ICCV 2025 Papers — Page 26
IEEE/CVF International Conference on Computer Vision · 2701 papers
UniPhys: Unified Planner and Controller with Diffusion for Flexible Physics-Based Character Control
Yan Wu (ETH Zurich), Siyu Tang (ETH Zurich)
Diffusion modelMultimodalityPhysics Related
🎯 What it does: This paper proposes UniPhys, a physics-based motion control framework that unifies planning and control into a single diffusion model.
UniPortrait: A Unified Framework for Identity-Preserving Single- and Multi-Human Image Personalization
Junjie He (Alibaba Group), Liefeng Bo (Alibaba Group)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: A unified single/multiple identity portrait personalization framework called UniPortrait is proposed, which can generate high-quality portrait images while maintaining high facial identity fidelity, editability, free text descriptions, and layout-free constraints.
UniRes: Universal Image Restoration for Complex Degradations
Mo Zhou (Google), Hossein Talebi (Google)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: An end-to-end diffusion model framework called UniRes is proposed to restore images containing various real-world mixed degradations.
UNIS: A Unified Framework for Achieving Unbiased Neural Implicit Surfaces in Volume Rendering
Junkai Deng (Nanyang Technological University), Ying He (Nanyang Technological University)
Neural Radiance FieldPoint CloudMeshBenchmark
🎯 What it does: Proposes the UNIS framework, unifying neural implicit surface methods, providing a family of density mapping functions that satisfy first-order unbiasedness, and validating new Algebraic and Softplus kernel functions.
UniversalBooth: Model-Agnostic Personalized Text-to-Image Generation
Songhua Liu (National University of Singapore), Xinchao Wang (National University of Singapore)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: We propose a model-agnostic personalized text-to-image generation method called UniversalBooth, which can be directly transferred to diffusion models of different architectures after training, without the need for retraining.
UniVerse: Unleashing the Scene Prior of Video Diffusion Models for Robust Radiance Field Reconstruction
Jin Cao (Zhejiang University), Sida Peng (Zhejiang University)
RestorationGenerationTransformerDiffusion modelNeural Radiance FieldImageVideo
🎯 What it does: For inconsistent multi-view images, we first use a Video Diffusion Model (VDM) to restore the images to a consistent state, and then utilize methods such as NeRF/3DGS for 3D reconstruction.
UniVG: A Generalist Diffusion Model for Unified Image Generation and Editing
Tsu-Jui Fu (Apple), Yinfei Yang (Meta)
SegmentationGenerationDepth EstimationDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: This paper presents UniVG, a unified diffusion model that supports multi-tasking including text-to-image, inpainting, instruction editing, layout generation, identity-preserving generation, depth estimation, and referential segmentation.
Unknown Text Learning for CLIP-based Few-Shot Open-set Recognition
Rui Ma (Tianjin University), Yahong Han (Tianjin University)
RecognitionTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: A method for Unknown Text Learning (UTL) is proposed, utilizing the CLIP model to simultaneously learn unknown text and contextual prompts in few-shot open-set recognition tasks.
Unlearning the Noisy Correspondence Makes CLIP More Robust
Haochen Han (Peng Cheng Laboratory), Fangming Liu (Peng Cheng Laboratory)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: Fine-tuning the pre-trained CLIP model using 'hardest negative samples' to achieve 'unlearning' of noise pairs (FP and FN), thereby enhancing the model's robustness to noise correspondence.
Unleashing High-Quality Image Generation in Diffusion Sampling Using Second-Order Levenberg-Marquardt-Langevin
Fangyikang Wang (Zhejiang University), Chen Li (WeChat Vision, Tencent Inc)
GenerationData SynthesisOptimizationDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A training-independent Levenberg-Marquardt-Langevin (LML) method is proposed, which approximates the Hessian of the diffusion model through low-rank approximation and damping techniques, and uses it for Langevin updates to improve sampling quality.
Unleashing the Temporal Potential of Stereo Event Cameras for Continuous-Time 3D Object Detection
Jae-Young Kang (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A continuous-time 3D object detection framework is proposed that uses only event stereo cameras, achieving fully asynchronous detection without LiDAR or RGB.
Unleashing Vecset Diffusion Model for Fast Shape Generation
Zeqiang Lai (Chinese University of Hong Kong), Xiangyu Yue (Chinese University of Hong Kong)
GenerationData SynthesisDiffusion modelAuto EncoderGenerative Adversarial NetworkPoint CloudMesh
🎯 What it does: Proposes the FlashVDM framework, which accelerates the pre-trained Vecset Diffusion Model (VDM) to 5 steps through Progressive Flow Distillation and an efficient VAE decoder, achieving high-quality 3D shape generation within 1 second;
Unlocking Constraints: Source-Free Occlusion-Aware Seamless Segmentation
Yihong Cao (Hunan University), Hui Zhang (Hunan University)
SegmentationDomain AdaptationAutonomous DrivingImage
🎯 What it does: A source-agnostic panoramic occlusion-free seamless segmentation task is proposed, and the UNLOCK framework is designed to achieve model adaptation in the target panoramic domain.
Unlocking the Potential of Diffusion Priors in Blind Face Restoration
Yunqi Miao (University of Warwick), Jiankang Deng (Imperial College London)
RestorationGenerationDiffusion modelAuto EncoderImageMultimodality
🎯 What it does: This paper presents FLIPNET, a unified model that can switch between recovery mode and degradation mode, utilizing diffusion models to achieve lossless face restoration and capable of synthesizing degraded images that conform to real-world scenarios.
