CVPR 2025 Papers — Page 28

Unlocking the Potential of Unlabeled Data in Semi-Supervised Domain Generalization

Dongkwan Lee (Seoul National University), Nojun Kwak (Seoul National University)

🎯 What it does: This paper proposes a module named UPCSC, which utilizes unlabeled samples with low model confidence in semi-supervised domain generalization to enhance the model's performance in unseen domains.

UNOPose: Unseen Object Pose Estimation with an Unposed RGB-D Reference Image

Xingyu Liu (Tsinghua University), Xiangyang Ji (Tsinghua University)

🎯 What it does: The UNOPose framework is proposed to solve the unknown object 6DoF pose estimation under a single unposed RGB-D reference image.

Unraveling Normal Anatomy via Fluid-Driven Anomaly Randomization

Peirong Liu (Harvard Medical School and Massachusetts General Hospital), Juan E. Iglesias (UCL)

🎯 What it does: The UNA model is proposed, utilizing fluid-driven anomaly randomization to achieve healthy reconstruction and anomaly detection of multimodal (CT/MRI) brain structures.

Unseen Visual Anomaly Generation

Han Sun (École Polytechnique Fédérale de Lausanne), Olga Fink (École Polytechnique Fédérale de Lausanne)

🎯 What it does: The AnomalyAny framework is proposed, utilizing Stable Diffusion to generate unseen, realistic, and diverse visual anomalies with only one normal image and an anomaly description.

Unsupervised Continual Domain Shift Learning with Multi-Prototype Modeling

Haopeng Sun (Chinese Academy of Sciences), Yiqiang Chen (Chinese Academy of Sciences)

🎯 What it does: A Multi-Prototype Modeling (MPM) is proposed for unsupervised continual domain shift learning, addressing domain generalization, domain adaptation, and catastrophic forgetting.

Unsupervised Discovery of Facial Landmarks and Head Pose

Satyajit Tourani (International Institute of Information Technology Hyderabad), Muhammad Haris Khan (Mohamed bin Zayed University of Artificial Intelligence)

🎯 What it does: This paper proposes an unsupervised method utilizing Stable Diffusion to jointly discover facial key points and head poses, refining key point localization through self-supervised cyclic consistency, image-aware text embedding, and 3D-based rotation rendering enhancement.

Unsupervised Foundation Model-Agnostic Slide-Level Representation Learning

Tim Lenz (EKFZ for Digital Health TU Dresden), Jakob N. Kather (EKFZ for Digital Health TU Dresden)

🎯 What it does: A novel unsupervised contrastive learning framework called COBRA is proposed, which generates full-slice low-dimensional representations in the feature space using multi-scale features from multiple base models.

Unveil Inversion and Invariance in Flow Transformer for Versatile Image Editing

Pengcheng Xu, Boyu Wang

🎯 What it does: This paper proposes an unsupervised image editing framework based on the flow transformer (MM-DiT), utilizing two-stage reverse and AdaLN control to achieve precise retention of invariant content from the original image and diverse editing.

Unveiling Differences in Generative Models: A Scalable Differential Clustering Approach

Jingwei Zhang (Chinese University of Hong Kong), Farzan Farnia (Chinese University of Hong Kong)

🎯 What it does: This paper proposes a scalable differential clustering method based on random Fourier features, called FINC, to identify sample types with significant differences in generation frequency between two generative models.

Unveiling the Ignorance of MLLMs: Seeing Clearly, Answering Incorrectly

Yexin Liu (Hong Kong University of Science and Technology), Bo Zhao (Shanghai Jiao Tong University)

🎯 What it does: This paper proposes the MMVU benchmark to evaluate the phenomenon where multimodal large language models still provide incorrect answers after visual understanding, and enhances the model's robustness by constructing positive and negative paired samples and prompt strategies.

Unveiling the Mist over 3D Vision-Language Understanding: Object-centric Evaluation with Chain-of-Analysis

Jiangyong Huang (State Key Laboratory of General Artificial Intelligence), Siyuan Huang (State Key Laboratory of General Artificial Intelligence)

🎯 What it does: The BEACON3D benchmark is proposed, providing an object-level multi-testing and chain analysis framework for evaluating the performance of 3D vision-language models in grounding and question-answering (QA) tasks.

Unveiling Visual Perception in Language Models: An Attention Head Analysis Approach

Jing Bi (University of Rochester), Chenliang Xu (University of Rochester)

🎯 What it does: This paper systematically analyzes the attention heads in multimodal large language models (MLLMs) and discovers the existence of visual attention heads that specifically focus on visual tokens. It quantitatively evaluates their behavior under different models, scales, and training strategies.

