CVPR 2024 Papers — Page 27
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
Video2Game: Real-time Interactive Realistic and Browser-Compatible Environment from a Single Video
Hongchi Xia, Shenlong Wang
GenerationData SynthesisAutonomous DrivingNeural Radiance FieldGaussian SplattingVideoMesh
🎯 What it does: Automatically generate a real-time interactive, physically simulated, and high-quality rendered game environment from a single video.
VideoBooth: Diffusion-based Video Generation with Image Prompts
Yuming Jiang (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationData SynthesisDiffusion modelImageVideoText
🎯 What it does: Utilizing image prompts and text prompts to generate multi-frame high-quality videos in diffusion models, achieving personalized video synthesis.
VideoCon: Robust Video-Language Alignment via Contrast Captions
Hritik Bansal (University of California Los Angeles), Aditya Grover (University of California Los Angeles)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningVideoTextMultimodality
🎯 What it does: The VideoCon dataset was constructed, and various semantically feasible contrastive subtitles and natural language explanations were automatically generated using a large language model, enhancing the robustness of video-text alignment models.
VideoCrafter2: Overcoming Data Limitations for High-Quality Video Diffusion Models
Haoxin Chen (Tencent AI Lab), Ying Shan (Tencent AI Lab)
GenerationData SynthesisDiffusion modelImageVideo
🎯 What it does: This paper proposes a method for training high-quality text-to-video diffusion models in the absence of high-quality video data.
VideoCutLER: Surprisingly Simple Unsupervised Video Instance Segmentation
Xudong Wang (University of California Berkeley), Trevor Darrell (University of California Berkeley)
Object DetectionSegmentationTransformerContrastive LearningVideo
🎯 What it does: Proposes VideoCutLER, a completely unsupervised video instance segmentation framework;
VideoGrounding-DINO: Towards Open-Vocabulary Spatio-Temporal Video Grounding
Syed Talal Wasim (Mohamed bin Zayed University of AI), Fahad Shahbaz Khan (Linkoping University)
RecognitionObject DetectionTransformerVision Language ModelVideoText
🎯 What it does: This paper proposes an open-source vocabulary spatiotemporal video grounding method called VideoGrounding-DINO, which can accurately locate targets in videos under both closed-set and open-set scenarios.
VideoLLM-online: Online Video Large Language Model for Streaming Video
Joya Chen (National University of Singapore), Mike Zheng Shou (National University of Singapore)
TransformerLarge Language ModelVision Language ModelVideoMultimodality
🎯 What it does: Proposed the LIVE framework and implemented VideoLLM-online, achieving real-time multimodal question answering in continuous video streams.
VideoMAC: Video Masked Autoencoders Meet ConvNets
Gensheng Pei (Nanjing University of Science and Technology), Yazhou Yao (Nanjing University of Science and Technology)
SegmentationRepresentation LearningConvolutional Neural NetworkAuto EncoderVideo
🎯 What it does: In self-supervised video pre-training, VideoMAC is proposed, a video mask autoencoder based on pure convolutional networks.
VideoRF: Rendering Dynamic Radiance Fields as 2D Feature Video Streams
Liao Wang (ShanghaiTech University), Minye Wu (KU Leuven)
GenerationCompressionNeural Radiance FieldVideo
🎯 What it does: Proposes VideoRF, enabling dynamic neural radiance fields to be decoded and rendered in real-time on mobile devices.
VideoSwap: Customized Video Subject Swapping with Interactive Semantic Point Correspondence
Yuchao Gu (National University of Singapore), Kevin Tang (Meta)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: The VideoSwap framework is proposed, enabling customizable video subject replacement, supporting shape changes while maintaining the motion trajectory of the source video through sparse semantic point correspondence.
VidLA: Video-Language Alignment at Scale
Mamshad Nayeem Rizve (Amazon Shopping), Trishul Chilimbi (University of Central Florida)
RetrievalTransformerLarge Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: This paper presents VidLA, a scalable video-language alignment model that utilizes hierarchical temporal tokens and large-scale LLMs to generate text data.
VidToMe: Video Token Merging for Zero-Shot Video Editing
Xirui Li (Shanghai Jiao Tong University), Ming-Hsuan Yang (UC Merced)
GenerationData SynthesisTransformerDiffusion modelVideo
🎯 What it does: Utilize a pre-trained image diffusion model for zero-shot video editing, achieving temporal consistency through cross-frame token merging.
View From Above: Orthogonal-View aware Cross-view Localization
Shan Wang (CSIRO), Hongdong Li (Australian National University)
Pose EstimationDomain AdaptationAutonomous DrivingOptimizationTransformerGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a cross-view localization method based on Top-to-Ground Aggregation (T2GA), addressing registration errors caused by neglecting and occluding ground features, significantly improving the alignment accuracy between ground cameras and satellite images.