UnMix-NeRF: Spectral Unmixing Meets Neural Radiance Fields
Fabian Perez (Universidad Industrial de Santander), Bernard Ghanem (KAUST)
SegmentationGenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: Combining spectral unmixing with NeRF to achieve spectral view synthesis and unsupervised material segmentation.
Unraveling the Effects of Synthetic Data on End-to-End Autonomous Driving
Junhao Ge (Shanghai Jiao Tong University), Siheng Chen (New York University)
Data SynthesisAutonomous DrivingGaussian SplattingPoint Cloud
🎯 What it does: Designed and implemented SceneCrafter, an end-to-end autonomous driving simulator based on 3D Gaussian Splatting, for high-fidelity synthetic data generation and closed-loop evaluation.
Unraveling the Smoothness Properties of Diffusion Models: A Gaussian Mixture Perspective
Yingyu Liang (University of Hong Kong), Yufa Zhou (University of Pennsylvania)
GenerationData SynthesisOptimizationDiffusion modelMultimodalityStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper provides a theoretical analysis of the Lipschitz continuity and second-order momentum properties of diffusion models under Gaussian mixture data distributions, and presents an upper bound for k uncorrelated components;
UnrealZoo: Enriching Photo-realistic Virtual Worlds for Embodied AI
Fangwei Zhong (Beijing Normal University), Yizhou Wang (Peking University)
Robotic IntelligenceLarge Language ModelReinforcement LearningVision Language ModelImageMultimodality
🎯 What it does: The UnrealZoo platform has been constructed, providing 100 high-fidelity virtual worlds based on Unreal Engine and 67 types of playable entities. It achieves efficient data collection and distributed training for evaluating embodied AI tasks such as visual navigation and tracking through the optimization of the UnrealCV+ and Gym interface.
Unsupervised Histopathological Image Semantic Segmentation with Overlapping Patches Consistency Constraint
Wentian Cai (South China University of Technology), Ying Gao (South China University of Technology)
SegmentationTransformerImage
🎯 What it does: An unsupervised semantic segmentation framework for pathological images is proposed, utilizing Overlapping Patch Consistency Constraint (OPCC) and Inter-layer Self-Attention Fusion (ILSAF) to enhance feature representation and clustering effectiveness.
Unsupervised Identification of Protein Compositions and Conformations via Implicit Content-Transformation Disentanglement
Mostofa Rafid Uddin (Carnegie Mellon University), Min Xu (Carnegie Mellon University)
RecognitionRepresentation LearningAuto EncoderContrastive LearningImage
🎯 What it does: Decoupling the content (protein composition) and transformation (conformation) of 3D cryo-ET images of protein mixtures in an unsupervised manner to achieve the identification of protein composition and conformation.
Unsupervised Imaging Inverse Problems with Diffusion Distribution Matching
Giacomo Meanti (Univ. Grenoble Alpes), Julien Mairal (Univ. Grenoble Alpes)
RestorationSuper ResolutionDiffusion modelFlow-based ModelImage
🎯 What it does: This paper proposes a method that utilizes a collection of unpaired clean images and corrupted images to learn the forward degradation operator A in image inverse problems through a diffusion model and Conditional Flow Matching (CFM). After obtaining the degradation operator, a non-blind method is used to restore the image. Additionally, applications to non-uniform blur, camera lens distortion, and single-image super-resolution are demonstrated.
Unsupervised Joint Learning of Optical Flow and Intensity with Event Cameras
Shuang Guo (TU Berlin), Guillermo Gallego (TU Berlin)
RestorationConvolutional Neural NetworkOptical FlowImageVideo
🎯 What it does: Proposes an unsupervised learning framework that utilizes a single network to simultaneously estimate the optical flow (motion) and image intensity (appearance) of event cameras.
Unsupervised Part Discovery via Descriptor-Based Masked Image Restoration with Optimized Constraints
Jiahao Xia (University of Technology Sydney), Jian Zhang (Southern University of Science and Technology)
Object DetectionSegmentationTransformerAuto EncoderImage
🎯 What it does: An unsupervised part discovery framework called Mask Part Autoencoder (MPAE) is proposed, which aligns part descriptors using low-level visual features by filling in part descriptors and recovering images on randomly occluded images, resulting in pixel-level segmentation that closely matches the true part shapes.
Unsupervised RGB-D Point Cloud Registration for Scenes with Low Overlap and Photometric Inconsistency
Yejun Shou (Zhejiang University), Yanlong Cao (Zhejiang University)
RecognitionSegmentationOptimizationGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes OG-UPCR, an unsupervised RGB-D point cloud registration framework designed for scenarios with low overlap and inconsistent lighting;
Unsupervised Visible-Infrared Person Re-identification under Unpaired Settings
Haoyu Yao (Wuhan University), Mang Ye (Wuhan University)
RecognitionRetrievalTransformerContrastive LearningImageMultimodality
🎯 What it does: A Mapping and Collaborative Learning (MCL) framework is proposed for training unsupervised visible-infrared person re-identification models under unpaired conditions.