UPME: An Unsupervised Peer Review Framework for Multimodal Large Language Model Evaluation

Qihui Zhang (Peking University), Li Yuan (Peking University)

🎯 What it does: This paper proposes a completely unsupervised multimodal large language model evaluation framework called UPME, which automatically generates questions using image data and evaluates model answers through peer review, eliminating the cost of manual Q&A alignment.

UrbanCAD: Towards Highly Controllable and Photorealistic 3D Vehicles for Urban Scene Simulation

Yichong Lu (Zhejiang University), Yiyi Liao (Zhejiang University)

🎯 What it does: This paper proposes the UrbanCAD framework, which can automatically generate highly controllable and realistic 3D digital twins of vehicles from a single urban image.

URWKV: Unified RWKV Model with Multi-state Perspective for Low-light Image Restoration

Rui Xu (Fuzhou University), Yuzhong Chen (Fuzhou University)

🎯 What it does: This paper proposes a multi-state unified low-light image enhancement and deblurring model URWKV, designed to handle dynamically coupled low-light and blur degradation.

Using Diffusion Priors for Video Amodal Segmentation

Kaihua Chen (Carnegie Mellon University), Tarasha Khurana (Carnegie Mellon University)

🎯 What it does: This paper proposes a two-stage video amodal segmentation and content completion method, which first uses a video diffusion model to generate a complete amodal mask based on visible occlusion information, and then completes the RGB content of the occluded areas based on this.

Using Powerful Prior Knowledge of Diffusion Model in Deep Unfolding Networks for Image Compressive Sensing

Chen Liao (Beijing Jiaotong University), Zhongli Wang (Beijing Jiaotong University)

🎯 What it does: A deep unfolding network DMP-DUN based on a pre-trained diffusion model is proposed, achieving high-quality reconstruction of image compressed sensing with very few iterations (only 2 steps).

USP-Gaussian: Unifying Spike-based Image Reconstruction, Pose Correction and Gaussian Splatting

Kang Chen (Peking University), Zhaofei Yu (Peking University)

🎯 What it does: A joint optimization USP-Gaussian framework is proposed, which integrates spike-based image reconstruction, pose correction, and 3D Gaussian splatting into a single end-to-end process.

UVGS: Reimagining Unstructured 3D Gaussian Splatting using UV Mapping

Aashish Rai (Brown University), Aayush Prakash (Meta Reality Labs)

🎯 What it does: By using spherical mapping, the unordered high-dimensional point set of 3D Gaussian Splatting (3DGS) is mapped to a two-dimensional UV map (UVGS), which is then compressed into a 3-channel Super UVGS using a multi-branch CNN, enabling it to directly work with two-dimensional pre-trained models (such as VAE and diffusion models) for generation, compression, and repair.

UWAV: Uncertainty-weighted Weakly-supervised Audio-Visual Video Parsing

Yung-Hsuan Lai (National Taiwan University), Moitreya Chatterjee (Mitsubishi Electric Research Labs)

🎯 What it does: A new framework for weakly supervised audio-visual video parsing (AVVP) called UWAV is proposed, which utilizes a pre-trained Transformer to generate temporally coherent pseudo-labels and combines uncertainty-weighted training.

v-CLR: View-Consistent Learning for Open-World Instance Segmentation

Chang-Bin Zhang (University of Hong Kong), Kai Han (University of Hong Kong)

🎯 What it does: A view-consistent learning (v-CLR) framework is proposed for open-world instance segmentation, enabling the model to learn appearance-invariant features from different perspectives (natural images, depth maps, stylized images, etc.).

V-Stylist: Video Stylization via Collaboration and Reflection of MLLM Agents

Zhengrong Yue (Shanghai Jiao Tong University), Yali Wang (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences)

🎯 What it does: A multi-agent system called V-Stylist is proposed, which can automatically convert long videos into the desired artistic style based on user open-ended descriptions.

V^2Dial: Unification of Video and Visual Dialog via Multimodal Experts

Adnen Abdessaied, Andreas Bulling

🎯 What it does: A unified multimodal expert model V Dial 2 is proposed, capable of simultaneously processing image and video inputs to generate dialogue responses.

V2V3D: View-to-View Denoised 3D Reconstruction for Light Field Microscopy

Jiayin Zhao (Tsinghua University), Hui Qiao (Tsinghua University)

🎯 What it does: An unsupervised perspective mutual mapping framework V2V3D is proposed for image denoising and 3D reconstruction in light field microscopy.

