ICCV 2025 Papers — Page 12
IEEE/CVF International Conference on Computer Vision · 2701 papers
Heatmap Regression without Soft-Argmax for Facial Landmark Detection
Chiao-An Yang (Purdue University), Raymond A. Yeh (Purdue University)
RecognitionPose EstimationImage
🎯 What it does: A new facial keypoint detection method is proposed, avoiding the use of Soft-argmax and achieving heatmap regression through a structured prediction-based training objective.
Heavy Labels Out! Dataset Distillation with Label Space Lightening
Ruonan Yu (National University of Singapore), Xinchao Wang (National University of Singapore)
CompressionKnowledge DistillationContrastive LearningImage
🎯 What it does: The paper proposes a lightweight label generation framework called HeLlO, which achieves online soft label generation during dataset distillation through an image-to-label projector, significantly reducing label storage costs.
Height-Fidelity Dense Global Fusion for Multi-modal 3D Object Detection
Hanshi Wang (Chinese Academy of Sciences), Zhipeng Zhang (Shanghai Jiao Tong University)
Object DetectionAutonomous DrivingMultimodalityPoint Cloud
🎯 What it does: A global dense fusion framework based on Hybrid Mamba Block and high-fidelity LiDAR encoding is proposed for multimodal 3D object detection.
HERMES: A Unified Self-Driving World Model for Simultaneous 3D Scene Understanding and Generation
Xin Zhou (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
GenerationAutonomous DrivingTransformerLarge Language ModelWorld ModelImagePoint Cloud
🎯 What it does: A unified driving world model HERMES is proposed, capable of simultaneous 3D scene understanding and future scene generation.
HERMES: temporal-coHERent long-forM understanding with Episodes and Semantics
Gueter Josmy Faure (National Taiwan University), Winston H. Hsu (National Taiwan University)
ClassificationRecognitionRetrievalCompressionComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideo
🎯 What it does: The HERMES framework is proposed, which includes two pluggable modules: the Episodic COmpressor (ECO) for compressing continuous frame sequences, and the Semantics reTRiever (SeTR) for extracting global semantic features. These two types of features are combined and input into a large language model to achieve long-term video understanding.
HERO: Human Reaction Generation from Videos
Chengjun Yu (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
GenerationData SynthesisTransformerVideo
🎯 What it does: A framework called HERO is proposed for generating 3D human response actions from RGB videos, supporting various interaction scenarios such as human-human, human-animal, and scene-human interactions, and can generate corresponding response actions based on emotional information in the video.
Heuristic-Induced Multimodal Risk Distribution Jailbreak Attack for Multimodal Large Language Models
Teng Ma (Sun Yat-Sen University), Wenqi Ren (Guangdong Key Laboratory of Information Security Technology)
Adversarial AttackTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageTextMultimodality
🎯 What it does: A black-box multimodal jailbreak method called HIMRD is proposed, which first splits malicious prompts into text and image components and embeds them separately. It then constructs understanding-enhancing prompts and inducing prompts through heuristic search, inducing multimodal large language models to reconstruct malicious semantics and output violating content.
HFD-Teacher: High-Frequency Depth Distillation from Depth Foundation Models for Enhanced Depth Completion
Zhiyuan Yang (Nanyang Technological University), Kezhi Mao (SIMTech, Agency for Science, Technology and Research)
RestorationDepth EstimationKnowledge DistillationTransformerImage
🎯 What it does: A teacher-student framework is proposed to enhance detail reconstruction in sparse depth completion through high-frequency knowledge distillation.
Hi-Gaussian: Hierarchical Gaussians under Normalized Spherical Projection for Single-View 3D Reconstruction
Binjian Xie (Chinese Academy of Sciences), Yihong Wu (Chinese Academy of Sciences)
GenerationDepth EstimationGaussian SplattingPoint Cloud
🎯 What it does: Proposes a Hi-Gaussian single-view 3D reconstruction method that can reconstruct a complete scene at once and supports novel view rendering.
Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridging
Chongjie Ye, Xiaoguang Han (Chinese University of Hong Kong, Shenzhen)
GenerationData SynthesisDiffusion modelImagePoint CloudMesh
🎯 What it does: The Hi3DGen framework is proposed, which decomposes a single image into three main modules through normal bridging: image → normal estimation, normal → geometry generation, and 3D data synthesis, achieving high-fidelity 3D geometric reconstruction.
Hierarchical 3D Scene Graphs Construction Outdoors
Jon Nyffeler (ETH Zurich), Daniel Barath (ETH Zurich)
Object DetectionSegmentationRetrievalLarge Language ModelPoint CloudGraph
🎯 What it does: A pipeline for constructing layered 3D scene graphs of outdoor scenes based on pose-aware images and 3D reconstruction is proposed, capable of automatically generating hierarchical maps for multi-scale objects from buildings and facades to windows.
Hierarchical Cross-modal Prompt Learning for Vision-Language Models
Hao Zheng (South China Normal University), Zhenhua Huang (Shenzhen Polytechnic University)
ClassificationTransformerPrompt EngineeringVision Language ModelImageText
🎯 What it does: A hierarchical cross-modal prompt learning framework HiCroPL is proposed, which improves the interaction between text and visual prompts through bidirectional knowledge flow and hierarchical knowledge mapping, addressing the issues of modality isolation and hierarchical semantic decay in traditional methods.