Unsupervised Visual Chain-of-Thought Reasoning via Preference Optimization
Kesen Zhao (Nanyang Technological University), Hanwang Zhang (Nanyang Technological University)
Object DetectionOptimizationTransformerLarge Language ModelContrastive LearningImageMultimodalityChain-of-Thought
🎯 What it does: This paper proposes an unsupervised visual chain reasoning framework called UV-CoT, which learns visual reasoning by comparing model-generated candidate boxes with an evaluator.
Unveiling the Invisible: Reasoning Complex Occlusions Amodally with AURA
Zhixuan Li (Nanyang Technological University), Weisi Lin (Nanyang Technological University)
Object DetectionSegmentationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: The AURA model is proposed to achieve segmentation of visible and invisible objects under user text input, and generate corresponding text answers.
UnZipLoRA: Separating Content and Style from a Single Image
Chang Liu (University of Illinois), Svetlana Lazebnik (University of Illinois)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a technique to decompose a single image into content (subject) LoRA and style LoRA, supporting independent or joint generation of images in different scenes.
UPP: Unified Point-Level Prompting for Robust Point Cloud Analysis
Zixiang Ai (Wangxuan Institute of Computer Technology Peking University), Jiahuan Zhou (Wangxuan Institute of Computer Technology Peking University)
ClassificationSegmentationTransformerPrompt EngineeringPoint Cloud
🎯 What it does: A unified point-level prompting framework (UPP) is proposed, which transforms denoising and completion tasks into prompts for downstream point cloud analysis, achieving robust analysis in the presence of noise and missing data while keeping the backbone network parameters frozen.
UPRE: Zero-Shot Domain Adaptation for Object Detection via Unified Prompt and Representation Enhancement
Xiao Zhang (Dalian University of Technology), Xiangxiang Chu (Alibaba Group)
Object DetectionDomain AdaptationPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes a framework for Unified Prompt and Representation Enhancement (UPRE) to achieve zero-shot domain adaptation for object detection under unlabelled target domain images.
UrbanLLaVA: A Multi-modal Large Language Model for Urban Intelligence
Jie Feng (Tsinghua University), Yong Li (Tsinghua University)
TransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper proposes UrbanLLaVA, which unifies the processing of four types of urban multimodal data (street view images, satellite images, spatial data, and trajectory data) and achieves cross-modal reasoning and decision-making in multi-city tasks.
USP: Unified Self-Supervised Pretraining for Image Generation and Understanding
Xiangxiang Chu (Alibaba Group), Yong Wang (Alibaba Group)
ClassificationSegmentationGenerationTransformerDiffusion modelAuto EncoderImage
🎯 What it does: A unified self-supervised pre-training framework USP is proposed, which first performs masked feature modeling in the VAE latent space, and then directly initializes the diffusion model and visual task network with the pre-trained weights.
UST-SSM: Unified Spatio-Temporal State Space Models for Point Cloud Video Modeling
Peiming Li (Peking University), Mengyuan Liu (Peking University)
RecognitionComputational EfficiencyTransformerVideoPoint Cloud
🎯 What it does: This paper proposes a Unified Spatiotemporal State Space Model (UST-SSM) for spatiotemporal encoding and action recognition of point cloud videos.
V.I.P. : Iterative Online Preference Distillation for Efficient Video Diffusion Models
Jisoo Kim (Yonsei University), Youngjae Yu (Yonsei University)
GenerationOptimizationKnowledge DistillationSupervised Fine-TuningDiffusion modelContrastive LearningVideo
🎯 What it does: A lightweight video diffusion model training framework V.I.P. based on iterative online preference distillation is proposed, and efficient distillation of the pruned model is achieved through the ReDPO loss.
V2M4: 4D Mesh Animation Reconstruction from a Single Monocular Video
Jianqi Chen (King Abdullah University of Science and Technology), Peter Wonka (King Abdullah University of Science and Technology)
GenerationOptimizationGaussian SplattingVideoMeshBenchmark
🎯 What it does: An end-to-end method named V2M4 is proposed, capable of directly generating high-quality, usable 4D mesh animation assets from monocular video captured from a single fixed camera perspective.
V2PE: Improving Multimodal Long-Context Capability of Vision-Language Models with Variable Visual Position Encoding
Junqi Ge (Tsinghua University), Xizhou Zhu (Tsinghua University)
RecognitionGenerationRetrievalTransformerSupervised Fine-TuningVision Language ModelTextMultimodality
🎯 What it does: This study investigates the performance of visual language models under long sequence inputs, constructs a long-context multimodal dataset, proposes and evaluates the Variable Visual Position Encoding (V2PE) method, and fine-tunes it on InternVL2.
V2XPnP: Vehicle-to-Everything Spatio-Temporal Fusion for Multi-Agent Perception and Prediction
Zewei Zhou (University of California), Jiaqi Ma (University of California)
Object DetectionAutonomous DrivingTransformerPoint CloudSequential
🎯 What it does: Proposes the V2XPnP framework, which fuses multi-vehicle and multi-frame LiDAR features through a Transformer for spatiotemporal integration, supporting end-to-end perception and prediction tasks.
V2XScenes: A Multiple Challenging Traffic Conditions Dataset for Large-Range Vehicle-Infrastructure Collaborative Perception
Bowen Wang (Shanghai Jiao Tong University), Chin Long Ng (Shanghai Jiao Tong University)
Object DetectionObject TrackingAutonomous DrivingTransformerSimultaneous Localization and MappingVideoPoint CloudBenchmark
🎯 What it does: This paper constructs the V2XScenes dataset and proposes a 3D detection and tracking benchmark for vehicle-roadside collaborative perception.