View-Category Interactive Sharing Transformer for Incomplete Multi-View Multi-Label Learning
Shilong Ou (Beijing University of Posts and Telecommunications), Junjiang Wu (Beijing University of Posts and Telecommunications)
ClassificationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: This paper proposes a View-Category Interactive Sharing Transformer (VIST) model for dual missing multi-view multi-label learning, which can simultaneously complete missing views, utilize the complementary information between views and labels for feature learning, and perform multi-label classification.
View-decoupled Transformer for Person Re-identification under Aerial-ground Camera Network
Quan Zhang (Sun Yat-Sen University), Jianhaung Lai
RecognitionRetrievalTransformerImage
🎯 What it does: This paper proposes a View Decoupling Transformer (VDT) for person re-identification in drone-ground camera networks and constructs a large-scale synthetic dataset called CARGO.
ViewDiff: 3D-Consistent Image Generation with Text-to-Image Models
Lukas Höllein (Technical University of Munich), Matthias Nießner (Technical University of Munich)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes the ViewDiff method, which utilizes a pre-trained text-to-image diffusion model. By incorporating cross-frame attention and projection layers into the U-Net, it achieves the generation of high-quality 3D object images with consistent multi-view perspectives and realistic backgrounds in a single forward pass, while also supporting autoregressive generation of any viewpoint.
ViewFusion: Towards Multi-View Consistency via Interpolated Denoising
Xianghui Yang (Amazon), Anton van den Hengel (University of Adelaide)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: We propose ViewFusion, a training-free autoregressive method that can be directly embedded into existing single-view diffusion models to generate multi-view consistent images.
Viewpoint-Aware Visual Grounding in 3D Scenes
Xiangxi Shi (Oregon State University), Stefan Lee (Oregon State University)
RecognitionObject DetectionTransformerTextPoint Cloud
🎯 What it does: A perspective-aware 3D visual localization model, VPP-Net, is proposed, which can automatically predict the speaker's perspective based on the pointing expression and perform scene transformations to eliminate spatial relationship ambiguities.
ViLa-MIL: Dual-scale Vision-Language Multiple Instance Learning for Whole Slide Image Classification
Jiangbo Shi (Xi'an Jiaotong University), Huazhu Fu (Institute of High Performance Computing)
ClassificationTransformerLarge Language ModelVision Language ModelImageBiomedical Data
🎯 What it does: A dual-scale visual-language multi-instance learning framework, ViLa-MIL, is proposed for whole slide image (WSI) classification.
VILA: On Pre-training for Visual Language Models
Ji Lin (NVIDIA), Song Han (NVIDIA)
Representation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a new visual language model (VLM) pre-training scheme—VILA, systematically exploring the key designs for visual language pre-training on LLMs, and achieving stronger multimodal reasoning and instruction-following capabilities based on this.
VINECS: Video-based Neural Character Skinning
Zhouyingcheng Liao (University of Hong Kong), Christian Theobalt (Max Planck Institute for Informatics)
GenerationPose EstimationNeural Radiance FieldVideoMesh
🎯 What it does: An end-to-end trainable method is proposed that automatically generates fully bound and skinned 3D characters with pose-related skin weights using only multi-view videos.
ViP-LLaVA: Making Large Multimodal Models Understand Arbitrary Visual Prompts
Mu Cai (University of Wisconsin-Madison), Yong Jae Lee (University of Wisconsin-Madison)
RecognitionSegmentationGenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodalityBenchmark
🎯 What it does: A large multimodal model ViP-LLaVA is proposed, which can achieve region-level understanding and dialogue by overlaying arbitrary visual prompts (such as arrows, borders, hand-drawn elements, etc.) on images;
Virtual Immunohistochemistry Staining for Histological Images Assisted by Weakly-supervised Learning
Jiahan Li (Harbin Institute of Technology), Yongbing Zhang (Harbin Institute of Technology)
Image TranslationGenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A weakly supervised learning-based unpaired virtual immunohistochemistry (IHC) staining framework (Confusion-GAN) is proposed, capable of generating high-quality, pathologically consistent IHC images without the need for H&E-IHC aligned images.
Vision-and-Language Navigation via Causal Learning
Liuyi Wang (Tongji University), Qijun Chen (Tongji University)
TransformerContrastive LearningMultimodality
🎯 What it does: A cross-modal navigation framework called GOAT is proposed, which enhances the generalization ability of the VLN system by adjusting observable and unobservable confounding factors through causal learning.
VISTA-LLAMA: Reducing Hallucination in Video Language Models via Equal Distance to Visual Tokens
Fan Ma (Zhejiang University), Yi Yang (Zhejiang University)
RecognitionGenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoText
🎯 What it does: VISTA-LLAMA is proposed, a framework that reduces the hallucination of video language models by maintaining an equal distance between visual tokens and all language tokens, and introduces a sequential visual projector for temporal modeling.
Visual Anagrams: Generating Multi-View Optical Illusions with Diffusion Models
Daniel Geng (University of Michigan), Andrew Owens (University of Michigan)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a zero-shot method using a pre-trained diffusion model to generate multi-view optical illusion images that change appearance under different perspectives (such as flipping, rotation, pixel permutation, etc.).