Hierarchical Divide-and-Conquer Grouping for Classification Adaptation of Pre-Trained Models
Ziqian Lu (Zhejiang Sci-Tech University), Jun Liu (Zhejiang University)
ClassificationData SynthesisDomain AdaptationTransformerLarge Language ModelSupervised Fine-TuningDiffusion modelTextMultimodality
🎯 What it does: A hierarchical divide-and-conquer strategy (HDG) is proposed, which first automatically divides the label space into base class and new class subspaces, and then recursively clusters within each subspace using text embeddings, combined with LLM + diffusion model to synthesize diverse new class samples and fine-tune lightweight classifiers specific to each subspace;
Hierarchical Event Memory for Accurate and Low-latency Online Video Temporal Grounding
Minghang Zheng (Wangxuan Institute of Computer Technology, Peking University), Yang Liu (Peking University)
RecognitionObject DetectionTransformerVideo
🎯 What it does: A hierarchical event memory framework is proposed, achieving accurate and low-latency predictions for online video temporal localization through event-level proposals and future prediction branches.
Hierarchical Material Recognition from Local Appearance
Matthew Beveridge (Columbia University), Shree K. Nayar (Columbia University)
ClassificationRecognitionGraph Neural NetworkImage
🎯 What it does: This paper proposes a hierarchical material classification system based on physical properties and uses Graph Attention Networks (GAT) for hierarchical material recognition of local appearances.
Hierarchical Variational Test-Time Prompt Generation for Zero-Shot Generalization
Zhaoyang Wu (Xidian University), Wenping Ma (Xidian University)
ClassificationDomain AdaptationTransformerPrompt EngineeringAuto EncoderContrastive LearningImageText
🎯 What it does: A hierarchical variational test-time prompt generation framework is proposed, which dynamically generates text and visual prompts during the inference phase to achieve adaptive reasoning for zero-shot tasks.
Hierarchical Visual Prompt Learning for Continual Video Instance Segmentation
Jiahua Dong (Mohamed bin Zayed University of Artificial Intelligence), Fahad Shahbaz Khan (Mohamed bin Zayed University of Artificial Intelligence)
Object DetectionSegmentationTransformerPrompt EngineeringVideo
🎯 What it does: This paper proposes the Continuous Video Instance Segmentation (CVIS) task and designs a Hierarchical Visual Prompt Learning (HVPL) model.
Hierarchical-aware Orthogonal Disentanglement Framework for Fine-grained Skeleton-based Action Recognition
Haochen Chang (Sun Yat-sen University), Erwei Yin (Sun Yat-sen University)
RecognitionGraph Neural NetworkContrastive LearningVideo
🎯 What it does: A hierarchical perception orthogonal decoupling framework HiOD is proposed for fine-grained skeleton action recognition.
Hierarchy UGP: Hierarchy Unified Gaussian Primitive for Large-Scale Dynamic Scene Reconstruction
Hongyang Sun (Zhejiang University), Sida Peng (Zhejiang University)
GenerationAutonomous DrivingComputational EfficiencyGaussian SplattingImagePoint Cloud
🎯 What it does: A Hierarchical Unified Gaussian Primitive (Hierarchy UGP) framework is proposed for high-quality, real-time rendering of large-scale dynamic urban scene reconstruction.
Hierarchy-Aware Pseudo Word Learning with Text Adaptation for Zero-Shot Composed Image Retrieval
Zhe Li (University of Science and Technology of China), Zhendong Mao (University of Science and Technology of China)
RetrievalTransformerContrastive LearningImageText
🎯 What it does: A hierarchical perception dynamic pseudo-word learning framework (HIT) is proposed, which learns multi-layer pseudo-words in zero-shot synthesized image retrieval tasks through semantic hierarchical parsing and text adaptive filtering.
HiERO: Understanding the Hierarchy of Human Behavior Enhances Reasoning on Egocentric Videos
Simone Alberto Peirone (Politecnico di Torino), Giuseppe Averta (Politecnico di Torino)
RecognitionRetrievalRepresentation LearningGraph Neural NetworkVideoText
🎯 What it does: A weakly supervised hierarchical graph model HiERO is proposed, which aligns video segments with narrative text and discovers functional threads of human behavior through spectral clustering, thereby enhancing the self-understanding ability of perspective videos.
HiGarment: Cross-modal Harmony Based Diffusion Model for Flat Sketch to Realistic Garment Image
Junyi Guo (Xi'an Jiaotong Liverpool University), Dongming Lu (Zhejiang University)
Image TranslationGenerationTransformerDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes the task of generating realistic clothing images from flat sketches and text prompts, termed FS2RG, and introduces the HiGarment framework based on this task.
High-Precision 3D Measurement of Complex Textured Surfaces Using Multiple Filtering Approach
Yuchong Chen (Southeast University), Feipeng Da (Southeast University)
Depth EstimationOptimizationComputational EfficiencyOptical FlowImagePoint Cloud
🎯 What it does: A multiple filtering algorithm (MFA) is proposed in structured light 3D measurement, which significantly reduces measurement errors caused by complex textures by applying multiple Gaussian blur processes to the captured fringe images and correcting phase errors using an error model.
High-Resolution Spatiotemporal Modeling with Global-Local State Space Models for Video-Based Human Pose Estimation
Runyang Feng (Jilin University), Yixing Gao (Jilin University)
Pose EstimationTransformerSupervised Fine-TuningVideo
🎯 What it does: This paper proposes the GLSMamba framework, which utilizes a global and local high-resolution spatiotemporal state space model for video human pose estimation.
