Xi'an Jiaotong-Liverpool University)
Anomaly DetectionTransformerImageBiomedical Data
🎯 What it does: This paper proposes a generic anomaly detection framework based on residual learning, which utilizes residual features to guide instance-level normality learning, achieving cross-domain anomaly detection and localization.
HVG-3D: Bridging Real and Simulation Domains for 3D-Conditional Hand-Object Interaction Video Synthesis
Mingjin Chen (Hong Kong Polytechnic University), Yi Wang (Hong Kong Polytechnic University)
GenerationDomain AdaptationTransformerDiffusion modelVideoPoint CloudSequential
🎯 What it does: Propose a 3D control-based hand-object interaction video synthesis framework called HVG-3D, which can generate high-quality, temporally consistent interactive videos from a single real image and 3D point clouds or trajectory conditions from any source.
Hybrid Agents for Image Restoration
Bingchen Li (University of Science and Technology of China), Zhibo Chen (University of Science and Technology of China)
RestorationTransformerLarge Language ModelSupervised Fine-TuningAgentic AIPrompt EngineeringImageMultimodality
🎯 What it does: Propose HybridAgent—a hybrid interactive image restoration framework combining fast/slow agents and feedback agents, capable of automatically identifying and handling mixed distortions;
Hybrid Robust Collaborative Perception with LiDAR-4D Radar Fusion under Adverse Weather Conditions
Yuquan Yang, Xiaohua Xu (University Of Science And Technology Of China)
Autonomous DrivingConvolutional Neural NetworkMultimodalityPoint Cloud
🎯 What it does: Propose a hybrid robust collaborative perception framework HRCP that integrates LiDAR and 4D radar fusion under adverse weather conditions to achieve multi-vehicle collaborative 3D object detection.
Hybrid Token Compression for Vision-Language Models
Jusheng Zhang (Sun Yat-sen University), Keze Wang (Sun Yat-sen University)
CompressionRepresentation LearningTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Propose HTC-VLM, a hybrid token compression framework that can compress visual images into a single token while preserving semantic and detail information.
HybridDriveVLA: Vision-Language-Action Model with Visual CoT reasoning and ToT Evaluation for Autonomous Driving
Yipene Cedric Francois Bassole (Dongguk University), Yunsick Sung (Dongguk University)
Autonomous DrivingLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Propose HybridDriveVLA, an end-to-end Vision-Language-Action model that predicts future scenes through visual Chain-of-Thought (V-CoT) and selects the optimal trajectory by evaluating multiple paths using Tree-of-Thought (ToT).
HyCal: A Training-Free Prototype Calibration Method for Cross-Discipline Few-Shot Class-Incremental Learning
Eunju Lee (Chung-Ang University), YoungBin Kim (Chung-Ang University)
ClassificationRecognitionDomain AdaptationComputational EfficiencyTransformerVision Language ModelImageBiomedical DataBenchmark
🎯 What it does: This paper investigates the Domain Gravity phenomenon generated by pre-trained vision-language models in few-shot incremental learning with imbalanced samples in heterogeneous domains, and proposes a training-agnostic hybrid prototype calibration method called HYCAL.
Hyper-PCN: Hypergraph-Based Point Cloud Completion via High-Order Correlation Modeling
Linfei Li (Tsinghua University), Yue Gao (Tsinghua University)
RestorationGenerationAutonomous DrivingGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: Designed and implemented a hypergraph-based point cloud completion framework, Hyper-PCN, which achieves complete and detailed 3D reconstruction by modeling high-order associations in incomplete point clouds.
Hyperbolic Busemann Neural Networks
Ziheng Chen (University of Trento), Nicu Sebe (University of Trento)
ClassificationImageGraphBiomedical Data
🎯 What it does: Proposed hyperbolic space polynomial logistic regression (BMLR) and fully connected layer (BFC) based on the Busemann function, which can directly operate on the Poincaré ball and Lorentz hyperbolic model, avoiding the parameter redundancy and low batch efficiency of traditional hyperplane MLR.