V2X-R: Cooperative LiDAR-4D Radar Fusion with Denoising Diffusion for 3D Object Detection

Xun Huang (Xiamen University), Chenglu Wen (Xiamen University)

🎯 What it does: Designed and released the first V2X collaborative perception dataset containing 4D radar, V2X-R, and implemented a LiDAR-4D radar multimodal fusion framework based on this dataset, while also proposing a Multi-modal Denoising Diffusion (MDD) denoising module.

Variance-Based Membership Inference Attacks Against Large-Scale Image Captioning Models

Daniel Samira (Ben Gurion University of the Negev), Asaf Shabtai (Ben Gurion University of the Negev)

🎯 What it does: For image description generation models, two variance-based membership inference attack methods (MVTA and C-WSA) are proposed, which do not require training shadow models and can determine whether a sample belongs to the training set using only image input.

VASparse: Towards Efficient Visual Hallucination Mitigation via Visual-Aware Token Sparsification

Xianwei Zhuang (Peking University), Yuexian Zou (Peking University)

🎯 What it does: This paper proposes VASparse, an efficient decoding strategy for large visual language models (LVLM) aimed at alleviating visual hallucination (VH).

VasTSD: Learning 3D Vascular Tree-state Space Diffusion Model for Angiography Synthesis

Zhifeng Wang (National University of Defense Technology), Kai Xu (National University of Defense Technology)

🎯 What it does: This paper proposes a three-dimensional vascular tree state space diffusion model, VasTSD, for synthesizing vascular angiography images from non-contrast medical volume data.

VDocRAG: Retrieval-Augmented Generation over Visually-Rich Documents

Ryota Tanaka (NTT Corporation), Jun Suzuki (Tohoku University)

🎯 What it does: This paper proposes a Retrieval-Augmented Generation (RAG) framework called VDocRAG, which directly understands multimodal visually rich documents in image form and answers questions, avoiding information loss caused by traditional text parsing.

VELOCITI: Benchmarking Video-Language Compositional Reasoning with Strict Entailment

Darshana Saravanan (International Institute of Information Technology Hyderabad), Makarand Tapaswi (International Institute of Information Technology Hyderabad)

🎯 What it does: Designed and released the VELOCITI benchmark to evaluate the reasoning ability of video-language models in short videos regarding agents, actions, and multi-event combinations.

VERA: Explainable Video Anomaly Detection via Verbalized Learning of Vision-Language Models

Muchao Ye (University of Iowa), Pan He (Auburn University)

🎯 What it does: This paper proposes the VERA framework, which utilizes a frozen visual-language model (VLM) to self-learn a set of guiding questions for anomaly patterns through 'semantic learning', achieving interpretable video anomaly detection (VAD) without parameter updates or additional reasoning modules. During the inference phase, a refined scoring process is employed that incorporates segmentation, scene context fusion, and temporal smoothing.

VerbDiff: Text-Only Diffusion Models with Enhanced Interaction Awareness

SeungJu Cha (Hanyang University), Dong-Jin Kim (Hanyang University)

🎯 What it does: A text-based diffusion model called VerbDiff is proposed, focusing on improving the accuracy and detail representation of human-object interactions.

vesselFM: A Foundation Model for Universal 3D Blood Vessel Segmentation

Bastian Wittmann (University of Zurich), Bjoern Menze (University of Zurich)

🎯 What it does: This paper proposes vesselFM, a foundational model for 3D vascular segmentation based on three heterogeneous data sources (real vascular images, domain-randomized synthetic images, and flow-matching generated images);

VEU-Bench: Towards Comprehensive Understanding of Video Editing

Bozheng Li (Brown University), Wenbo Zhu (Berkeley)

🎯 What it does: A VUE-Bench video editing understanding benchmark covering 10 dimensions and 19 fine-grained tasks has been constructed, and an excellent Vid-LLM model called Oscars has been trained based on this.

VGGT: Visual Geometry Grounded Transformer

Jianyuan Wang (Visual Geometry Group University of Oxford), David Novotny (Meta AI)

🎯 What it does: This paper presents the Visual Geometry Grounded Transformer (VGGT), a large feedforward transformer capable of predicting camera parameters, depth maps, point clouds, and point tracking, among other 3D attributes, from any number of views at once.

VI^3NR: Variance Informed Initialization for Implicit Neural Representations

Chamin Hewa Koneputugodage (Australian National University), Stephen Gould (Australian National University)

🎯 What it does: A universal weight initialization method for implicit neural representations (INR) is proposed, ensuring stable pre-activation distribution across layers and applicability to any activation function.