Visual Concept Connectome (VCC): Open World Concept Discovery and their Interlayer Connections in Deep Models
Matthew Kowal (York University), Konstantinos G. Derpanis (York University)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkTransformerImage
🎯 What it does: A Visual Concept Connectivity (VCC) framework is proposed, which can automatically discover interpretable concepts at different layers of deep networks and their inter-layer connections without relying on labels.
Visual Delta Generator with Large Multi-modal Models for Semi-supervised Composed Image Retrieval
Young Kyun Jang (Meta AI), Ser-Nam Lim
Data SynthesisRetrievalTransformerLarge Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a visual delta generator (VDG) based on large language models, which is used to automatically synthesize the textual differences (visual delta) between query images and target images, thereby generating pseudo-triplet enhanced semi-supervised composite image retrieval (CIR) models;
Visual Fact Checker: Enabling High-Fidelity Detailed Caption Generation
Yunhao Ge (NVIDIA), Yin Cui (NVIDIA)
Object DetectionGenerationTransformerLarge Language ModelVision Language ModelDiffusion modelImageMultimodalityPoint Cloud
🎯 What it does: This paper presents VisualFactChecker (VFC) — a training-free multimodal pipeline that combines multimodal image-text models, LLMs, object detection, and VQA tools to achieve high-fidelity detail descriptions of 2D images and 3D objects.
Visual In-Context Prompting
Feng Li (Hong Kong University of Science and Technology), Jianfeng Gao (Microsoft Research)
SegmentationTransformerPrompt EngineeringImageVideo
🎯 What it does: DINOv is proposed, a unified visual context prompting framework that can simultaneously perform referential segmentation and general segmentation tasks.
Visual Layout Composer: Image-Vector Dual Diffusion Model for Design Layout Generation
Mohammad Amin Shabani (Simon Fraser University), Yasutaka Furukawa (Adobe Research)
GenerationTransformerDiffusion modelImage
🎯 What it does: A dual-domain diffusion model for images and vectors is proposed, which simultaneously generates high-quality images and editable vector frameworks during layout generation.
Visual Objectification in Films: Towards a New AI Task for Video Interpretation
Julie Tores (Universite Cote dAzur), Sarah Lecossais (Universite Sorbonne Paris Nord)
RecognitionObject DetectionTransformerVision Language ModelVideo
🎯 What it does: Introduced a visual task for detecting objectification of characters in films and constructed the ObyGaze12 dataset.
Visual Point Cloud Forecasting enables Scalable Autonomous Driving
Zetong Yang (OpenDriveLab), Hongyang Li (OpenDriveLab)
Autonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: A visual point cloud prediction pre-training framework called ViDAR is proposed, using future point cloud prediction as a pre-training task to enhance the performance of vision-driven perception, prediction, and planning.
Visual Program Distillation: Distilling Tools and Programmatic Reasoning into Vision-Language Models
Yushi Hu (Google Research), Ariel Fuxman (Google Research)
Object DetectionDepth EstimationKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextMultimodalityChain-of-Thought
🎯 What it does: This paper proposes Visual Program Distillation (VPD), which generates executable multi-tool programs through LLM and transforms the program execution trajectory into natural language Chain-of-Thought, thereby injecting visual reasoning and tool skills into VLM in a stepwise manner.
Visual Programming for Zero-shot Open-Vocabulary 3D Visual Grounding
Zhihao Yuan (Chinese University of Hong Kong Shenzhen), Zhen Li (Chinese University of Hong Kong Shenzhen)
Object DetectionSegmentationTransformerLarge Language ModelVision Language ModelImagePoint Cloud
🎯 What it does: This paper proposes a zero-shot open vocabulary 3D visual localization method, which achieves the localization of target objects in 3D scenes by converting natural language descriptions into executable visual programs.
Visual Prompting for Generalized Few-shot Segmentation: A Multi-scale Approach
Mir Rayat Imtiaz Hossain (University of British Columbia), James J. Little (University of British Columbia)
SegmentationTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes to regulate the multi-scale Transformer decoder through learning visual prompts, achieving end-to-end inference for the Generalized Few-Shot Segmentation (GFSS) task.
Visual-Augmented Dynamic Semantic Prototype for Generative Zero-Shot Learning
Wenjin Hou (Huazhong University of Science and Technology), Xinge You (Huazhong University of Science and Technology)
GenerationData SynthesisTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes a Visual-Augmented Dynamic Semantic Prototype (VADS) method to improve feature synthesis in generative zero-shot learning.
ViT-CoMer: Vision Transformer with Convolutional Multi-scale Feature Interaction for Dense Predictions
Chunlong Xia (Baidu Inc), Yifeng Shi (Baidu Inc)
Object DetectionSegmentationTransformerImageMultimodality
🎯 What it does: Proposes the ViT-CoMer basic structure, which integrates pure Vision Transformer and multi-scale CNN features to achieve efficient feature interaction for dense prediction tasks;
ViT-Lens: Towards Omni-modal Representations
Weixian Lei (National University of Singapore), Mike Zheng Shou (National University of Singapore)
ClassificationRetrievalRepresentation LearningTransformerContrastive LearningImageMultimodalityAudio
🎯 What it does: This paper proposes VIT-LENS, an omni-modal representation learning framework that utilizes a pre-trained ViT to project any new modality into the visual space through a Lens module and align it with foundational models like CLIP.