Hyperbolic Defect Feature Synthesis for Few-Shot Defect Classification
Huimin Li (Beihang University), Junlin Hu (Beihang University)
ClassificationAnomaly DetectionVision Language ModelContrastive LearningImageOrdinary Differential Equation
🎯 What it does: This paper proposes a defect feature synthesis method based on hyperbolic space (HypDFS), which constructs defect prototypes on a hyperbolic sphere, models defect distributions, generates synthetic features, further fuses pre-trained CLIP features using a residual adapter, and improves few-shot defect classification performance with a hierarchical contrastive loss.
Hyperbolic Gramian Volumes for Multimodal Alignment
Saiyang Na (University of Texas at Arlington), Junzhou Huang (University of Texas at Arlington)
RetrievalRepresentation LearningContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: Propose a HyperGRAM framework that extends the Gramian volume alignment method to hyperbolic space and mixes it with Euclidean space for zero-shot video-text retrieval.
Hyperbolic Prototype Learning with Uncertainty-Aware Consistency for Continual Test-Time Segmentation
Siddhant Gole (Indian Institute of Technology Bombay), Biplab Banerjee (Indian Institute of Technology Bombay)
SegmentationRepresentation LearningImage
🎯 What it does: Propose a continuous test-time semantic segmentation method that learns class prototypes in hyperbolic space and incorporates an uncertainty consistency adaptation mechanism, named HyperProtoSeg + HBCA.
Hyperbolic Relational Prompts for Intersectional Fairness in Medical VLMs
Jiayu Qian (City University of Hong Kong), Zhi-An Huang (City University of Hong Kong)
Graph Neural NetworkPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodalityBiomedical DataElectronic Health Records
🎯 What it does: Proposed a fair-aware relational prompt framework FRP to address fairness bias in medical vision-language models across race and gender intersectional groups.
HyperGait: Unleashing the Power of Parsing for Gait Recognition in the Wild via Hypergraph
Jinkai Zheng (Hangzhou Dianzi University), Wu Liu (University of Science and Technology of China)
RecognitionConvolutional Neural NetworkGraph Neural NetworkVideo
🎯 What it does: Propose the HyperGait framework, which leverages hypergraph convolution to learn high-order spatiotemporal features on gait parsing sequences, thereby improving gait recognition accuracy.
HyperGaussians: High-Dimensional Gaussian Splatting for High-Fidelity Animatable Face Avatars
Gent Serifi (ETH Zurich), Marcel C. Buehler (ETH Zurich)
GenerationGaussian SplattingVideo
🎯 What it does: Propose HyperGaussians, extending 3D Gaussian Splatting to high-dimensional multivariate Gaussians and using local embeddings for conditioning, achieving higher-resolution modeling of facial expressions, lighting, and details.
Hypergraph-State Collaborative Reasoning for Multi-Object Tracking
Zikai Song (Huazhong University of Science and Technology), Xinchao Wang (La Trobe Uniersity)
Object TrackingGraph Neural NetworkVideo
🎯 What it does: Propose a collaborative motion estimation framework based on hypergraphs and state space models (HyperSSM) for motion prediction and association in multi-target tracking.
HyperNAS: Enhancing Architecture Representation for NAS Predictor via Hypernetwork
Jindi Lv (Sichuan University), Jiancheng Lv (Sichuan University)
Representation LearningNeural Architecture SearchGraph Neural NetworkImageBenchmark
🎯 What it does: Built the HyperNAS predictor, combining global encoding and shared hypernetworks to achieve precise prediction and parameter generation of architectural performance.
HyperST: Hierarchical Hyperbolic Learning for Spatial Transcriptomics Prediction
Chen Zhang (National Institute for Data Science in Health and Medicine, Xiamen University), Rongshan Yu (National Institute for Data Science in Health and Medicine, Xiamen University)
Representation LearningSupervised Fine-TuningContrastive LearningImageBiomedical Data
🎯 What it does: Propose the HyperST framework, which utilizes multi-level image and gene feature extraction to align in a hyperbolic space for predicting spatial transcriptomics data.