VA-MoE: Variables-Adaptive Mixture of Experts for Incremental Weather Forecasting
Hao Chen (Hong Kong University of Science and Technology), Lei Bai (Shanghai AI Laboratory)
TransformerMixture of ExpertsTime Series
🎯 What it does: A Variable Adaptive Mixture of Experts (VA-MoE) framework is proposed for incremental weather forecasting, which supports the dynamic addition of new meteorological variables without retraining the entire model.
VACE: All-in-One Video Creation and Editing
Zeyinzi Jiang (Alibaba Group), Yu Liu (Alibaba Group)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderVideoTextMultimodalityBenchmark
🎯 What it does: VACE is a unified video generation and editing framework that supports various tasks such as text-to-video, reference video generation, video editing, mask editing, and multi-task combinations, all achievable with a single model.
VAFlow: Video-to-Audio Generation with Cross-Modality Flow Matching
Xihua Wang (Renmin University of China), Yunfeng Wang (ZHI-TECH GROUP)
GenerationData SynthesisTransformerDiffusion modelFlow-based ModelAuto EncoderVideoMultimodalityAudio
🎯 What it does: This paper proposes VAFlow, a video-to-audio generation framework directly based on flow matching.
VAGUE: Visual Contexts Clarify Ambiguous Expressions
Heejeong Nam (Brown University), Youngjae Yu (Yonsei University)
RecognitionData-Centric LearningTransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposes the VAGUE benchmark to evaluate the intent reasoning ability of multimodal systems in resolving ambiguous expressions within visual contexts.
VALLR: Visual ASR Language Model for Lip Reading
Marshall Thomas (University of Surrey), Richard Bowden (University of Surrey)
RecognitionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoText
🎯 What it does: A two-stage visual speech recognition framework is proposed: first, using a visual Transformer + CTC to map lip-reading videos to phoneme sequences, and then reconstructing the phoneme sequences into complete sentences using a fine-tuned LLM.
Vamba: Understanding Hour-Long Videos with Hybrid Mamba-Transformers
Weiming Ren (University of Waterloo), Wenhu Chen (University of Waterloo)
TransformerLarge Language ModelSupervised Fine-TuningVideoBenchmark
🎯 What it does: A hybrid Mamba-Transformer architecture (VAMBA) is proposed for understanding videos lasting several hours, avoiding the quadratic complexity of traditional Transformers and large-scale token compression.
Variance-Based Pruning for Accelerating and Compressing Trained Networks
Uranik Berisha (Robert Bosch GmbH), Alexandru Paul Condurache (University of Lubeck)
CompressionComputational EfficiencyTransformerImage
🎯 What it does: This paper proposes a single-shot pruning method based on activation variance (Variance-Based Pruning, VBP). It first calculates the mean and variance of the activations in the MLP hidden layers of the Transformer, then removes the neurons with the smallest variance, compensating their mean into the biases of subsequent linear layers to compress the model without significantly affecting performance.
VCA: Video Curious Agent for Long Video Understanding
Zeyuan Yang (University of Massachusetts), Chuang Gan (University of Massachusetts)
Computational EfficiencyRobotic IntelligenceReinforcement Learning from Human FeedbackTransformerPrompt EngineeringVision Language ModelVideoTextChain-of-Thought
🎯 What it does: Designed and implemented a curiosity-driven video agent (VCA) based on a video language model, which actively explores long videos through a tree search structure and self-generated intrinsic rewards to efficiently answer question-and-answer tasks.
Vector Contrastive Learning For Pixel-Wise Pretraining In Medical Vision
Yuting He (Case Western Reserve University), Shuo Li (Case Western Reserve University)
SegmentationRepresentation LearningContrastive LearningMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes Vector Contrastive Learning and the COVER framework to address the issue of over-dispersion in pixel-level pre-training of medical images.
VEGGIE: Instructional Editing and Reasoning Video Concepts with Grounded Generation
Shoubin Yu (Adobe Research University of Michigan), Mohit Bansal (UNC Chapel Hill)
GenerationData SynthesisLarge Language ModelDiffusion modelImageVideoMultimodalityBenchmark
🎯 What it does: We propose VEGGIE, a unified video editing framework that enables pixel-level video editing based on diverse instructions (such as adding, deleting, changing, style transfer, localization, etc.).
VehicleMAE: View-asymmetry Mutual Learning for Vehicle Re-identification Pre-training via Masked AutoEncoders
Qi Wang (Nanchang University), Ruihua Zhou (Nanchang University)
RecognitionRetrievalKnowledge DistillationConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes VehicleMAE, a self-supervised pre-training framework based on perspective-asymmetric masking and past-present mutual learning, which is pre-trained on the large-scale multi-view vehicle Re-ID dataset DiffVERI and evaluated on the VeRi-776 and VehicleID benchmarks.
Verbalized Representation Learning for Interpretable Few-Shot Generalization
Cheng-Fu Yang (University of California), Kai-Wei Chang (University of California)
ClassificationExplainability and InterpretabilityRepresentation LearningTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: A Verbalized Representation Learning (VRL) method is proposed, which utilizes visual-language models (VLM) to automatically generate interpretable natural language features from a small number of samples and maps them to numerical vectors for downstream classification tasks.