ViTamin: Designing Scalable Vision Models in the Vision-Language Era
Jieneng Chen (Johns Hopkins University), Liang-Chieh Chen (ByteDance)
ClassificationObject DetectionSegmentationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a scalable visual model called ViTamin in the era of visual-language models and establishes an evaluation protocol based on CLIP.
ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models
Jeong-gi Kwak (University of British Columbia), Kwang Moo Yi (University of British Columbia)
GenerationData SynthesisDiffusion modelOptical FlowImageVideo
🎯 What it does: A new perspective synthesis method that does not require training is proposed, treating the perspective generation of a single image as a short video where the camera scans along a smooth trajectory, and utilizing the noise fusion of a pre-trained video diffusion model and a perspective diffusion model for denoising.
VkD: Improving Knowledge Distillation using Orthogonal Projections
Roy Miles (Huawei), Jiankang Deng (Huawei)
ClassificationObject DetectionGenerationKnowledge DistillationTransformerImage
🎯 What it does: This study proposes an orthogonal projection-based knowledge distillation method that can directly project features while ensuring the invariance of intra-batch feature similarity, thereby maximizing knowledge transfer, applicable to classification, detection, and generation tasks.
Vlogger: Make Your Dream A Vlog
Shaobin Zhuang (Shanghai Jiao Tong University), Yali Wang (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences)
GenerationData SynthesisLarge Language ModelDiffusion modelVideoText
🎯 What it does: A general AI system named Vlogger is proposed, capable of automatically generating minute-long video logs (vlogs) based on user text descriptions, using LLMs (such as GPT-4) as directors for the four-stage process of scriptwriting, actor design, segment generation, and voiceover.
VLP: Vision Language Planning for Autonomous Driving
Chenbin Pan (Syracuse University), Liu Ren (Bosch Research North America)
Autonomous DrivingTransformerLarge Language ModelVision Language ModelContrastive LearningMultimodality
🎯 What it does: This paper proposes a Vision-Language Planning (VLP) framework that integrates language models (LM) into the BEV memory and planning stages of autonomous driving systems (ADS), achieving a unified end-to-end perception, prediction, and planning.
VMC: Video Motion Customization using Temporal Attention Adaption for Text-to-Video Diffusion Models
Hyeonho Jeong (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisOptimizationKnowledge DistillationDiffusion modelVideo
🎯 What it does: A Video Motion Customization (VMC) framework is proposed, which utilizes a single target video to transfer motion patterns to a new video with different backgrounds or appearances through one-time fine-tuning of only the temporal attention layer.
VMINer: Versatile Multi-view Inverse Rendering with Near- and Far-field Light Sources
Fan Fei (Peking University), Boxin Shi (Peking University)
RestorationGenerationOptimizationNeural Radiance FieldImageMesh
🎯 What it does: A multi-view inverse rendering framework VMiner is proposed, which can utilize both near-field and far-field lighting to reconstruct 3D geometry, materials, and generate texture triangle meshes that can be directly imported into industrial rendering under different lighting conditions.
VoCo: A Simple-yet-Effective Volume Contrastive Learning Framework for 3D Medical Image Analysis
Linshan Wu (Hong Kong University of Science and Technology), Hao Chen (Hong Kong University of Science and Technology)
SegmentationRepresentation LearningConvolutional Neural NetworkContrastive LearningImageBiomedical DataComputed Tomography
🎯 What it does: This paper proposes the VoCo framework for self-supervised pre-training of 3D medical images, predicting the contextual position of sub-volumes based on volume contrast learning.
Volumetric Environment Representation for Vision-Language Navigation
Rui Liu (Zhejiang University), Yi Yang (Zhejiang University)
Object DetectionRepresentation LearningTransformerReinforcement LearningPoint Cloud
🎯 What it does: This paper proposes a voxelized 3D environment representation (VER) by projecting multi-view 2D features into a 3D voxel grid, and performing coarse-to-fine feature extraction and multi-task learning in this space to achieve joint predictions of 3D occupancy, room layout, and 3D detection, thereby providing a more complete scene understanding for Visual Language Navigation (VLN);
VOODOO 3D: Volumetric Portrait Disentanglement For One-Shot 3D Head Reenactment
Phong Tran (Mohammed bin Zayed University of Artificial Intelligence), Hao Li (Pinscreen)
GenerationData SynthesisConvolutional Neural NetworkTransformerNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: A single-image 3D head reproduction technology based on a fully convolutional volumetric three-plane representation is proposed, achieving high-fidelity separation and real-time synthesis of expressions between the source image and the driving image.
VP3D: Unleashing 2D Visual Prompt for Text-to-3D Generation
Yang Chen (HiDream.ai Inc.), Tao Mei (HiDream.ai Inc.)