HypeVPR: Exploring Hyperbolic Space for Perspective to Equirectangular Visual Place Recognition
Suhan Woo (Yonsei University), Euntai Kim (Yonsei University)
RetrievalContrastive LearningImage
🎯 What it does: Proposed and implemented HypeVPR, a hierarchical embedding framework based on hyperbolic space, for solving the visual place recognition from viewpoint to panorama (P2E VPR) problem.
HySeg: Learning Generative Priors for Structure-Aware Remote Sensing Segmentation
Jie Qiu (Fujian Agriculture and Forestry University), Jianzhang Chen (Fujian Agriculture and Forestry University)
SegmentationConvolutional Neural NetworkTransformerFlow-based ModelImage
🎯 What it does: Explores the introduction of generative structural priors in remote sensing image semantic segmentation, constructing a generative-discriminative hybrid framework to achieve structure-aware segmentation.
I-Scene: 3D Instance Models are Implicit Generalizable Spatial Learners
Lu Ling (Purdue University), Aniket Bera (NVIDIA Research)
GenerationData SynthesisDiffusion modelRectified FlowMesh
🎯 What it does: Reprogramming pre-trained single-instance generation models into scene-level spatial learners by leveraging perspective-centric space and scene context attention to achieve unsupervised spatial relation learning;
I'm a Map! Interpretable Motion-Attentive Maps: Spatio-Temporally Localizing Concepts in Video Diffusion Transformers
Youngjun Jun (Yonsei University), Seong Jae Hwang (Yonsei University)
SegmentationExplainability and InterpretabilityTransformerDiffusion modelVideo
🎯 What it does: For video diffusion Transformers, a training-free and gradient-free motion attention mapping (IMAP) is proposed, which can perform spatiotemporal localization of any motion concept in videos.
I2I-Bench: A Comprehensive Benchmark Suite for Image-to-Image Editing Models
Juntong Wang (Shanghai Jiao Tong University), Xiongkuo Min (Shanghai Jiao Tong University)
Image TranslationLarge Language ModelPrompt EngineeringVision Language ModelImageTextBenchmark
🎯 What it does: Propose I2I-Bench, a comprehensive image editing benchmark containing 1000 diverse instructions and 30 fine-grained evaluation dimensions, systematically evaluating single-image and multi-image editing models while open-sourcing data and code.
IAFMNet: Information-Aware Feature Modulation for Efficient Super-Resolution
Junwei Xu (Xidian University), Weisheng Dong (Xidian University)
RestorationSuper ResolutionConvolutional Neural NetworkImageBenchmark
🎯 What it does: This paper proposes a lightweight super-resolution network called IAFMNet based on information density estimation, which uses information entropy to learn an information density map to guide feature enhancement and sparse convolution, achieving efficient single-image super-resolution.
IAG: Input-aware Backdoor Attack on VLM-based Visual Grounding
Junxian Li (Shanghai Jiao Tong University), Di Zhang (Fudan University)
Adversarial AttackConvolutional Neural NetworkVision Language ModelMultimodality
🎯 What it does: This paper proposes a text-guided input-aware reverse gated attack (IAG), which can dynamically generate stealthy triggers in visual-linguistic models (VLM) for visual localization tasks, forcing the model to ignore user queries and forcibly localize any specified target.
IBISAgent: Reinforcing Pixel-Level Visual Reasoning in MLLMs for Universal Biomedical Object Referring and Segmentation
Yankai Jiang (Zhejiang University Shanghai Artificial Intelligence Laboratory), Jianwei Yin (Zhejiang University Shanghai Artificial Intelligence Laboratory)
SegmentationSupervised Fine-TuningReinforcement LearningVision Language ModelVision-Language-Action ModelBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Propose a multi-modal large language model (MLLM) agent named IBISAgent, which utilizes multi-step decision-making (Markov decision process) for pixel-level visual reasoning and segmentation on medical images, and supports complex tasks such as mask refinement.
ICTPolarReal: A Polarized Reflection and Material Dataset of Real World Objects
Jing Yang (Institute for Creative Technologies University of Southern California), Yajie Zhao (Institute for Creative Technologies University of Southern California)
Data SynthesisImageBenchmark
🎯 What it does: This paper constructs a comprehensive real-world polarized reflection and material dataset containing 218 daily objects, 346 lighting conditions, 8 viewpoints, and approximately 1.2 million high-resolution images.