Versatile Transition Generation with Image-to-Video Diffusion
Zuhao Yang (Nanyang Technological University), Song Bai (ByteDance Inc.)
GenerationData SynthesisDiffusion modelImageVideoBenchmark
🎯 What it does: This paper proposes a unified VTG framework that utilizes a pre-trained image-to-video diffusion model to achieve four types of transition video generation: object deformation, motion prediction, concept blending, and scene transition.
VertexRegen: Mesh Generation with Continuous Level of Detail
Xiang Zhang, Henry Howard-Jenkins
GenerationTransformerMesh
🎯 What it does: A vertex-splitting-based generative advanced mesh framework, VertexRegen, is proposed, which can automatically generate 3D meshes at continuous levels of detail.
VFlowOpt: A Token Pruning Framework for LMMs with Visual Information Flow-Guided Optimization
Sihan Yang (Shanghai AI Laboratory), Jiangmiao Pang (Shanghai AI Laboratory)
OptimizationTransformerLarge Language ModelVision Language ModelImageVideoMultimodality
🎯 What it does: To address the issue of visual token redundancy in large multimodal models, the VFlowOpt token pruning framework is proposed.
VGGSounder: Audio-Visual Evaluations for Foundation Models
Daniil Zverev (Technical University of Munich), A. Sophia Koepke (Technical University of Munich)
ClassificationRecognitionLarge Language ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: Introducing VGGSounder - a re-annotation of the VGGSound test set with multi-label, modality labels, and meta-labels, creating a multimodal perception benchmark for evaluating audio-visual foundational models.
VGMamba: Attribute-to-Location Clue Reasoning for Quantity-Agnostic 3D Visual Grounding
Yihang Zhu (Xidian University), Cheng Deng (Xidian University)
Object DetectionContrastive LearningPoint CloudBenchmark
🎯 What it does: A new 3D visual localization network called VGMamba is proposed, which employs an attribute-location clue reasoning mechanism to achieve target localization in 3D scenes through SVD decomposition of features, Mamba for long-range modeling, and bidirectional directive fusion.
ViCTr: Vital Consistency Transfer for Pathology Aware Image Synthesis
Onkar Susladkar (Northwestern University), Ulas Bagci (Stanford University)
GenerationData SynthesisDiffusion modelRectified FlowImageMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes the ViCTr two-stage framework for generating medical images that balance anatomical integrity and pathological details, particularly for the synthesis of abdominal CT/MRI and liver cirrhosis MRI.
Vid-Group: Temporal Video Grounding Pretraining from Unlabeled Videos in the Wild
Peijun Bao (Nanyang Technological University), Alex Kot (Nanyang Technological University)
RecognitionRetrievalTransformerLarge Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: A massive video temporal annotation dataset, Vid-Group, was constructed without the need for manual labeling, and the ReCorrect algorithm was proposed, which significantly improves the performance of Temporal Video Grounding (TVG) during the pre-training phase through self-correction (semantic-guided refinement + memory consistency correction).
Video Color Grading via Look-Up Table Generation
Seunghyun Shin, Joon-Young Lee
Image TranslationImage HarmonizationGenerationLarge Language ModelDiffusion modelVideoText
🎯 What it does: A color grading framework based on reference videos is proposed, which explicitly generates LUTs using diffusion models and adjusts user preferences through text prompts.
Video Individual Counting for Moving Drones
Yaowu Fan (Sun Yat-sen University), Andy J. Ma (Sun Yat-sen University)
Object DetectionObject TrackingConvolutional Neural NetworkTransformerVideo
🎯 What it does: For crowded scenes captured by high-speed moving drones, a Video Individual Counting (VIC) method is proposed, which uses a shared density map-guided deep cross-frame attention network to directly estimate the shared density map and derive the inflow and outflow density maps, ultimately counting the unique individuals in the video.
Video Motion Graphs
Haiyang Liu (University of Tokyo), Yang Zhou (Adobe Research)
GenerationRetrievalDiffusion modelVideoMultimodality
🎯 What it does: Proposes the Video Motion Graphs system, which can retrieve video clips based on multimodal conditions such as reference videos, music, and action labels, and generate coherent human motion videos through HMInterp interpolation.
Video-T1: Test-time Scaling for Video Generation
Fangfu Liu (Tsinghua University), Yueqi Duan (Tsinghua University)
GenerationData SynthesisDiffusion modelVideoBenchmark
🎯 What it does: Proposes the use of Test-Time Scaling (TTS) strategy in video generation, enhancing video quality by utilizing more computational resources during the inference phase;
Video2BEV: Transforming Drone Videos to BEVs for Video-based Geo-localization
Hao Ju (University of Macau), Zhedong Zheng (University of Macau)
RecognitionGenerationRetrievalTransformerDiffusion modelContrastive LearningGaussian SplattingVideo
🎯 What it does: Proposes the Video2BEV scheme, which converts drone videos into bird's-eye views (BEV) to enhance cross-platform geographic positioning accuracy.
VideoAds for Fast-Paced Video Understanding
Zheyuan Zhang (Northwestern University), Boqing Gong (Boston University)
RecognitionData-Centric LearningTransformerLarge Language ModelVision Language ModelVideoMultimodalityChain-of-Thought
🎯 What it does: The VideoAds dataset is proposed, specifically designed to evaluate the multi-step temporal reasoning and summarization capabilities of multimodal large language models (MLLMs) in advertising videos.