Highlight What You Want: Weakly-Supervised Instance-Level Controllable Infrared-Visible Image Fusion
Zeyu Wang (Dalian Minzu University), Haoran Duan (Tsinghua University)
Image TranslationObject DetectionTransformerImageTextMultimodality
🎯 What it does: Achieved text instruction-controlled infrared-visible image fusion through weakly supervised two-stage training, and realized instance-level target highlighting in the fusion results.
HiMTok: Learning Hierarchical Mask Tokens for Image Segmentation with Large Multimodal Model
Tao Wang (Uni-Ubi), Wuyue Zhao (Uni-Ubi)
Object DetectionSegmentationTransformerLarge Language ModelPrompt EngineeringImageMultimodality
🎯 What it does: This paper proposes the HiMTok hierarchical mask tokenizer, which compresses segmentation masks into a maximum of 32 one-dimensional tokens, enabling large multimodal models to directly generate masks through language and decode them into fine-grained segmentation.
HiNeuS: High-fidelity Neural Surface Mitigating Low-texture and Reflective Ambiguity
Yida Wang (Li Auto Inc), Xianpeng Lang (Li Auto Inc)
RestorationGenerationNeural Radiance FieldPoint CloudMesh
🎯 What it does: We propose HiNeuS, a unified neural surface reconstruction framework for high-fidelity reconstruction that addresses issues of reflection, textureless regions, and detail-geometry conflicts.
Hints of Prompt: Enhancing Visual Representation for Multimodal LLMs in Autonomous Driving
Hao Zhou (University of Chinese Academy of Sciences), Honggang Qi (University of Chinese Academy of Sciences)
Domain AdaptationAutonomous DrivingTransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoMultimodality
🎯 What it does: The HoP framework is proposed, which enhances the visual representation and reasoning capabilities of multimodal large models in autonomous driving visual question answering by adding three types of prompts: Affinity, Semantic, and Question.
HiP-AD: Hierarchical and Multi-Granularity Planning with Deformable Attention for Autonomous Driving in a Single Decoder
Yingqi Tang (Nullmax), Erkang Cheng (Nullmax)
Autonomous DrivingTransformerPoint Cloud
🎯 What it does: A unified decoder end-to-end autonomous driving framework HiP-AD is proposed, integrating perception, prediction, and planning tasks, and achieving multi-granularity trajectory prediction through multi-scale planning queries and a deformable attention mechanism.
Hipandas: Hyperspectral Image Joint Denoising and Super-Resolution by Image Fusion with the Panchromatic Image
Shuang Xu (Northwestern Polytechnical University), Deyu Meng (Macau University of Science and Technology)
RestorationSuper ResolutionImage
🎯 What it does: This paper proposes an end-to-end unsupervised framework called Hipandas, which jointly uses panchromatic images to denoise and recover super-resolution from low-resolution, noisy hyperspectral images.
HIS-GPT: Towards 3D Human-In-Scene Multimodal Understanding
Jiahe Zhao (Institute of Computing Technology, Chinese Academy of Sciences), Shiguang Shan (Institute of Computing Technology, Chinese Academy of Sciences)
Object DetectionPose EstimationTransformerLarge Language ModelSupervised Fine-TuningMultimodalityPoint CloudBenchmark
🎯 What it does: Proposes the Human-In-Scene Question Answering (HIS-QA) task and the corresponding multimodal benchmark HIS-Bench, and designs the HIS-GPT model based on this task, which can simultaneously analyze 3D scenes and human movements;
HOLa: Zero-Shot HOI Detection with Low-Rank Decomposed VLM Feature Adaptation
Qinqian Lei (National University of Singapore), Robby T. Tan
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: The HOLa method is proposed, which enhances zero-shot human-object interaction detection by adapting weights through low-rank decomposition of VLM text features, while combining LLM-generated action regularization and human-object tokens to improve unseen class generalization and action differentiation.
Holistic Tokenizer for Autoregressive Image Generation
Anlin Zheng (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
Image TranslationRestorationGenerationTransformerLarge Language ModelImage
🎯 What it does: A global-local visual tokenizer named Hita is proposed, which is seamlessly integrated with autoregressive (AR) image generation models like Llama to achieve image generation, style transfer, and unsupervised restoration.
Holistic Unlearning Benchmark: A Multi-Faceted Evaluation for Text-to-Image Diffusion Model Unlearning
Saemi Moon (POSTECH), Dongwoo Kim (POSTECH)
GenerationData SynthesisLarge Language ModelVision Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: Proposes the Holistic Unlearning Benchmark (HUB), a systematic evaluation of the concept unlearning effects of text-to-image diffusion models, covering six dimensions: authenticity, alignment, directionality, directionality accuracy, multilingual robustness, attack robustness, and efficiency.
HoliTracer: Holistic Vectorization of Geographic Objects from Large-Size Remote Sensing Imagery
Yu Wang (Wuhan University), Yansheng Li (Wuhan University)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes the HoliTracer framework, which directly extracts vector maps of multiple types of geographic objects from large-scale remote sensing images.
HOMO-Feature: Cross-Arbitrary-Modal Image Matching with Homomorphism of Organized Major Orientation
Chenzhong Gao (Beijing Institute of Technology), Desheng Weng (Beijing Institute of Technology)
RecognitionRetrievalImageMultimodality
🎯 What it does: A fully manual feature extraction and matching framework called HOMO is proposed, which can achieve cross-modal feature matching between images of any modality.