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldImageText
🎯 What it does: This paper proposes a visual prompt-guided text-to-3D generation framework called VP3D, which first generates high-quality images as visual prompts using a 2D diffusion model, and then combines Score Distillation Sampling (SDS) with additional visual consistency and human feedback rewards to achieve zero-shot generation of high-quality 3D models from text.
VRetouchEr: Learning Cross-frame Feature Interdependence with Imperfection Flow for Face Retouching in Videos
Wen Xue (South China University of Technology), Hau San Wong (City University of Hong Kong)
Image TranslationRestorationTransformerGenerative Adversarial NetworkOptical FlowVideo
🎯 What it does: A video-oriented defect removal framework called VRetouchEr is designed, utilizing defect flow estimation and multi-frame mask attention to achieve stable and high-quality removal of facial defects in videos.
VRP-SAM: SAM with Visual Reference Prompt
Yanpeng Sun (Nanjing University of Science and Technology), Zechao Li (Nanjing University of Science and Technology)
SegmentationMeta LearningConvolutional Neural NetworkPrompt EngineeringImage
🎯 What it does: This paper proposes a Visual Reference Prompt (VRP) encoder, which is integrated with the Segment Anything Model (SAM) to form VRP-SAM, thereby supporting semantic segmentation of target images directly using annotated reference images (points, boxes, lines, masks).
VS: Reconstructing Clothed 3D Human from Single Image via Vertex Shift
Leyuan Liu (National Engineering Research Center for E-Learning Central China Normal University), Jingying Chen (National Engineering Research Center for E-Learning Central China Normal University)
GenerationPose EstimationGraph Neural NetworkImageMesh
🎯 What it does: A method for 3D human body reconstruction from a single image based on a two-stage vertex shift has been proposed, which achieves high-fidelity and defect-free 3D reconstruction of humans wearing loose clothing while maintaining the structure of the human body.
VSCode: General Visual Salient and Camouflaged Object Detection with 2D Prompt Learning
Ziyang Luo (Northwestern Polytechnical University), Junwei Han (Institute of Artificial Intelligence, Hefei Comprehensive National Science Center)
Object DetectionTransformerPrompt EngineeringImageMultimodality
🎯 What it does: A general visual salient object and camouflage object detection framework, VSCode, is proposed, capable of handling multi-modal SOD and COD tasks in one go.
VSRD: Instance-Aware Volumetric Silhouette Rendering for Weakly Supervised 3D Object Detection
Zihua Liu (Tokyo Institute of Technology), Masatoshi Okutomi (Tokyo Institute of Technology)
Object DetectionAutonomous DrivingImage
🎯 What it does: A weakly supervised 3D object detection framework VSRD based on multi-view automatic labeling is proposed, which optimizes 3D bounding boxes using 2D instance masks and generates pseudo-labels to train a monocular 3D detector.
VTimeLLM: Empower LLM to Grasp Video Moments
Bin Huang (Tsinghua University), Wenwu Zhu (Tsinghua University)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoText
🎯 What it does: This paper proposes VTimeLLM, a large language model (LLM) with video temporal boundary awareness, capable of accurately locating the start and end times of video events and generating descriptions.
VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning
Kang Chen (Harbin Institute of Technology), Xiangqian Wu (Harbin Institute of Technology)
RecognitionObject DetectionData SynthesisRecurrent Neural NetworkTransformerVision Language ModelImageTextMultimodality
🎯 What it does: The VTQA dataset is proposed, which combines real images and long texts for multi-hop cross-media question answering, and implements the KECMRN model based on key entity extraction and cross-media reasoning.
WALT3D: Generating Realistic Training Data from Time-Lapse Imagery for Reconstructing Dynamic Objects Under Occlusion
Khiem Vuong (Carnegie Mellon University), Srinivasa G. Narasimhan (Carnegie Mellon University)
Object DetectionObject TrackingSegmentationData SynthesisPose EstimationAutonomous DrivingConvolutional Neural NetworkTransformerGenerative Adversarial NetworkImageVideo
🎯 What it does: By automatically mining unobstructed vehicle and pedestrian instances from the sequential images of static cameras, pseudo-real annotations are generated using existing 2D and 3D prediction methods. These instances are then composited back into the original image in a physically consistent 3D clipping manner, resulting in a synthetic training set with complete 2D/3D occlusion information.
WANDR: Intention-guided Human Motion Generation
Markos Diomataris (Max Planck Institute for Intelligent Systems), Michael J. Black (Max Planck Institute for Intelligent Systems)
GenerationRobotic IntelligenceAuto EncoderSequential
🎯 What it does: This paper presents WANDR, an autoregressive conditional variational autoencoder that generates natural full-body movements based on the initial pose and 3D target position, allowing the wrist to accurately reach the target.
WateRF: Robust Watermarks in Radiance Fields for Protection of Copyrights
Youngdong Jang (Korea University), Sangpil Kim (Google Research)
GenerationData SynthesisSafty and PrivacyNeural Radiance FieldImage
🎯 What it does: This paper proposes a method to embed a robust watermark that can be extracted from rendered images in both implicit and explicit NeRF models.