ViCaS: A Dataset for Combining Holistic and Pixel-level Video Understanding using Captions with Grounded Segmentation

Ali Athar (ByteDance Inc), Liang-Chieh Chen (ByteDance Inc)

Vid2Avatar-Pro: Authentic Avatar from Videos in the Wild via Universal Prior

Chen Guo (Meta), Chen Cao (Meta)

🎯 What it does: Generate animatable and realistic 3D human avatars from monocular outdoor videos, capable of rendering in any pose and perspective.

Vid2Sim: Generalizable, Video-based Reconstruction of Appearance, Geometry and Physics for Mesh-free Simulation

Chuhao Chen (University of Pennsylvania), Lingjie Liu (University of Pennsylvania)

🎯 What it does: The Vid2Sim framework is proposed, which can quickly reconstruct texture geometry and physical properties from multi-view videos and achieve efficient meshless, LBS-based elastic simulation.

Vid2Sim: Realistic and Interactive Simulation from Video for Urban Navigation

Ziyang Xie (University of Illinois Urbana-Champaign), Bolei Zhou (University of California)

🎯 What it does: This paper presents the Vid2Sim framework, which utilizes monocular video to construct realistic and interactive 3D simulation environments for training urban navigation agents.

VidBot: Learning Generalizable 3D Actions from In-the-Wild 2D Human Videos for Zero-Shot Robotic Manipulation

Hanzhi Chen (Technical University of Munich), Stefan Leutenegger (ETH Zurich)

🎯 What it does: Proposes the VidBot framework, which utilizes everyday RGB video to learn transferable 3D affordances, achieving zero-shot robot manipulation;

VidComposition: Can MLLMs Analyze Compositions in Compiled Videos?

Yunlong Tang (University of Rochester), Chenliang Xu (University of Rochester)

🎯 What it does: The VidComposition benchmark is proposed to evaluate the compositional understanding ability of multimodal large language models in compiling videos.

Video Depth Anything: Consistent Depth Estimation for Super-Long Videos

Sili Chen (ByteDance), Bingyi Kang (ByteDance)

🎯 What it does: This paper proposes Video Depth Anything, achieving seamless and consistent depth estimation for long videos.

Video Depth without Video Models

Bingxin Ke (ETH Zurich), Konrad Schindler (ETH Zurich)

🎯 What it does: This paper proposes RollingDepth, a rolling inference framework based on a single-image diffusion model that can generate temporally consistent depth videos without a video model.

Video Language Model Pretraining with Spatio-temporal Masking

Yue Wu (Computer Vision Foundation), Shuhui Wang (Computer Vision Foundation)

🎯 What it does: This paper proposes a spatiotemporal consistency-based masking and reconstruction strategy (STM) for self-supervised pre-training of video-language models;

Video Motion Transfer with Diffusion Transformers

Alexander Pondaven (University of Oxford), Fabio Pizzati (Mohamed bin Zayed University of Artificial Intelligence)

🎯 What it does: The DiTFlow method is proposed, which utilizes the attention of the Diffusion Transformer to extract Attention Motion Flow (AMF) and transfers the motion from the reference video to the newly generated video through an untrained optimization approach.

Video Summarization with Large Language Models

Min Jung Lee (Pohang University of Science and Technology), Minsu Cho (Pohang University of Science and Technology)

🎯 What it does: A video summarization framework called LLMVS is proposed, which utilizes a large language model (LLM) to generate frame descriptions, then evaluates the importance of key frames within a local window using the LLM, and finally aggregates the results into a final summary through global self-attention.

Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding

Duo Zheng (Chinese University of Hong Kong), Liwei Wang (Chinese University of Hong Kong)

🎯 What it does: This paper proposes Video-3D LLM, which directly constructs location-aware video representations using video frames and corresponding 3D coordinates to accomplish tasks such as 3D visual localization, dense description, and question answering.

Video-Bench: Human-Aligned Video Generation Benchmark

Hui Han (Shanghai Jiao Tong University), Yongxin Ni (National University of Singapore)

🎯 What it does: This paper presents Video-Bench, a comprehensive evaluation benchmark for video generation that combines a multi-dimensional evaluation system (video-condition alignment and video quality) and achieves automated assessment through a multi-modal large language model (MLLM);

Video-ColBERT: Contextualized Late Interaction for Text-to-Video Retrieval

Arun Reddy (Johns Hopkins University), Rama Chellappa (Johns Hopkins University)

🎯 What it does: A text-to-video retrieval model called Video-ColBERT is proposed, which can compute similarity using unidirectional approximate interactions during retrieval, maintaining efficiency;

Video-Guided Foley Sound Generation with Multimodal Controls

Ziyang Chen (University of Michigan), Justin Salamon (Adobe Research)

🎯 What it does: Design and implement MultiFoley, a multimodal (text, audio, video) conditional Foley sound generation model that can generate synchronized and controllable full-bandwidth (48kHz) sound effects for silent videos.

Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis

Chaoyou Fu (Nanjing University), Xing Sun

🎯 What it does: The Video-MME benchmark is proposed to evaluate the capabilities of large language models in video analysis through multimodal question answering.

Video-Panda: Parameter-efficient Alignment for Encoder-free Video-Language Models

Jinhui Yi (University of Bonn), Juergen Gall (University of Bonn)

🎯 What it does: An efficient video-language model called Video-Panda is proposed, which utilizes a Spatio-Temporal Alignment Block (STAB) to directly extract spatio-temporal features from video frames and align them with the language model.

Video-XL: Extra-Long Vision Language Model for Hour-Scale Video Understanding

Yan Shu (Shanghai Jiao Tong University), Bo Zhao (Shanghai Jiao Tong University)

🎯 What it does: This study focuses on long video understanding and proposes the Video-XL model, which utilizes Visual Summary Tokens (VST) to compress videos to fit the context length limitations of large multimodal language models.

VideoAutoArena: An Automated Arena for Evaluating Large Multimodal Models in Video Analysis through User Simulation

Ziyang Luo (Salesforce AI Research), Junnan Li (Salesforce AI Research)

🎯 What it does: Designed and implemented VideoAutoArena and VideoAutoBench, which utilize LMM-driven user simulations to automatically generate open-ended, adaptive questions for competitions and automatic judging, thereby evaluating the capabilities of large multimodal models (LMM) in video analysis;

VideoComp: Advancing Fine-Grained Compositional and Temporal Alignment in Video-Text Models

Dahun Kim (Google DeepMind), Anelia Angelova (Google DeepMind)

🎯 What it does: This paper proposes the Compositionality Benchmark (ActivityNet-Comp and YouCook2-Comp) for evaluating and enhancing fine-grained temporal consistency between video and text, along with a corresponding training framework, aiming to enable models to better capture the temporal and semantic relationships in multi-event videos.

VideoDirector: Precise Video Editing via Text-to-Video Models

Yukun Wang (Sun Yat-sen University), Yulan Guo (Sun Yat-sen University)

🎯 What it does: Proposes VideoDirector, which directly uses text-to-video models for precise video editing;

VideoDPO: Omni-Preference Alignment for Video Diffusion Generation

Runtao Liu (Hong Kong University of Science and Technology), Qifeng Chen (Hong Kong University of Science and Technology)

🎯 What it does: This paper proposes VideoDPO, which optimizes video diffusion models using automatically generated preference alignment data to better align with user preferences.

VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection

Songhao Han (Beihang University), Si Liu (Beihang University)

🎯 What it does: This paper constructs a large-scale video question-answering dataset named VideoEspresso, which automatically generates fine-grained reasoning questions and annotates multimodal chain-of-thought (CoT), while proposing a hybrid LVLM collaboration framework to achieve efficient video reasoning.

VideoGEM: Training-free Action Grounding in Videos

Felix Vogel (Goethe University Frankfurt), Hilde Kuehne (MPI for Informatics)

🎯 What it does: A training-free visual language model video action localization method called VideoGEM is proposed, which can directly utilize pre-trained CLIP, OpenCLIP, and ViCLIP to spatially locate actions in videos.

VideoGigaGAN: Towards Detail-rich Video Super-Resolution

Yiran Xu (Adobe Research), Difan Liu (University of Maryland)

🎯 What it does: This paper presents VideoGigaGAN, a GAN-based video super-resolution model that can generate high-frequency details while maintaining temporal consistency.

VideoGLaMM : A Large Multimodal Model for Pixel-Level Visual Grounding in Videos

Shehan Munasinghe (Mohamed bin Zayed University of Artificial Intelligence), Salman Khan (Australian National University)

🎯 What it does: We propose and train VideoGLaMM, an end-to-end multimodal video large model that can generate natural language answers containing pixel-level spatiotemporal segmentation masks in response to user queries.