VideoAuteur: Towards Long Narrative Video Generation
Junfei Xiao (Johns Hopkins University), Lu Jiang (ByteDance)
GenerationTransformerLarge Language ModelDiffusion modelVideoMultimodality
🎯 What it does: Designed and implemented an end-to-end long-form narrative video generation framework called VideoAuteur, and created a large-scale cooking video dataset named CookGen.
VideoLLaMB: Long Streaming Video Understanding with Recurrent Memory Bridges
Yuxuan Wang (State Key Laboratory of General Artificial Intelligence), Zilong Zheng (State Key Laboratory of General Artificial Intelligence)
SegmentationRetrievalComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideoText
🎯 What it does: This paper proposes the VideoLLaMB framework, which achieves efficient understanding of long video streams using a recursive memory bridge layer and scene chunking algorithm, and can generate video question-answering and planning results in real-time on a single GPU.
VideoMiner: Iteratively Grounding Key Frames of Hour-Long Videos via Tree-based Group Relative Policy Optimization
Xinye Cao (Beijing University of Posts and Telecommunications), Yutong Gao (Minzu University of China)
SegmentationOptimizationReinforcement LearningVision Language ModelVideo
🎯 What it does: This paper presents VideoMiner, which utilizes a tree structure for hierarchical segmentation, description, and clustering of hourly-level videos, allowing for precise keyframe localization layer by layer.
VideoOrion: Tokenizing Object Dynamics in Videos
Yicheng Feng (Peking University), Zongqing Lu (Peking University)
Object DetectionObject TrackingSegmentationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodality
🎯 What it does: This paper presents VideoOrion, a video large language model that dynamically encodes objects in videos as object tokens using a detect-segment-track pipeline, and inputs them along with context tokens into the LLM, enhancing video understanding and video reference tasks.
VideoRFSplat: Direct Scene-Level Text-to-3D Gaussian Splatting Generation with Flexible Pose and Multi-View Joint Modeling
Hyojun Go (EverEx), Changick Kim (Yonsei University)
GenerationData SynthesisPose EstimationTransformerDiffusion modelRectified FlowGaussian SplattingVideoText
🎯 What it does: A dual-stream architecture and asynchronous sampling strategy based on a video generation model is proposed, which directly generates high-quality 3D Gaussian scattering field (3DGS) scenes from text.
VideoSetDiff: Identifying and Reasoning Similarities and Differences in Similar Videos
Yue Qiu (National Institute of Advanced Industrial Science and Technology), Ryusuke Sagawa (National Institute of Advanced Industrial Science and Technology)
RecognitionRetrievalTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVideoTextMultimodalityBenchmark
🎯 What it does: A VideoSetDiff dataset has been constructed (containing 3,783 four-video sets and 76,859 QA pairs) and a baseline called VidSetReasoner has been proposed for identifying and reasoning about similarities and differences among multiple videos.
VideoVAE+: Large Motion Video Autoencoding with Cross-modal Video VAE
Yazhou Xing (Hong Kong University of Science and Technology), Qifeng Chen (Hong Kong University of Science and Technology)
GenerationCompressionAuto EncoderGenerative Adversarial NetworkVideoText
🎯 What it does: This paper proposes VideoVAE+, a variational autoencoder capable of high-quality self-encoding in videos with significant motion, combining spatiotemporal compression and cross-modal text guidance.
ViewSRD: 3D Visual Grounding via Structured Multi-View Decomposition
Ronggang Huang (South China University of Technology), Shengfeng He (South China University of Technology)
RecognitionObject DetectionTransformerLarge Language ModelTextPoint Cloud
🎯 What it does: This work addresses the 3D visual grounding (3DVG) task and proposes the ViewSRD framework, which decomposes complex multi-anchor queries into single-anchor representations and shares viewpoint information across multiple perspectives to achieve more accurate target localization.
VIGFace: Virtual Identity Generation for Privacy-Free Face Recognition Dataset
Minsoo Kim (Korea Institute of Science and Technology), Ig-Jae Kim (Korea Institute of Science and Technology)
RecognitionGenerationData SynthesisPose EstimationDiffusion modelImage
🎯 What it does: VIGFace proposes a method for pre-allocating virtual identities in the feature space and generating privacy-friendly synthetic facial images using diffusion models.
ViLLa: Video Reasoning Segmentation with Large Language Model
Rongkun Zheng (University of Hong Kong), Hengshuang Zhao (University of Hong Kong)
Object TrackingSegmentationTransformerLarge Language ModelVision Language ModelVideoMultimodality
🎯 What it does: A multimodal model called ViLLa based on large language models is designed for instance segmentation and tracking in complex video scenes according to natural language instructions.
ViLU: Learning Vision-Language Uncertainties for Failure Prediction
Marc Lafon (Conservatoire National des Arts et Métiers), Nicolas Thome (Sorbonne Université)
ClassificationAnomaly DetectionTransformerVision Language ModelImageMultimodality
🎯 What it does: Proposes the ViLU framework for failure prediction and uncertainty quantification in visual-language models (VLMs).