Watermark-embedded Adversarial Examples for Copyright Protection against Diffusion Models
Peifei Zhu (LY Corporation), Hirokatsu Kataoka (LY Corporation)
GenerationAdversarial AttackDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Generate adversarial samples with embedded visible watermarks through training conditional generative adversarial networks to prevent diffusion models from copying unauthorized images.
WaveFace: Authentic Face Restoration with Efficient Frequency Recovery
Yunqi Miao (University of Warwick), Jungong Han (University of Sheffield)
RestorationDiffusion modelImage
🎯 What it does: The WaveFace method is proposed, transferring the blind face restoration task to the frequency domain, utilizing discrete wavelet decomposition to separately restore low-frequency and high-frequency components.
Wavelet-based Fourier Information Interaction with Frequency Diffusion Adjustment for Underwater Image Restoration
Chen Zhao (Nanjing Normal University), Chengwei Hu (Nanjing Normal University)
RestorationTransformerDiffusion modelImage
🎯 What it does: This paper proposes a seabed image enhancement framework WF-Diff based on frequency domain features and diffusion models, which can first repair color distortion in the frequency domain and then refine details using a frequency domain residual diffusion model.
WaveMo: Learning Wavefront Modulations to See Through Scattering
Mingyang Xie (University of Maryland), Christopher A. Metzler (University of Maryland)
RestorationNeural Radiance FieldImageBiomedical Data
🎯 What it does: This paper proposes an end-to-end learning framework that jointly optimizes phase pre-modulation (wavefront modulation) and a proxy reconstruction network to improve imaging quality under scattering media and to enable the modulation scheme to generalize to unknown scattering conditions.
Weak-to-Strong 3D Object Detection with X-Ray Distillation
Alexander Gambashidze (Artificial Intelligence Research Institute), Ilya Makarov (Artificial Intelligence Research Institute)
Object DetectionAutonomous DrivingKnowledge DistillationPoint CloudTime Series
🎯 What it does: Proposes the X-Ray Teacher framework, which enhances 3D detection performance in sparse and occluded scenes by utilizing LiDAR time series to generate Object-Complete Frames and employing Teacher-Student knowledge distillation.
Weakly Misalignment-free Adaptive Feature Alignment for UAVs-based Multimodal Object Detection
Chen Chen (National University of Defense Technology), Ping Zhong (National University of Defense Technology)
Object DetectionConvolutional Neural NetworkImageMultimodality
🎯 What it does: The study focuses on RGB-IR multimodal object detection based on UAVs, proposing the OAFA method to achieve adaptive feature alignment and address the weak alignment issue.
Weakly Supervised Monocular 3D Detection with a Single-View Image
Xueying Jiang (Nanyang Technological University), Shijian Lu (Nanyang Technological University)
Object DetectionDepth EstimationAutonomous DrivingKnowledge DistillationImage
🎯 What it does: A weakly supervised monocular 3D detection framework SKD-WM3D is proposed, which can complete 3D box prediction using only pseudo-depth labels from a single image.
Weakly Supervised Point Cloud Semantic Segmentation via Artificial Oracle
Hyeokjun Kweon (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
SegmentationPoint Cloud
🎯 What it does: A weakly supervised point cloud semantic segmentation framework (REAL) is proposed, which combines artificial prior label generation based on SAM (Segment Anything Model) with active learning.
Weakly Supervised Video Individual Counting
Xinyan Liu (University of Chinese Academy of Science), Qingming Huang (University of Chinese Academy of Science)
Object DetectionObject TrackingConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: This paper proposes the Weakly Supervised Video Individual Counting (WVIC) task and presents a baseline model called CGNet, addressing the issue of expensive trajectory annotations required by traditional VIC.
Weakly-Supervised Audio-Visual Video Parsing with Prototype-based Pseudo-Labeling
Kranthi Kumar Rachavarapu (Indian Institute of Technology Madras), Rajagopalan A. N. (Indian Institute of Technology Madras)
RecognitionSegmentationOptimizationContrastive LearningVideoMultimodalityAudio
🎯 What it does: A prototype-based pseudo-labeling method is proposed for weakly supervised audio-video parsing (AVVP), significantly improving the temporal and modal localization of events through multi-prototype clustering and contrastive learning.
Weakly-Supervised Emotion Transition Learning for Diverse 3D Co-speech Gesture Generation
Xingqun Qi (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)
GenerationData SynthesisTransformerGenerative Adversarial NetworkMultimodalitySequentialAudio
🎯 What it does: This study investigates the generation of 3D co-speech gestures for emotion transfer under weak supervision, constructing two new emotion transfer datasets and proposing a corresponding generation framework.