HORT: Monocular Hand-held Objects Reconstruction with Transformers
Zerui Chen (Inria), Cordelia Schmid (Inria)
Object DetectionPose EstimationDepth EstimationComputational EfficiencyTransformerImagePoint Cloud
🎯 What it does: A Transformer-based coarse-fine hierarchical framework called HORT is proposed, which efficiently generates dense 3D point clouds of handheld objects from a single RGB image.
HouseCrafter: Lifting Floorplans to 3D Scenes with 2D Diffusion Models
Yiwen Chen (Northeastern University), Huaizu Jiang (Northeastern University)
GenerationData SynthesisDiffusion modelMesh
🎯 What it does: This paper proposes a method called HouseCrafter, which can generate complete 3D indoor scenes from 2D floor plans using a diffusion model and outputs them in a mesh format.
HouseTour: A Virtual Real Estate A(I)gent
Ata Çelen (ETH Zurich), Iro Armeni (Stanford University)
GenerationData SynthesisVision Language ModelDiffusion modelGaussian SplattingImageVideoText
🎯 What it does: Proposes the HouseTour method, which generates smooth 3D trajectories and natural language descriptions in real estate style based on sparsely captured property images and camera poses, and can synthesize virtual house tour videos.
How Can Objects Help Video-Language Understanding?
Zitian Tang (Brown University), Chen Sun (Brown University)
Object DetectionObject TrackingTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoMultimodality
🎯 What it does: The ObjectMLLM framework is proposed, which enhances video question-answering performance by injecting detected object bounding boxes and location information into multimodal large language models in text or embedding form.
How Do Multimodal Large Language Models Handle Complex Multimodal Reasoning? Placing Them in An Extensible Escape Game
Ziyue Wang (Tsinghua University), Yang Liu (Tsinghua University)
TransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelTextMultimodalityBenchmark
🎯 What it does: This paper proposes the MM-Escape benchmark and the EscapeCraft environment for evaluating the complete reasoning process of multimodal large language models (MLLMs) in escape room tasks.
How Do Optical Flow and Textual Prompts Collaborate to Assist in Audio-Visual Semantic Segmentation?
Yujian Lee (Hong Kong Baptist University), Wentao Fan (Hong Kong Baptist University)
SegmentationPrompt EngineeringOptical FlowVideoTextMultimodalityAudio
🎯 What it does: This paper studies the audio-visual semantic segmentation (AVSS) task and proposes a framework named Stepping Stone Plus (SSP), which utilizes optical flow, text prompts, and visual-text alignment techniques to improve segmentation accuracy.
How Far are AI-generated Videos from Simulating the 3D Visual World: A Learned 3D Evaluation Approach
Chirui Chang (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
GenerationData SynthesisExplainability and InterpretabilityConvolutional Neural NetworkContrastive LearningOptical FlowVideo
🎯 What it does: This paper proposes a 3D convolutional network L3DE trained on monocular 3D cues (depth, optical flow, appearance) to quantify the 3D visual consistency of AI-generated videos and provides distortion areas through gradient visualization.
How To Make Your Cell Tracker Say "I dunno!"
Richard D. Paul (Forschungszentrum Julich), Hanno Scharr (Forschungszentrum Julich)
Object TrackingTransformerTime Series
🎯 What it does: This paper proposes and implements a general framework that adds uncertainty estimation to cell tracking algorithms based on linear assignment using Bayesian inference and a multi-class perspective, enabling the tracker to provide confidence in its predictions.
How Would It Sound? Material-Controlled Multimodal Acoustic Profile Generation for Indoor Scenes
Mahnoor Fatima Saad (University of Utah), Ziad Al-Halah (University of Utah)
GenerationData SynthesisAuto EncoderMultimodalityAudio
🎯 What it does: This paper proposes a task to generate the target room impulse response (RIR) in indoor scenes by providing arbitrary material configurations, and designs a corresponding generation method.
HPSv3: Towards Wide-Spectrum Human Preference Score
Yuhang Ma (Mizzen AI), Hongsheng Li (Chinese University of Hong Kong)
GenerationRecommendation SystemTransformerReinforcement LearningVision Language ModelImageBenchmarkChain-of-Thought
🎯 What it does: A broad-spectrum human preference dataset HPDv3 was constructed, and a human preference scoring model HPSv3 was trained to evaluate text-to-image generation models; at the same time, an iterative generation method based on HPSv3, called Chain-of-Human-Preference (CoHP), was proposed to enhance image quality.
HQ-CLIP: Leveraging Large Vision-Language Models to Create High-Quality Image-Text Datasets and CLIP Models
Zhixiang Wei (University of Science and Technology of China), Fengyun Rao (Tencent Inc)
Data SynthesisRetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Utilize large-scale visual language models to generate high-quality multi-granularity image-text pairs, constructing a 150M image-text dataset and training an improved CLIP model.
HRScene: How Far Are VLMs from Effective High-Resolution Image Understanding?
Yusen Zhang (Penn State University), Rui Zhang (Penn State University)
RecognitionSegmentationTransformerVision Language ModelImageMultimodalityBenchmark
🎯 What it does: A unified high-resolution image understanding benchmark HRScene has been constructed.
HUG: Hierarchical Urban Gaussian Splatting with Block-Based Reconstruction for Large-Scale Aerial Scenes
Mai Su (Peking University), Guoping Wang (Peking University)
RestorationData SynthesisGaussian SplattingPoint Cloud
🎯 What it does: A hierarchical urban Gaussian projection method named HUG has been developed for high-quality reconstruction and real-time rendering of large-scale aerial city scenes, utilizing visibility-based rapid chunking, hierarchical Gaussian representation, and multi-level image supervision.