ViM-VQ: Efficient Post-Training Vector Quantization for Visual Mamba
Juncan Deng (Zhejiang University), Kejie Huang (Zhejiang University)
ClassificationObject DetectionSegmentationCompressionKnowledge DistillationTransformerImage
🎯 What it does: An efficient post-training vector quantization method, ViM-VQ, is proposed for the visual Mamba network to achieve low-bit compression while maintaining model performance.
VIPerson: Flexibly Generating Virtual Identity for Person Re-Identification
Xiao-Wen Zhang (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
RecognitionGenerationData SynthesisRetrievalDomain AdaptationDiffusion modelImage
🎯 What it does: Designed and implemented the VIPerson virtual identity generation pipeline, capable of synthesizing pedestrian images with camera-realistic styles and scalable cross-camera diversity without manual annotation, significantly enhancing the generalization ability of person re-identification models.
VisHall3D: Monocular Semantic Scene Completion from Reconstructing the Visible Regions to Hallucinating the Invisible Regions
Haoang Lu (Xian Jiaotong University), Le Wang (Xian Jiaotong University)
RestorationSegmentationAutonomous DrivingConvolutional Neural NetworkAuto EncoderPoint Cloud
🎯 What it does: This paper studies a two-stage monocular semantic scene completion framework called VisHall3D, which first recovers visible areas and then infers invisible areas.
Vision-Language Interactive Relation Mining for Open-Vocabulary Scene Graph Generation
Yukuan Min (Xidian University), Cheng Deng (Xidian University)
Object DetectionGenerationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: Proposes the Vision-Language Interactive Relation Mining model (VL-IRM), which achieves open vocabulary scene graph generation through interactive learning between vision and text.
Vision-Language Models Can't See the Obvious
Ngoc Dung Huynh (Technology Innovation Institute), Sanath Narayan (Technology Innovation Institute)
ClassificationRecognitionRetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: A SalBench benchmark is proposed to evaluate the capabilities of large visual language models in low-level perception (color, direction, size, etc.), defining three tasks: Odd-One-Out Detection, Referring Odd-One-Out, and Visual Referring Odd-One-Out.
Vision-Language Neural Graph Featurization for Extracting Retinal Lesions
Taimur Hassan (Abu Dhabi University), Naoufel Werghi (Khalifa University)
ClassificationSegmentationGraph Neural NetworkVision Language ModelImageMultimodality
🎯 What it does: Proposes an unsupervised audiovisual language neural graph feature characterization method that uses graph structures to segment retinal images and identify different lesions.
VISION-XL: High Definition Video Inverse Problem Solver using Latent Image Diffusion Models
Taesung Kwon (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
RestorationSuper ResolutionOptimizationComputational EfficiencyDiffusion modelAuto EncoderVideo
🎯 What it does: A high-resolution video inverse problem solving framework utilizing the latent diffusion model (SDXL) is proposed, supporting various spatiotemporal degradations (deblurring, super-resolution, inpainting, and their combinations with frame averaging) and achieving efficient real-time reconstruction on a single GPU.
VisionMath: Vision-Form Mathematical Problem-Solving
Zongyang Ma (Institute of Automation, Chinese Academy of Sciences), Weiming Hu (Institute of Automation, Chinese Academy of Sciences)
TransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Developed the VisionMath model to solve visually presented mathematical problems.
VisNumBench: Evaluating Number Sense of Multimodal Large Language Models
Tengjin Weng (Shenzhen University), Zhong Ming (Shenzhen University)
TransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: A benchmark called VisNumBench is proposed to evaluate the intuitive numerical sense capabilities of multimodal large models in visual numerical sense.
VISO: Accelerating In-orbit Object Detection with Language-Guided Mask Learning and Sparse Inference
Meiqi Wang (Tsinghua University), Han Qiu (Tsinghua University)
Object DetectionVision Language ModelImageText
🎯 What it does: The VISO model is proposed, which combines language-guided mask learning and sparse inference to achieve high-precision detection of small targets in satellite orbital environments and significantly improve inference speed.
ViSpeak: Visual Instruction Feedback in Streaming Videos
Shenghao Fu (Sun Yat-sen University), Wei-Shi Zheng
Large Language ModelSupervised Fine-TuningVision Language ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: A new task called Visual Instruction Feedback is proposed, which requires the model to actively recognize and respond to visual gestures and commands in streaming video.
VisRL: Intention-Driven Visual Perception via Reinforced Reasoning
Zhangquan Chen (Tsinghua University), Dongsheng Li (Microsoft Research Asia)
TransformerSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Proposes the VisRL framework, which utilizes reinforcement learning (step-level DPO) to achieve intention-driven visual perception and reasoning through self-evolution via task rewards without any bounding box annotations.
VistaDream: Sampling multiview consistent images for single-view scene reconstruction
Haiping Wang (Wuhan University), Bisheng Yang (Wuhan University)
GenerationDepth EstimationVision Language ModelDiffusion modelGaussian SplattingImage
🎯 What it does: Utilizing diffusion models and 3D Gaussian Splatting to reconstruct renderable 3D scenes from single-view images, and enhancing reconstruction quality through Multiview Consistency Sampling (MCS).