WHAM: Reconstructing World-grounded Humans with Accurate 3D Motion
Soyong Shin (Max Planck Institute for Intelligent Systems), Michael J. Black (Max Planck Institute for Intelligent Systems)
Pose EstimationRecurrent Neural NetworkVideo
🎯 What it does: This paper presents WHAM, which can accurately and in real-time recover full-body 3D posture, shape, and its trajectory in the world coordinate system from monocular videos captured by a moving camera.
What Do You See in Vehicle? Comprehensive Vision Solution for In-Vehicle Gaze Estimation
Yihua Cheng (University of Birmingham), Hyung Jin Chang (Huazhong University of Science and Technology)
ClassificationPose EstimationAutonomous DrivingTransformerImage
🎯 What it does: The first in-car gaze dataset, IVGaze, is proposed, along with a visual-based gaze collection and annotation method. Subsequently, a dual-stream gaze pyramid transformer (GazeDPTR) is developed for gaze estimation and extended to gaze region classification tasks.
What How and When Should Object Detectors Update in Continually Changing Test Domains?
Jayeon Yoo (Seoul National University), Nojun Kwak (Seoul National University)
Object DetectionDomain AdaptationComputational EfficiencyImage
🎯 What it does: A Continuous Testing Adaptation (CTA) method for object detection is proposed, achieving online adaptation in a constantly changing testing domain.
What If the TV Was Off? Examining Counterfactual Reasoning Abilities of Multi-modal Language Models
Letian Zhang (Tongji University), Bingchen Zhao (University of Edinburgh)
Data SynthesisTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: A C-VQA dataset is proposed and constructed, specifically to evaluate the counterfactual reasoning ability of multimodal large language models;
What Sketch Explainability Really Means for Downstream Tasks?
Hmrishav Bandyopadhyay (University of Surrey), Yi-Zhe Song (University of Surrey)
GenerationRetrievalExplainability and InterpretabilityAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a lightweight, pluggable interpretability framework that implements stroke-level (SLA) and partial stroke-level (P-SLA) attribution methods for hand-drawn sketches, addressing the non-differentiable issues of traditional rasterization, and seamlessly integrating this attribution tool into various downstream tasks (retrieval, generation, assisted drawing, adversarial attacks).
What When and Where? Self-Supervised Spatio-Temporal Grounding in Untrimmed Multi-Action Videos from Narrated Instructions
Brian Chen (Columbia University), Hilde Kuehne (University of Bonn)
RecognitionObject DetectionTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: A framework based on weakly supervised multimodal (video + ASR subtitles) learning is proposed to achieve unsupervised multi-action video spatiotemporal localization.
What You See is What You GAN: Rendering Every Pixel for High-Fidelity Geometry in 3D GANs
Alex Trevithick (University of California, San Diego), Koki Nagano (NVIDIA)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A new 3D GAN training framework is studied, capable of rendering high-detail 3D generative models at native 2D resolution without the need for multi-view or 3D data.
When StyleGAN Meets Stable Diffusion: a W+ Adapter for Personalized Image Generation
Xiaoming Li (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a W+ adapter that aligns the W+ space of StyleGAN with Stable Diffusion, enabling personalized text-to-image generation from a single reference image and supporting editable facial attributes.
When Visual Grounding Meets Gigapixel-level Large-scale Scenes: Benchmark and Approach
Tao Ma (Tsinghua University), Lu Fang (Peking University)
RecognitionObject DetectionTransformerImageVideoBenchmark
🎯 What it does: This paper proposes the GigGrounding dataset and the GlaZing model, focusing on visual localization in large scenes at the pixel level.
Why Not Use Your Textbook? Knowledge-Enhanced Procedure Planning of Instructional Videos
Kumaranage Ravindu Yasas Nagasinghe (Mohamed bin Zayed University of Artificial Intelligence), Muhammad Haris Khan (Mohamed bin Zayed University of Artificial Intelligence)
Diffusion modelVideo
🎯 What it does: This paper proposes a knowledge-enhanced program planning framework called KEPP, which utilizes a probabilistic program knowledge graph generated from the training set to assist in planning action sequences from the initial visual state to the target visual state.
WildlifeMapper: Aerial Image Analysis for Multi-Species Detection and Identification
Satish Kumar (University of California Santa Barbara), B.S. Manjunath (University of California Santa Barbara)
RecognitionObject DetectionTransformerImage
🎯 What it does: WildlifeMapper (WM) has been developed, an end-to-end model based on Transformer for detecting, locating, and identifying multiple wildlife species in high-resolution aerial images. A large-scale annotated dataset of the Masai Mara ecosystem has been publicly released, containing 21 species and 28k target boxes.
WinSyn: : A High Resolution Testbed for Synthetic Data
Tom Kelly (King Abdullah University of Science and Technology), Peter Wonka (King Abdullah University of Science and Technology)
SegmentationData SynthesisDomain AdaptationImageBenchmark
🎯 What it does: A high-resolution window image dataset WinSyn has been constructed (75,739 4K–6K photos, 9,002 with segmentation annotations) along with an adjustable procedural model based on CGA/shape grammar, generating 21,290 synthetic window images for semantic segmentation benchmarking.