ID-Crafter: VLM-Grounded Online RL for Compositional Multi-Subject Video Generation
Panwang Pan (Xiamen University), Yadong Mu (Xiamen University)
GenerationTransformerReinforcement LearningVision Language ModelDiffusion modelAuto EncoderVideoTextMultimodality
🎯 What it does: Developed the ID-Crafter framework for multi-agent video generation, maintaining identity consistency, semantic coherence, and temporal consistency.
ID-Sim: An Identity-Focused Similarity Metric
Julia Chae (MIT), Cusuh Ham (Adobe Research)
RecognitionData SynthesisRetrievalTransformerContrastive LearningImageBenchmark
🎯 What it does: Proposes an identity-focused similarity metric ID-Sim based on Transformer, which is highly sensitive to subtle identity changes while maintaining invariance to background, pose, and lighting.
Identity-Preserving Image-to-Video Generation via Reward-Guided Optimization
Liao Shen (Huazhong University of Science and Technology), Bo Zheng (Taobao & Tmall Group of Alibaba)
GenerationTransformerReinforcement LearningPrompt EngineeringDiffusion modelVideo
🎯 What it does: Propose an IPRO framework based on reward-guided methods, which directly optimizes the sampling process of image-to-video diffusion models through facial identity rewards, thereby enhancing identity consistency in videos.
IDESplat: Iterative Depth Probability Estimation for Generalizable 3D Gaussian Splatting
Wei Long (University Of Electronic Science And Technology Of China), Shuhang Gu (University Of Electronic Science And Technology Of China)
Depth EstimationComputational EfficiencyGaussian SplattingImage
🎯 What it does: Proposed a generalizable 3D Gaussian scattering model called IDESplat, which achieves multi-level fusion of depth probability through iterative warp and depth probability boosting unit (DPBU), thereby accurately predicting Gaussian means;
IDperturb: Enhancing Variation in Synthetic Face Generation via Angular Perturbations
Fadi Boutros, Naser Damer (Fraunhofer Institute for Computer Graphics Research IGD)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Proposed the IDPERTURB method, which enhances intra-class diversity of synthetic faces by performing angle-restricted spherical cap sampling within the identity embedding space, generating more generalizable training data.
IEBGL:An Interpretability-Enhanced Brain Graph Learning Framework with LLM-Instructed Topology and Literature-Augmented Semantics
Yihang Duan (Nanjing Forestry University), Li Zhang (Nanjing Forestry University)
ClassificationExplainability and InterpretabilityGraph Neural NetworkTransformerLarge Language ModelContrastive LearningTextGraphBiomedical DataMagnetic Resonance ImagingRetrieval-Augmented Generation
🎯 What it does: Propose an interpretable enhanced brain mapping learning framework, IEBGL, which improves the topology and node features of rs-fMRI brain networks for brain disease diagnosis through Topology Reconstruction via LLM Instructions (LITR) and Semantic Aggregation of Medical Literature (LASA).
IF-Bench: Benchmarking and Enhancing MLLMs for Infrared Images with Generative Visual Prompting
Tao Zhang (MAIS, Institute of Automation), Chunhong Pan (MAIS, Institute of Automation)
Large Language ModelPrompt EngineeringMixture of ExpertsVision Language ModelImageMultimodalityBenchmark
🎯 What it does: This study first constructs the first high-quality infrared image multimodal understanding benchmark, IF-Bench, containing 499 infrared images and 680 visual question answering (VQA) pairs; subsequently, it systematically evaluates the infrared image understanding capabilities of over 40 open-source and closed-source multimodal large language models (MLLMs); finally, it proposes a training-agnostic generated visual prompting (GenViP) method, which converts infrared images into semantically aligned RGB images using image editing models and inputs them together with the original images, significantly enhancing the performance of various MLLMs on infrared tasks.