Human-in-the-Loop Local Corrections of 3D Scene Layouts via Infilling
Christopher Xie (Meta Reality Labs), Jakob Engel (Meta Reality Labs)
OptimizationTransformerPoint Cloud
🎯 What it does: This paper proposes a human-computer interaction-based method for estimating 3D scene layouts, utilizing human clicks in the perspective to identify local errors and employing a 'filling' mechanism to allow the model to automatically correct these errors, thus achieving iterative, one-click layout optimization.
Human-Object Interaction from Human-Level Instructions
Zhen Wu (Stanford University), C. Karen Liu (Stanford University)
Robotic IntelligenceLarge Language ModelReinforcement LearningDiffusion modelText
🎯 What it does: Utilizing large language models to plan scenes and execute plans, and then generating synchronized full-body, finger, and object movements through a multi-stage diffusion model, while using reinforcement learning tracking to achieve physically feasible long-term human-machine interaction.
HumanOLAT: A Large-Scale Dataset for Full-Body Human Relighting and Novel-View Synthesis
Timo Teufel (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)
GenerationData SynthesisImageVideoBenchmark
🎯 What it does: This paper establishes the HumanOLAT, a large-scale full-body OLAT (One-Light-at-a-Time) dataset, and evaluates various baseline relighting and novel view synthesis methods on this dataset.
Humans as a Calibration Pattern: Dynamic 3D Scene Reconstruction from Unsynchronized and Uncalibrated Videos
Changwoon Choi (Seoul National University), Young Min Kim (Seoul National University)
GenerationPose EstimationNeural Radiance FieldVideo
🎯 What it does: Using human motion as a calibration pattern, a dynamic 3D scene was reconstructed from unsynchronized and uncalibrated multi-view videos.
Humans as Checkerboards: Calibrating Camera Motion Scale for World-Coordinate Human Mesh Recovery
Fengyuan Yang (National University of Singapore), Angela Yao (National University of Singapore)
Pose EstimationDepth EstimationSimultaneous Localization and MappingPoint CloudMesh
🎯 What it does: A non-optimization scale calibration framework HAC is proposed, which uses the human body as a 'checkerboard' to calibrate the scale of monocular SLAM by utilizing the absolute depth predicted by HMR, thereby achieving global human mesh recovery.
HumanSAM: Classifying Human-centric Forgery Videos in Human Spatial, Appearance, and Motion Anomaly
Chang Liu (National University of Defense Technology), Zhiping Cai (National University of Defense Technology)
ClassificationAnomaly DetectionConvolutional Neural NetworkTransformerSupervised Fine-TuningVideo
🎯 What it does: Designed the HumanSAM framework for fine-grained human video forgery detection, categorizing forged videos into three types: spatial, appearance, and motion.
HumorDB: Can AI understand graphical humor?
Vedaant V Jain (University of Illinois Urbana-Champaign), Felipe dos Santos Alves Feitosa (University of São Paulo)
ClassificationRecognitionData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodality
🎯 What it does: A multimodal visual humor dataset, HumorDB, containing pairs of control images was constructed and evaluated, utilizing human experiments and various visual/visual-language models for binary classification, ranking scoring, and comparative judgment tasks.
HUMOTO: A 4D Dataset of Mocap Human Object Interactions
Jiaxin Lu (University of Texas at Austin), Yi Zhou (Adobe Research)
Pose EstimationRobotic IntelligenceLarge Language ModelVideo
🎯 What it does: This paper constructs a high-quality 4D human-object interaction dataset called HUMOTO, which includes 735 segments of multi-object interaction animations, covering 63 finely modeled objects and recording complete body and hand movements.
HUST: High-Fidelity Unbiased Skin Tone Estimation via Texture Quantization
Zimin Ran (University of Technology Sydney), Jiankang Deng (Imperial College London)
GenerationDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: By constructing a high-resolution texture codebook and utilizing dual discriminators for UV texture recovery, combined with a cross-attention module and group identity loss, high-fidelity unbiased skin color estimation from a single or multiple images is achieved.
HVPUNet: Hybrid-Voxel Point-cloud Upsampling Network
Juhyung Ha (Indiana University), David J. Crandall (Indiana University)
GenerationSuper ResolutionConvolutional Neural NetworkPoint Cloud
🎯 What it does: An end-to-end point cloud upsampling network called HVPUNet is proposed, which utilizes Hybrid Voxels to simultaneously encode discrete occupancy information and continuous offset vectors, completing shape completion and super-resolution in two stages of point cloud refinement.
Hybrid Layout Control for Diffusion Transformer: Fewer Annotations, Superior Aesthetics
Keming Wu (Tsinghua University), Yuhui Yuan (Canva)
GenerationData SynthesisTransformerSupervised Fine-TuningDiffusion modelImage
🎯 What it does: A three-stage hybrid layout control framework is proposed, which significantly reduces the reliance on area prompts through three steps: anonymous layout, semantic layout, and quality optimization, achieving high-quality layout-to-image generation.
Hybrid-grained Feature Aggregation with Coarse-to-fine Language Guidance for Self-supervised Monocular Depth Estimation
Wenyao Zhang (Shanghai Jiao Tong University), Xin Jin
Depth EstimationAutonomous DrivingContrastive LearningImage
🎯 What it does: A self-supervised monocular depth estimation framework named Hybrid-depth is proposed, utilizing multi-granularity features from CLIP and DINO, combined with language guidance to achieve depth inference from coarse to fine.