VideoGuide: Improving Video Diffusion Models without Training Through a Teacher's Guide

Dohun Lee (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)

🎯 What it does: Enhancement of temporal consistency for pre-trained video diffusion models without the need for additional training or fine-tuning.

VideoHandles: Editing 3D Object Compositions in Videos Using Video Generative Priors

Juil Koo (KAIST), Minhyuk Sung (KAIST)

🎯 What it does: Proposes VideoHandles, a zero-shot method for 3D object composition editing using the priors of video generation models;

VideoICL: Confidence-based Iterative In-context Learning for Out-of-Distribution Video Understanding

Kangsan Kim (Korea Advanced Institute of Science and Technology), Sung Ju Hwang (Korea Advanced Institute of Science and Technology)

🎯 What it does: A training-agnostic video context learning framework (VideoICL) for out-of-distribution (OOD) video tasks is proposed, which enhances the generalization ability of video multimodal models by dynamically selecting the most relevant examples during inference and performing iterative reasoning.

VideoMage: Multi-Subject and Motion Customization of Text-to-Video Diffusion Models

Chi-Pin Huang (National Taiwan University), Yu-Chiang Frank Wang (NVIDIA)

🎯 What it does: This paper presents the VideoMage framework, which implements multi-entity and multi-motion customization for text-to-video diffusion models, allowing users to specify multiple character images and interactive actions to generate videos that align with text prompts.

VideoRefer Suite: Advancing Spatial-Temporal Object Understanding with Video LLM

Yuqian Yuan (Zhejiang University), Lidong Bing (Shanda AI Research Institute)

🎯 What it does: The VideoRefer Suite is proposed, which includes a large high-quality object-level video instruction dataset VideoRefer-700K, a VideoRefer model capable of capturing spatial-temporal fine-grained information, and VideoRefer-Bench for evaluating region-level video understanding.

VideoScene: Distilling Video Diffusion Model to Generate 3D Scenes in One Step

Hanyang Wang (Tsinghua University), Yueqi Duan (Tsinghua University)

🎯 What it does: Utilize video diffusion models for first-order 3D scene generation from sparse views;

VideoSPatS: Video SPatiotemporal Splines for Disentangled Occlusion, Appearance and Motion Modeling and Editing

Juan Luis Gonzalez (Flawless AI), Pablo Garrido (Flawless AI)

🎯 What it does: This paper proposes an implicit video representation method called VideoSPatS, which can separate occlusion, appearance, and motion from monocular videos and supports consistent video editing.

VideoTree: Adaptive Tree-based Video Representation for LLM Reasoning on Long Videos

Ziyang Wang (University of North Carolina Chapel Hill), Mohit Bansal (University of North Carolina Chapel Hill)

🎯 What it does: This paper presents VIDEOTREE, a training-free adaptive hierarchical video representation framework for LLM inference on long videos.

VideoWorld: Exploring Knowledge Learning from Unlabeled Videos

Zhongwei Ren (Beijing Jiaotong University), Xiaojie Jin (ByteDance Seed)

🎯 What it does: Utilizing unlabeled video to train autoregressive video generation models, exploring the learning of rules, reasoning, and planning knowledge from pure visual input, proposing the VideoWorld framework and Latent Dynamics Model (LDM);

VidHalluc: Evaluating Temporal Hallucinations in Multimodal Large Language Models for Video Understanding

Chaoyu Li (Arizona State University), Pooyan Fazli (Arizona State University)

🎯 What it does: This paper studies the hallucination problem of multimodal large language models in video understanding, proposing the largest hallucination benchmark VIDHALLUC and introducing the training-independent DINO-HEAL method.

VidMuse: A Simple Video-to-Music Generation Framework with Long-Short-Term Modeling

Zeyue Tian (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)

🎯 What it does: A large-scale video-music pair dataset V2M was constructed, and the VidMuse framework was proposed to generate music solely based on video.

VidSeg: Training-free Video Semantic Segmentation based on Diffusion Models

Qian Wang (King Abdullah University of Science and Technology), Peter Wonka (Max Planck Institute for Informatics)

🎯 What it does: Proposes the first training-free video semantic segmentation method based on diffusion models, VidSeg.

VidTwin: Video VAE with Decoupled Structure and Dynamics

Yuchi Wang (Peking University), Jiang Bian (Microsoft Research Asia)

🎯 What it does: This paper proposes VidTwin, a model based on video autoencoders that can encode videos into two complementary latent spaces: Structure and Dynamics, for efficient compression and generation.