IF-Prune: Information-Flow Guided Token Pruning for Efficient Vision-Language Models
Guohao Sun (Snap Inc), Jian Wang (Snap Inc)
Computational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: Proposed an information-flow-based posterior-guided visual token pruning framework called IF-Prune, which estimates the importance of each visual token in a single forward pass using a lightweight information bottleneck fine-tuned small VLM, and applies the pruning results to large VLMs.
IFCSR: Inference-Free Fidelity-Realism Control for One-Step Diffusion-based Real-World Image Super-Resolution
Jonghee Back (BLUEDOT Inc), Minyong Jeon (BLUEDOT Inc)
Super ResolutionConvolutional Neural NetworkDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes IFCSR, a real image super-resolution method based on first-order diffusion, which can freely switch between fidelity and realism through adjustable parameters without additional inference.
IGen: Scalable Data Generation for Robot Learning from Open-World Images
Chenghao Gu (Tsinghua University), Zhi Wang
Data SynthesisRobotic IntelligenceLarge Language ModelVision Language ModelImageTextPoint Cloud
🎯 What it does: Propose the IGen framework, which generates 3D scenes from open-world images, plans robot trajectories, renders corresponding visual observations, and produces visual-action pairs directly usable for training;
Illuminating Visual Identity in Universal Multimodal Embeddings
Jiawei Cao (Shanghai Jiao Tong University), Jieping Ye (Alibaba Group)
RecognitionRetrievalRepresentation LearningLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Propose a visual identity discrimination (VisID) framework, define four types of meta-task, and construct a large-scale MVEB benchmark; propose a unified identity-aware sampling and contrastive learning training method, enabling the general multimodal embedding (UME) to simultaneously possess strong identity recognition and cross-modal retrieval capabilities.
Illumination-Consistent Human-Scene Reconstruction from Monocular Video
Rongbin Zheng (Sun Yat-Sen University), Chengying Gao (Sun Yat-Sen University)
Data SynthesisGaussian SplattingVideoMesh
🎯 What it does: Jointly reconstruct relightable 3D humans and static scenes from monocular videos, achieving consistency in dynamic shadows and spatially varying illumination.
Illustrator's Depth: Monocular Layer Index Prediction for Image Decomposition
Nissim Maruani (Inria), Wang Yifan (Adobe Research)
SegmentationDepth EstimationTransformerImage
🎯 What it does: This paper proposes the concept of 'Illustrator's Depth,' utilizing a neural network to predict the layer index for each pixel in a single image, enabling image layering and editing;
iLRM: An Iterative Large 3D Reconstruction Model
Gyeongjin Kang (Sungkyunkwan University), Eunbyung Park (Yonsei University)
GenerationTransformerGaussian SplattingImage
🎯 What it does: Propose an iterative large-scale 3D reconstruction model called iLRM, which realizes a feed-forward 3D Gaussian splatting framework that combines multi-view images with iterable perspective embeddings.
Image Diffusion Preview with Consistency Solver
Fu-Yun Wang (Google DeepMind), Long Zhao (Google DeepMind)
GenerationComputational EfficiencyReinforcement LearningDiffusion modelImageOrdinary Differential Equation
🎯 What it does: Propose the Diffusion Preview framework and design ConsistencySolver as a trainable high-order ODE solver to generate high-quality previews with low-step sampling, enabling an interactive image generation workflow.
Image Generation from Contextually-Contradictory Prompts
Saar Huberman (Tel Aviv University), Daniel Cohen-Or (Tel Aviv University)
GenerationLarge Language ModelPrompt EngineeringDiffusion modelImageText
🎯 What it does: Propose a training-agnostic Stage-Aware Prompting (SAP) method that leverages large language models to automatically split and inject proxy prompts at different denoising stages, addressing context contradictions in text-to-image generation.
Image Guides Images: Consistent Video Amodal Completion with Rectified In-Context Exemplar Guidance
Xiaoyu Kong (Beihang University), Miao Wang (Beihang University)
RestorationTransformerVision Language ModelDiffusion modelRectified FlowVideo
🎯 What it does: Propose a zero-shot framework IC-Amodal that achieves video amodal completion using a pre-trained image inpainting model and visual context learning, ensuring spatiotemporal consistency.