Hybrid-Tower: Fine-grained Pseudo-query Interaction and Generation for Text-to-Video Retrieval
Bangxiang Lan (Renmin University of China), Xirong Li (Renmin University of China)
RetrievalTransformerContrastive LearningVideo
🎯 What it does: This paper proposes the Hybrid-Tower framework and designs the Fine-grained Pseudo-query Interaction and Generation (PIG) method, which utilizes pseudo-queries to achieve fine-grained interaction before video retrieval, enhancing retrieval performance.
Hybrid-TTA: Continual Test-time Adaptation via Dynamic Domain Shift Detection
Hyewon Park (Ewha Womans University), Dongbo Min (Ewha Womans University)
SegmentationDomain AdaptationTransformerSupervised Fine-TuningImage
🎯 What it does: A hybrid tuning framework called Hybrid-TTA is designed to adaptively switch between full parameter fine-tuning and adapter fine-tuning during continuous testing moments, integrating a masked image reconstruction auxiliary task.
Hydra-NeXt: Robust Closed-Loop Driving with Open-Loop Training
Zhenxin Li (Fudan University), Jose M. Alvarez (NVIDIA)
Autonomous DrivingKnowledge DistillationTransformerReinforcement LearningDiffusion modelSequential
🎯 What it does: This paper proposes Hydra-NeXt, a multi-branch unified planning framework that integrates trajectory prediction, control prediction, and trajectory refinement into a single model to achieve closed-loop autonomous driving.
HypDAE: Hyperbolic Diffusion Autoencoders for Hierarchical Few-shot Image Generation
Lingxiao Li (Chinese University of Hong Kong), Xiangyu Yue (Chinese University of Hong Kong)
GenerationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: A Hyperbolic Diffusion Autoencoders (HypDAE) model is proposed, which combines hyperbolic space with diffusion models to generate high-quality, diverse, and category-consistent images from a very small number of samples.
Hyper-Depth: Hypergraph-based Multi-Scale Representation Fusion for Monocular Depth Estimation
Lin Bie (Tsinghua University), Yue Gao (Tsinghua University)
Depth EstimationAutonomous DrivingGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: A multi-scale representation fusion framework based on hypergraphs, called Hyper-Depth, is proposed for monocular depth estimation.
HyperGCT: A Dynamic Hyper-GNN-Learned Geometric Constraint for 3D Registration
Xiyu Zhang (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
Object DetectionGraph Neural NetworkSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A high-order geometric constraint method based on dynamic hypergraph neural networks, HyperGCT, is proposed for 3D point cloud registration.
Hypergraph Clustering Network with Partial Attribute Imputation
Qianqian Wang (Xidian University), Quanxue Gao (Northwest A&F University)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: In the task of hypergraph clustering with missing node attributes, the HCN-PAI model is proposed.
HyPiDecoder: Hybrid Pixel Decoder for Efficient Segmentation and Detection
Fengzhe Zhou (Georgia Institute of Technology), Humphrey Shi (Georgia Institute of Technology)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes HyPiDecoder, a hybrid pixel decoder that integrates a convolutional FPN structure with multi-scale linear attention to enhance the inference speed and accuracy of semantic, instance, panoptic segmentation, and object detection models.
HyTIP: Hybrid Temporal Information Propagation for Masked Conditional Residual Video Coding
Yi-Hsin Chen (National Yang Ming Chiao Tung University), Wen-Hsiao Peng (National Yang Ming Chiao Tung University)
CompressionRecurrent Neural NetworkOptical FlowVideo
🎯 What it does: The HyTIP framework is proposed, which mixes decoded frames with a small amount of implicit features in a buffer for mask-conditioned residual video coding, improving the rate-distortion performance.
I Am Big, You Are Little; I Am Right, You Are Wrong
David A. Kelly (Kings College London), Nathan Blake (Kings College London)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper conducts a study on the minimum sufficient pixel set (MPS) of 15 different architectures of image classification models, comparing their size, location, and the differences in MPS size during misclassification on ImageNet.
I2-World: Intra-Inter Tokenization for Efficient Dynamic 4D Scene Forecasting
Zhimin Liao (Xi'an Jiaotong University), Ziyang Ren (Xi'an Jiaotong University)
GenerationCompressionAutonomous DrivingComputational EfficiencyTransformerAuto EncoderPoint Cloud
🎯 What it does: Proposes the I2-World framework for 4D vehicle occupancy prediction, dividing the scene into intra-scene and inter-scene tokenizers, and using an encoder-decoder architecture to generate future 3D scenes.
I2V3D: Controllable Image-to-video Generation with 3D Guidance
Zhiyuan Zhang (City University of Hong Kong), Jing Liao (City University of Hong Kong)
GenerationData SynthesisDiffusion modelImageVideoMesh
🎯 What it does: A framework for generating controllable videos from a single image is proposed: first, a rough 3D scene is obtained using single-view reconstruction, then animation trajectories are generated and rendered into keyframes using traditional CG animation tools. Finally, a diffusion model is used in two stages (3D-guided keyframe generation + 3D-guided video interpolation) to produce high-quality, temporally coherent videos.
I2VControl: Disentangled and Unified Video Motion Synthesis Control
Wanquan Feng, Qian He
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: The I2VControl framework is proposed, achieving unified and decoupled control of camera movement, object dragging, and motion brush actions within the same model.