Visual Chronicles: Using Multimodal LLMs to Analyze Massive Collections of Images
Boyang Deng (Stanford University), Thomas Funkhouser (Google DeepMind)
RecognitionRetrievalAnomaly DetectionOptimizationTransformerLarge Language ModelVision Language ModelImageMultimodalityTime Series
🎯 What it does: Using a multimodal large language model (MLLM) for unsupervised open trend discovery on tens of millions of timestamped images from Google Street View, automatically identifying visual change patterns in cities over a decade;
Visual Intention Grounding for Egocentric Assistants
Pengzhan Sun (National University of Singapore), Angela Yao (National University of Singapore)
Object DetectionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a task of inferring user intentions from a first-person perspective and locating corresponding target objects in a scene. A large-scale dataset called EgoIntention is constructed, and based on this, instruction fine-tuning of multimodal large language models is performed.
Visual Interestingness Decoded: How GPT-4o Mirrors Human Interests
Fitim Abdullahu (Zurich University of Applied Sciences), Helmut Grabner (Zurich University of Applied Sciences)
Recommendation SystemKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringImageVideoMultimodality
🎯 What it does: This study investigates the performance of large multimodal models (LMM) like GPT-4o in recognizing visual interestingness, evaluates its consistency with human assessments, and utilizes model-generated image comparison labels to train a learning ranking model to predict image interestingness.
Visual Modality Prompt for Adapting Vision-Language Object Detectors
Heitor R. Medeiros (ETS Montreal), Marco Pedersoli (ETS Montreal)
Object DetectionDomain AdaptationConvolutional Neural NetworkTransformerPrompt EngineeringImageMultimodality
🎯 What it does: This paper proposes ModPrompt, a visual prompt strategy based on an encoder-decoder framework, designed to transfer open-source vision-language object detectors (such as YOLO-World and Grounding DINO) to non-RGB visual modalities like infrared, depth, event, and LiDAR without compromising zero-shot capabilities.
Visual Relation Diffusion for Human-Object Interaction Detection
Ping Cao (Beijing Jiaotong University), Yao Zhao (Beijing Jiaotong University)
Object DetectionDiffusion modelImage
🎯 What it does: This paper proposes a Visual Relation Diffusion Model (VRDiff) that utilizes a diffusion process based solely on visual conditions to provide fine-grained feedback on human-object interaction features, achieving high-precision HOI detection.
Visual Surface Wave Elastography: Revealing Subsurface Physical Properties via Visible Surface Waves
Alexander C. Ogren (California Institute of Technology), Chiara Daraio (California Institute of Technology)
OptimizationOptical FlowVideoUltrasound
🎯 What it does: A visual surface wave elastography method (VSWE) is proposed, which extracts the dispersion relationship from visible surface fluctuations in videos and infers the thickness and elastic modulus of structures through physical simulation and optimization.
Visual Test-time Scaling for GUI Agent Grounding
Tiange Luo (University of Michigan), Honglak Lee (University of Michigan)
Vision Language ModelImage
🎯 What it does: This paper proposes RegionFocus, a method for dynamically focusing on GUI sub-regions during interaction to enhance the localization accuracy of visual language model agents.
Visual Textualization for Image Prompted Object Detection
Yongjian Wu (Beihang University), Yan Xu (Beihang University)
Object DetectionTransformerVision Language ModelImageText
🎯 What it does: A visual textualization method is proposed to map a small number of support images into the text feature space, enabling image prompting for open-set object detection without altering the structure of the pre-trained OVLM model.
Visual-Oriented Fine-Grained Knowledge Editing for MultiModal Large Language Models
Zhen Zeng (Hefei University of Technology), Meng Wang (Hefei University of Technology)
Knowledge DistillationTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Proposes a vision-guided fine-grained multimodal knowledge editing task, constructs the FGVEdit evaluation benchmark, and designs the Multimodal Scope Classifier-based Knowledge Editor (MSCKE) framework to implement this task.
Visual-RFT: Visual Reinforcement Fine-Tuning
Ziyu Liu (Shanghai Jiao Tong University), Jiaqi Wang (Shanghai Jiao Tong University)
ClassificationObject DetectionTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelImageMultimodality
🎯 What it does: This paper proposes the Visual-RFT method, which applies verifiable rewards for reinforcement fine-tuning on large-scale visual language models (LVLM), enabling the model to enhance its visual reasoning and perception capabilities through self-trial and error with limited data.
VisualCloze: A Universal Image Generation Framework via Visual In-Context Learning
Zhong-Yu Li (NKIARI), Ming-Ming Cheng (NKIARI)
RestorationGenerationGraph Neural NetworkDiffusion modelImageGraph
🎯 What it does: This paper proposes VisualCloze, a general image generation framework achieved through learning from visual context.
ViT-EnsembleAttack: Augmenting Ensemble Models for Stronger Adversarial Transferability in Vision Transformers
Hanwen Cao (Huazhong University of Science and Technology), Kun He (Huazhong University of Science and Technology)
Adversarial AttackTransformerImage
🎯 What it does: This paper proposes ViT-EnsembleAttack, an ensemble adversarial attack method targeting Vision Transformers, which enhances attack transferability through model adversarial augmentation.
ViT-Linearizer: Distilling Quadratic Knowledge into Linear-Time Vision Models
Guoyizhe Wei (Johns Hopkins University), Rama Chellappa (Johns Hopkins University)
ClassificationSegmentationKnowledge DistillationRecurrent Neural NetworkTransformerImage
🎯 What it does: This paper proposes a cross-architecture distillation framework called ViT-Linearizer, which transfers the quadratic complexity knowledge of ViT to linear-time RNN-style visual models (such as Adventurer).