Image-based Outlier Synthesis With Training Data
Sudarshan Regmi (Dartmouth)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: Propose a unified out-of-distribution (OOD) detection framework named ASCOOD without requiring external data, which generates virtual out-of-distribution samples by leveraging gradient attribution values of in-distribution (ID) samples, and subsequently jointly optimizes ID classification and prediction uncertainty for out-of-sample during training.
Image-Guided Geometric Stylization of 3D Meshes
Changwoon Choi (Seoul National University), Young Min Kim (Seoul National University)
GenerationDiffusion modelScore-based ModelAuto EncoderImageMesh
🎯 What it does: Propose a 3D mesh geometry stylization framework based on image style reference, which extracts style through LoRA fine-tuning of Diffusion models, and employs Score Distillation Sampling (SDS) combined with an approximate VAE encoder to guide Jacobian-based coarse-to-fine hierarchical deformation while maintaining mesh topology and semantic consistency;
Image-to-Point Cloud Feature Back-Projection for Multimodal Training of 3D Semantic Segmentation
Jiawei Han (Beijing Institute Of Technology), Wei Li (Beijing Institute Of Technology)
SegmentationDepth EstimationAutonomous DrivingKnowledge DistillationImageMultimodalityPoint Cloud
🎯 What it does: Achieved seamless fusion of images and point clouds by back-projecting aggregated image features with depth information into the 3D space of LiDAR point clouds, and jointly training them in a single-branch network;
ImageRAGTurbo: Towards One-step Text-to-Image Generation with Retrieval-Augmented Diffusion Models
Peijie Qiu (Washington University in St. Louis), Rahul Bhagat (Amazon Core Search)
GenerationRetrievalDiffusion modelMultimodalityRetrieval-Augmented Generation
🎯 What it does: Achieved one-step text-to-image generation by directly injecting retrieved text-image pairs into the high-level features of UNet via a retrieval-enhanced H-space adapter, significantly improving the image quality and text alignment of few-step diffusion models.
Imagine Before Concentration: Diffusion-Guided Registers Enhance Partially Relevant Video Retrieval
Jun Li, Bin Chen (Harbin Institute Of Technology)
RetrievalVision Language ModelDiffusion modelVideoTextMultimodality
🎯 What it does: Propose the DreamPRVR framework, which first generates global registration through a text-supervised diffusion model, and then refines video representations using registration-enhanced Gaussian Attention to address partial relevance video retrieval problems.
IMAIA: Interactive Maps AI Assistant for Travel Planning and Geo-Spatial Intelligence
Jieren Deng (Microsoft), Chiqun Zhang (Microsoft)
Data-Centric LearningLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Proposed the IMAIA framework, achieving a unified interactive map AI assistant for desktop map exploration and on-site navigation near the endpoint (last-100m);
Imbalanced View Contribution Evaluation and Refinement for Deep Incomplete Multi-View Clustering
Taichun Zhou (National University of Defense Technology), Kunlun He (Chinese PLA General Hospital)
OptimizationExplainability and InterpretabilityRepresentation LearningData-Centric LearningAuto EncoderMultimodality
🎯 What it does: This paper proposes an Imbalanced View Contribution Evaluation and Refinement (ICER) framework specifically designed to address the issue of contribution imbalance caused by missing views in incomplete multi-view clustering.
ImmerIris: A Large-Scale Dataset and Benchmark for Off-Axis and Unconstrained Iris Recognition in Immersive Applications
Yuxi Mi (Fudan University), Shuigeng Zhou (Fudan University)
RecognitionConvolutional Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: Constructed a large-scale, off-axis, unconstrained immersive iris dataset (ImmerIris) and proposed an end-to-end recognition paradigm that does not use traditional iris normalization steps.