IAP: Invisible Adversarial Patch Attack through Perceptibility-Aware Localization and Perturbation Optimization
Subrat Kishore Dutta (CISPA Helmholtz Center for Information Security), Xiao Zhang (CISPA Helmholtz Center for Information Security)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A framework for invisible adversarial patch attacks based on perceptual adaptive positioning and perturbation optimization (IAP) is proposed.
ICE-Bench: A Unified and Comprehensive Benchmark for Image Creating and Editing
Yulin Pan (Alibaba Group), Yu Liu (Alibaba Group)
GenerationTransformerVision Language ModelDiffusion modelImageBenchmark
🎯 What it does: A unified evaluation framework called ICE-Bench is proposed for a systematic assessment of 31 fine-grained tasks related to image generation and editing.
IDEATOR: Jailbreaking and Benchmarking Large Vision-Language Models Using Themselves
Ruofan Wang (Fudan University), Yu-Gang Jiang (Fudan University)
GenerationAdversarial AttackTransformerVision Language ModelDiffusion modelImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper presents IDEATOR, a framework that utilizes Visual Language Models (VLM) and diffusion models to automatically generate image-text pairs for black-box jailbreak, and based on this framework, constructs the VLJailbreakBench evaluation benchmark.
Identity Preserving 3D Head Stylization with Multiview Score Distillation
Bahri Batuhan Bilecen (Bilkent University), Aysegul Dundar (Bilkent University)
GenerationData SynthesisKnowledge DistillationGenerative Adversarial NetworkImage
🎯 What it does: Using a PanoHead-based 3D GAN, identity preservation and artistic effects are unified through multi-view score distillation on real facial images, proposing techniques such as negative log-likelihood distillation, mirror gradients, grid score denoising, and score rank weighting.
Identity-aware Language Gaussian Splatting for Open-vocabulary 3D Semantic Segmentation
SungMin Jang (Konkuk University), Wonjun Kim (Konkuk University)
SegmentationGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes an Identity-Aware Language Gaussian Rendering (ILGS) method, which achieves high-precision predictions for open vocabulary 3D semantic segmentation by embedding language and identity embeddings into 3D Gaussian primitives, and introduces a progressive mask expansion strategy to refine boundaries.
IDF: Iterative Dynamic Filtering Networks for Generalizable Image Denoising
Dongjin Kim (Hanyang University), Tae Hyun Kim (Hanyang University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: Utilizing dynamic convolution kernels for prediction and iterative refinement to achieve efficient denoising of unknown noise.
IDFace: Face Template Protection for Efficient and Secure Identification
Sunpill Kim (Hanyang University), Jae Hong Seo (Hanyang University)
RecognitionRetrievalSafty and PrivacyComputational EfficiencyImage
🎯 What it does: A facial template protection method named IDFace based on homomorphic encryption (HE) is proposed, which enables efficient matching of encrypted templates in large-scale recognition scenarios.
IFAdapter: Instance Feature Control for Grounded Text-to-Image Generation
Yinwei Wu (Tencent), Xinchao Wang (National University of Singapore)
GenerationData SynthesisVision Language ModelDiffusion modelImageText
🎯 What it does: Designed and implemented the IFAdapter plugin to enhance the control of instance positioning and fine-grained features in the layout-to-image generation of pre-trained diffusion models;
IGD: Instructional Graphic Design with Multimodal Layer Generation
Yadong Qu (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
GenerationData SynthesisLarge Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Based on natural language instructions, the Instructional Graphic Designer (IGD) has been designed and implemented to generate editable multimodal (text, images, graphics) layers in one go, outputting complete design files.
IGL-Nav: Incremental 3D Gaussian Localization for Image-goal Navigation
Wenxuan Guo (Tsinghua University), Jiwen Lu (Tsinghua University)
Pose EstimationRobotic IntelligenceConvolutional Neural NetworkGaussian SplattingSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes an image target navigation method IGL-Nav based on Incremental 3D Gaussian Splatting (3DGS), which can construct scene models in real-time during exploration and achieve coarse and fine target localization in two steps.
ILLUME: Illuminating Your LLMs to See, Draw, and Self-Enhance
Chunwei Wang (Huawei Noah's Ark Lab), Hang Xu (Huawei Noah's Ark Lab)
RecognitionGenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: ILLUME has been developed, a unified multimodal large language model capable of simultaneous image understanding and image generation, supporting various visual tasks such as documents and tables.
IM-LUT: Interpolation Mixing Look-Up Tables for Image Super-Resolution
Sejin Park (Korea University), Seung-Won Jung (Korea University)
RestorationSuper ResolutionImage
🎯 What it does: This paper proposes an IM-LUT framework that can achieve super-resolution of images at any scale by learning the mixed weights of various interpolation methods for adaptive resampling of local textures.
Im2Haircut: Single-view Strand-based Hair Reconstruction for Human Avatars
Vanessa Sklyarova (Max Planck Institute for Intelligent Systems), Justus Thies (Max Planck Institute for Intelligent Systems)
GenerationOptimizationTransformerGaussian SplattingImage
🎯 What it does: Generate high-quality, line-based 3D hair geometry from a single photo using a prior model and local optimization;
IM360: Large-scale Indoor Mapping with 360 Cameras
Dongki Jung (University of Maryland), Dinesh Manocha (University of Maryland)
Depth EstimationOptimizationSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes a panoramic structure-from-motion reconstruction and rendering pipeline (IM360) based on panoramic cameras, utilizing a spherical camera model and dense matching to achieve camera localization, geometric reconstruction, and texture optimization in sparsely scanned indoor environments.