Viewpoint Rosetta Stone: Unlocking Unpaired Ego-Exo Videos for View-invariant Representation Learning

Mi Luo (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)

🎯 What it does: By combining paired and unpaired perspective videos, VIEWPOINTROSETTA proposes a viewpoint-invariant video representation.

ViiNeuS: Volumetric Initialization for Implicit Neural Surface Reconstruction of Urban Scenes with Limited Image Overlap

Hala Djeghim (Huawei Paris Research Center), Désiré Sidibé (Univ. Evry Paris-Saclay)

🎯 What it does: We propose ViiNeuS, a hybrid implicit surface reconstruction framework that can quickly construct large-scale urban 3D scenes from limited overlapping street view RGB images.

ViKIENet: Towards Efficient 3D Object Detection with Virtual Key Instance Enhanced Network

Zhuochen Yu (Nanyang Technological University), Andy W. H. Khong (Nanyang Technological University)

🎯 What it does: This paper proposes a LiDAR+RGB 3D object detection framework named ViKIENet, which achieves efficient multimodal feature fusion through virtual key instances (VKIs). It includes three main modules: Semantic Key Instance Selection (SKIS), Virtual Instance Focused Fusion (VIFF), and Virtual Instance to Real Attention (VIRA), and presents a variant ViKIENet-R that incorporates rotation-invariant features.

VILA-M3: Enhancing Vision-Language Models with Medical Expert Knowledge

Vishwesh Nath (NVIDIA), Daguang Xu (NVIDIA)

🎯 What it does: Proposes the VILA-M3 framework, integrating medical expert model knowledge into a multimodal language model, achieving four-stage training and improving performance on medical tasks.

VinaBench: Benchmark for Faithful and Consistent Visual Narratives

Silin Gao (EPFL), Antoine Bosselut (Sony Group Corporation)

🎯 What it does: A benchmark called VinaBench is proposed to evaluate and improve the authenticity and consistency of visual narrative generation models.

VinTAGe: Joint Video and Text Conditioning for Holistic Audio Generation

Saksham Singh Kushwaha (University of Texas at Dallas), Yapeng Tian (University of Texas at Dallas)

🎯 What it does: A Transformer model based on flow matching, VinTAGe, is proposed to simultaneously utilize video and text inputs to generate complete audio (including both in-scene and out-of-scene sounds).

VIRES: Video Instance Repainting via Sketch and Text Guided Generation

Shuchen Weng (Peking University), Boxin Shi (Peking University)

🎯 What it does: A VIRES framework based on ControlNet and diffusion models is proposed, enabling video instance re-painting, replacement, generation, and deletion guided by sketches and text.

VISCO: Benchmarking Fine-Grained Critique and Correction Towards Self-Improvement in Visual Reasoning

Xueqing Wu (University of California), Nanyun Peng (Stanford University)

🎯 What it does: A VISCO benchmark is proposed for systematically evaluating the fine-grained capabilities of visual language models in self-critique and correction.

Vision-Guided Action: Enhancing 3D Human Motion Prediction with Gaze-informed Affordance in 3D Scenes

Ting Yu (Hangzhou Normal University), Qiongjie Cui (Singapore University of Technology and Design)

🎯 What it does: A framework named GAP3DS is proposed, which is based on human gaze information for operability-aware 3D human motion prediction. It can generate future motion trajectories and poses with greater physical feasibility and semantic consistency in complex 3D scenes.

Vision-Language Embodiment for Monocular Depth Estimation

Jinchang Zhang (University of Georgia), Guoyu Lu (Binghamton University)

🎯 What it does: A monocular depth estimation framework is proposed that combines the intrinsic parameters of the camera model with language priors, utilizing the physical characteristics of the camera to generate embedded scene depth and fuse it with RGB features.

Vision-Language Gradient Descent-driven All-in-One Deep Unfolding Networks

Haijin Zeng (Harvard University), Jie Liu (Harbin Institute of Technology)

🎯 What it does: A unified VLU-Net deep unfolding network is proposed, utilizing fine-tuned CLIP for visual-language guidance to automatically identify and restore various image distortions (noise, blur, haze, rain, low light, etc.).

Vision-Language Model IP Protection via Prompt-based Learning

Lianyu Wang (Key Laboratory of Brain-Machine Intelligence Technology Ministry of Education China), Daoqiang Zhang (Key Laboratory of Brain-Machine Intelligence Technology Ministry of Education China)

🎯 What it does: This paper proposes a prompt-based IP protection framework called IP-CLIP, designed to limit the reasoning capabilities of visual language models like CLIP in unauthorized domains.