Immunizing Models Against Harmful Long-Horizon Fine-Tuning via Contractive Optimization Dynamics
Najibul Haque Sarker (Virginia Tech), Chris Thomas (Virginia Tech)
OptimizationSafty and PrivacyMeta LearningImageTextBenchmark
🎯 What it does: Prevent models from harmful fine-tuning over long cycles by applying contractive contraction constraints on pre-trained models.
iMontage: Unified, Versatile, Highly Dynamic Many-to-many Image Generation
Zhoujie Fu (Nanyang Technological University), Guosheng Lin (Nanyang Technological University)
GenerationData SynthesisDiffusion modelFlow-based ModelAuto EncoderImageVideoMultimodality
🎯 What it does: Proposes iMontage, a unified multi-input multi-output image generation framework that generates high dynamic and consistent image sets using pre-trained video diffusion models.
Improved Mean Flows: On the Challenges of Fastforward Generative Models
Zhengyang Geng (Carnegie Mellon University), Kaiming He (Massachusetts Institute of Technology)
GenerationTransformerDiffusion modelFlow-based ModelAuto EncoderImage
🎯 What it does: This paper proposes an improved version of MeanFlow (iMF), achieving more stable training and flexible guidance during one-step generation by rewriting the training objective as regression of instantaneous velocity and allowing conditional CFG.
Improving Adversarial Transferability with Local Perturbation Augmentation
Jian-Xun Mi (Chongqing University of Posts and Telecommunications), Weisheng Li (Chongqing University of Posts and Telecommunications)
Adversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposes Local Perturbation Augmentation Attack (LPAA), which enhances the cross-model transferability of adversarial examples by performing local perturbation augmentation within multiple subspaces and providing a specialized perturbation initialization strategy.
Improving Calibration in Test-Time Prompt Tuning for Vision-Language Models via Data-Free Flatness-Aware Prompt Pretraining
Hyeonseo Jang (Yonsei University), Kibok Lee (Yonsei University)
OptimizationTransformerPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose an unsupervised pre-training framework (FPP) that initializes prompts to flatter loss landscape regions before test-time prompt tuning (TPT), aiming to improve the calibration performance of vision-language models.
Improving Controllable Generation: Faster Training and Better Performance via x0-Supervision
Amadou S. Sangare (Universit' e Paris-Saclay), Bertrand Luvison (Universit' e Paris-Saclay)
GenerationDiffusion modelFlow-based ModelImage
🎯 What it does: This paper proposes a new supervision method called x₀-supervision to accelerate the training of controllable image generation models (diffusion/flow matching) and improve the quality and control accuracy of the final generated images.
Improving Diffusion Generalization with Weak-to-Strong Segmented Guidance
Liangyu Yuan (Westlake University), Chi Zhang (Westlake University)
GenerationDiffusion modelImage
🎯 What it does: The paper proposes a Segmented Guidance (SGG) mechanism that combines conditional dependent and conditional independent guidance to enhance the generalization performance of diffusion models and migrate the weak-to-strong guidance principle into the training phase;
Improving Motion in Image-to-Video Models via Adaptive Low-Pass Guidance
June Suk Choi (KAIST), Kimin Lee (KAIST)
Image TranslationGenerationDiffusion modelFlow-based ModelVideo
🎯 What it does: Enhance video motion dynamics through adaptive low-pass filtering during the sampling process of the I2V model
Improving Sparse Autoencoder with Dynamic Attention
Dongsheng Wang (Shenzhen University), Hui Huang (Shenzhen University)
ClassificationRepresentation LearningTransformerAuto EncoderImageTextMultimodality
🎯 What it does: This paper proposes a sparse autoencoder based on Transformer (Sparsemax SAE), which utilizes the sparsemax attention mechanism to automatically determine the required number of concepts in each feature vector during the autoencoding process, thereby achieving higher quality concept learning and reconstruction.
Improving Text-to-Image Generation with Intrinsic Self-Confidence Rewards
Seungwook Kim (Pohang University of Science and Technology), Minsu Cho (Pohang University of Science and Technology)
Image TranslationGenerationReinforcement LearningDiffusion modelRectified FlowImageTextMultimodality
🎯 What it does: Post-training of text-to-image diffusion/flow matching models using the model's own confidence (noise restoration error) as a reward signal to enhance generation quality.