Image as an IMU: Estimating Camera Motion from a Single Motion-Blurred Image
Jerred Chen (University of Oxford), Ronald Clark (University of Oxford)
Pose EstimationDepth EstimationOptimizationConvolutional Neural NetworkSimultaneous Localization and MappingOptical FlowImage
🎯 What it does: By utilizing the blur information in a single frame motion blurred image to predict pixel motion flow and depth map, and then solving a linear least squares problem to obtain the instantaneous camera velocity, achieving 'IMU-like' measurement of camera speed from a single frame.
Image Intrinsic Scale Assessment: Bridging the Gap Between Quality and Resolution
Vlad Hosu (Sony AI), Dietmar Saupe (University of Konstanz)
Image TranslationContrastive LearningImage
🎯 What it does: This paper introduces the concept of Image Intrinsic Scale (IIS) and studies the perceived quality of images at different scales as a new task—Image Intrinsic Scale Assessment (IISA). It also constructs the first IISA dataset, IISA-DB, and proposes a weak label generation method, WIISA, to enhance the predictive performance of IQA models.
Image-Guided Shape-from-Template Using Mesh Inextensibility Constraints
Thuy Tran (CNRS), Shaifali Parashar (CNRS)
Depth EstimationOptimizationNeural Radiance FieldImageMesh
🎯 What it does: A template-based unsupervised 3D reconstruction method is proposed, which utilizes image color, gradient, and contour information combined with mesh non-stretch constraints to achieve object shape recovery.
ImageGem: In-the-wild Generative Image Interaction Dataset for Generative Model Personalization
Yuanhe Guo (New York University), Hongyi Wen (Stanford University)
GenerationRetrievalRecommendation SystemVision Language ModelDiffusion modelImageText
🎯 What it does: Proposed the ImageGem dataset and used it for personalized text-to-image generation, retrieval, and LoRA editing.
ImageGen-CoT: Enhancing Text-to-Image In-context Learning with Chain-of-Thought Reasoning
Jiaqi Liao (Microsoft), Lijuan Wang (Microsoft)
GenerationTransformerLarge Language ModelSupervised Fine-TuningImageTextMultimodalityChain-of-Thought
🎯 What it does: This paper proposes the incorporation of chain-of-thought reasoning before image generation (ImageGen-CoT) in unified multimodal large language models (MLLMs) and constructs an automated ImageGen-CoT dataset for fine-tuning to enhance text-to-image contextual learning (T2I-ICL) capabilities.
Images as Noisy Labels: Unleashing the Potential of the Diffusion Model for Open-Vocabulary Semantic Segmentation
Fan Li (Northwestern Polytechnical University), Yuelei Xu (Northwestern Polytechnical University)
SegmentationTransformerDiffusion modelImageAgriculture Related
🎯 What it does: A denoising learning framework based on diffusion models, called DEDOS, is proposed to treat images as noisy labels. During the denoising process, it gradually constructs the scene skeleton and filters texture noise using CLIP features, ultimately achieving open vocabulary semantic segmentation.
iManip: Skill-Incremental Learning for Robotic Manipulation
Zexin Zheng (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
Knowledge DistillationRobotic IntelligenceTransformerReinforcement LearningMultimodalityTime Series
🎯 What it does: This paper studies the incremental learning problem of robots gradually acquiring new manipulation skills without retraining.
Imbalance in Balance: Online Concept Balancing in Generation Models
Yukai Shi, Kun Gai
GenerationData SynthesisDiffusion modelAuto EncoderImageTextBenchmark
🎯 What it does: An online concept balancing training strategy called IMBA loss is proposed to address the concept imbalance problem in text-to-image generation models, and a new benchmark called Inert-CompBench is constructed.
IMG: Calibrating Diffusion Models via Implicit Multimodal Guidance
Jiayi Guo (Georgia Tech), Humphrey Shi (Georgia Tech)
GenerationData SynthesisLarge Language ModelDiffusion modelImageMultimodality
🎯 What it does: The Implicit Multimodal Guidance (IMG) framework is proposed to automatically identify and correct the misalignment between prompts and images during image generation with diffusion models, thereby enhancing the quality of multimodal alignment.
ImHead: A Large-scale Implicit Morphable Model for Localized Head Modeling
Rolandos Alexandros Potamias (Imperial College London), Stefanos Zafeiriou (Imperial College London)
GenerationData SynthesisMesh
🎯 What it does: This paper presents imHead, a large-scale implicit 3D shape model for generating and editing full-head 3D facial avatars.
IMoRe: Implicit Program-Guided Reasoning for Human Motion Q&A
Chen Li (Institute of High Performance Computing), Basura Fernando (Nanyang Technological University)
Pose EstimationTransformerVision Language ModelTextSequential
🎯 What it does: This paper proposes an implicit program-guided motion reasoning framework IMoRe, which utilizes structured program functions to uniformly handle multiple types of queries, dynamically selects multi-layer Vision Transformer motion features for reasoning, and ultimately achieves multi-step fine-grained question answering for motion sequences.
Implicit Counterfactual Learning for Audio-Visual Segmentation
Mingfeng Zha (University of Electronic Science and Technology of China), Heng Tao Shen
SegmentationDiffusion modelContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper proposes an Implicit Adversarial Framework (ICF) that utilizes multi-granularity implicit text and a latent diffusion model to generate orthogonal adversarial pseudo-samples, alleviating the issues of modality imbalance and mismatch in audio-visual segmentation.