Vision-Language Models Do Not Understand Negation

Kumail Alhamoud (Massachusetts Institute of Technology), Marzyeh Ghassemi (Massachusetts Institute of Technology)

🎯 What it does: A NegBench benchmark was constructed to evaluate the understanding of negation by visual language models, and it was found that existing models perform poorly on negation tasks.

VisionArena: 230k Real World User-VLM Conversations with Preference Labels

Christopher Chou (Stanford University), Wei-Lin Chiang (University of California Berkeley)

🎯 What it does: This paper constructs the VisionArena dataset, collecting 230K dialogues between real users and visual language models (VLMs), and generates a VLM competition ranking through user preference voting.

VisionPAD: A Vision-Centric Pre-training Paradigm for Autonomous Driving

Haiming Zhang (FNii), Zhen Li (SSE)

🎯 What it does: This paper proposes a self-supervised pre-training framework named VisionPAD, which learns 3D geometry and motion representations for vehicle vision tasks using only multi-frame multi-view images.

VisionZip: Longer is Better but Not Necessary in Vision Language Models

Senqiao Yang (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)

🎯 What it does: This paper proposes VisionZip, a method that significantly reduces the number of visual tokens and improves efficiency without sacrificing model performance by selecting dominant visual tokens and merging the remaining tokens.

VISTA: Enhancing Long-Duration and High-Resolution Video Understanding by Video Spatiotemporal Augmentation

Weiming Ren (University of Waterloo), Wenhu Chen (University of Waterloo)

🎯 What it does: The VISTA framework is proposed, which generates long-term, high-definition video instruction pairs by stitching existing video-caption data in both space and time, thereby constructing a high-quality synthetic instruction dataset VISTA-400K with 400,000 entries.

VISTA3D: A Unified Segmentation Foundation Model For 3D Medical Imaging

Yufan He (NVIDIA), Wenqi Li (NVIDIA)

🎯 What it does: VISTA3D is proposed, a unified 3D medical image segmentation foundation model that achieves high-precision automatic segmentation across 127 supported categories and enables interactive error correction and zero-shot segmentation through 3D point annotations.

VISTREAM: Improving Computation Efficiency of Visual Streaming Perception via Law-of-Charge-Conservation Inspired Spiking Neural Network

Kang You (Shanghai Jiao Tong University), Zhezhi He (Shanghai Jiao Tong University)

🎯 What it does: The VISTREAM framework is researched and proposed, achieving efficient inference of visual flow perception through differential coding and LoCC-SNN.

Visual Agentic AI for Spatial Reasoning with a Dynamic API

Damiano Marsili (California Institute of Technology), Georgia Gkioxari (California Institute of Technology)

🎯 What it does: This paper presents VADAR—a framework for agentic program synthesis based on LLM, capable of dynamically generating and implementing Python APIs to accomplish 3D visual spatial reasoning tasks.

Visual and Semantic Prompt Collaboration for Generalized Zero-Shot Learning

Huajie Jiang (Beijing University of Technology), Yuankai Qi (Macquarie University)

🎯 What it does: This paper proposes a Visual and Semantic Prompt Collaborative Network (VSPCN) that jointly learns more semantically relevant visual features through visual and semantic prompts, thereby enhancing the performance of Generalized Zero-Shot Learning (GZSL).

Visual Consensus Prompting for Co-Salient Object Detection

Jie Wang (Tianjin University), Yahong Han (Tianjin University)

🎯 What it does: This paper proposes a co-salient object detection method based on Visual Consensus Prompt (VCP), freezing the pre-trained Transformer and only adjusting a minimal number of prompt parameters to complete the CoSOD task.

Visual Lexicon: Rich Image Features in Language Space

XuDong Wang (Google DeepMind), Cordelia Schmid (Google DeepMind)

🎯 What it does: ViLex is proposed to map images to text space, achieving visual representation that maintains both semantics and details, supporting unsupervised image reconstruction and multimodal generation, and enhancing the performance of visual language models.

Visual Persona: Foundation Model for Full-Body Human Customization

Jisu Nam (KAIST), Yang Zhou (Adobe Research)

🎯 What it does: This paper presents Visual Persona, a full-body portrait customization model based on large-scale paired alignment data, which can generate diverse and identity-consistent full-body images based on a single portrait and text prompts.

Visual Prompting for One-shot Controllable Video Editing without Inversion

Zhengbo Zhang (Singapore University of Technology and Design), Lin Geng Foo (Singapore University of Technology and Design)

🎯 What it does: A one-click controllable video editing method is proposed that does not require DDIM inversion and can be completed using only visual prompts;