Wired Perspectives: Multi-View Wire Art Embraces Generative AI
Zhiyu Qu (University of Surrey), Yi-Zhe Song (University of Surrey)
GenerationData SynthesisKnowledge DistillationDiffusion modelScore-based ModelImageText
🎯 What it does: This paper proposes a DreamWire system based on diffusion models, allowing users to generate multi-view wire sculptures (MVWA) for each perspective through text or sketch prompts.
Wonder3D: Single Image to 3D using Cross-Domain Diffusion
Xiaoxiao Long (University of Hong Kong), Wenping Wang
GenerationData SynthesisDiffusion modelImageMesh
🎯 What it does: Generate high-quality texture meshes from a single image, using a cross-domain diffusion model to produce multi-view normal maps and color maps, and then quickly recover 3D shapes through geometric-aware normal fusion.
WonderJourney: Going from Anywhere to Everywhere
Hong-Xing Yu, Charles Herrmann
SegmentationGenerationDepth EstimationTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextPoint Cloud
🎯 What it does: This work proposes WonderJourney, a framework that can generate long, diverse, and coherent 3D scenes from any text or image starting point, achieving a journey 'from anywhere to anywhere';
WorDepth: Variational Language Prior for Monocular Depth Estimation
Ziyao Zeng (Yale University), Alex Wong (Yale University)
Depth EstimationTransformerAuto EncoderImageText
🎯 What it does: Utilizing language descriptions as a variational prior, the scale accuracy of monocular depth estimation is improved through the joint learning of a variational autoencoder (VAE) and an image-based conditional sampler.
WOUAF: Weight Modulation for User Attribution and Fingerprinting in Text-to-Image Diffusion Models
Changhoon Kim (Arizona State University), Yezhou Yang (Arizona State University)
GenerationData SynthesisSafty and PrivacyConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A user fingerprint embedding method based on weight modulation, WOUAF, is proposed, which can quickly generate user-specific fingerprints on the Stable Diffusion decoder, enabling user attribution for the model distributor.
Would Deep Generative Models Amplify Bias in Future Models?
Tianwei Chen (Osaka University), Yuta Nakashima (CyberAgent Inc.)
GenerationData SynthesisRetrievalTransformerVision Language ModelDiffusion modelImage
🎯 What it does: This paper gradually replaces the original images in COCO and CC3M with images generated by Stable Diffusion, training OpenCLIP and image description models to evaluate the amplification or mitigation of biases in dimensions such as gender, race, age, and skin color.
WWW: A Unified Framework for Explaining What Where and Why of Neural Networks by Interpretation of Neuron Concepts
Yong Hyun Ahn (Kyung Hee University), Seong Tae Kim (Kyung Hee University)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A unified framework called WWW is proposed, which can simultaneously explain the three layers of meaning of a model: 'what', 'where', and 'why'.
X-3D: Explicit 3D Structure Modeling for Point Cloud Recognition
Shuofeng Sun (Beijing University of Posts and Telecommunications), Haibin Yan (Tsinghua University)
ClassificationObject DetectionSegmentationPoint Cloud
🎯 What it does: This study proposes X-3D, which significantly improves point cloud classification, segmentation, and detection performance by explicitly constructing local geometric structures in the original space and generating shared dynamic structural kernels based on this.
X-Adapter: Adding Universal Compatibility of Plugins for Upgraded Diffusion Model
Lingmin Ran (Show Lab), Mike Zheng Shou (National University of Singapore)
GenerationDiffusion modelImageStochastic Differential Equation
🎯 What it does: This paper proposes X-Adapter, a universal adapter that allows plugins for training legacy diffusion models to be used directly on upgraded models without the need for retraining.
X-MIC: Cross-Modal Instance Conditioning for Egocentric Action Generalization
Anna Kukleva (Meta Reality Labs), Shugao Ma (Meta Reality Labs)
ClassificationRecognitionTransformerPrompt EngineeringVideoText
🎯 What it does: This study investigates action recognition across datasets in first-person perspective videos and proposes the X-MIC framework for cross-modal instance conditioning in a frozen CLIP embedding space.
XCube: Large-Scale 3D Generative Modeling using Sparse Voxel Hierarchies
Xuanchi Ren (NVIDIA), Francis Williams (University of Toronto)
GenerationData SynthesisAutonomous DrivingDiffusion modelAuto EncoderPoint CloudMesh
🎯 What it does: This paper proposes a high-resolution sparse 3D voxel hierarchy generation model called XCube, which can generate millions of voxels within 30 seconds and provide attributes such as semantics, normals, and TSDF.
XFeat: Accelerated Features for Lightweight Image Matching
Guilherme Potje (Universidade Federal de Minas Gerais), Erickson R. Nascimento (Microsoft)
Computational EfficiencyKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A lightweight CNN architecture called XFeat is proposed, achieving sparse and semi-dense image matching while balancing speed and accuracy.
XFibrosis: Explicit Vessel-Fiber Modeling for Fibrosis Staging from Liver Pathology Images
Chong Yin (Hong Kong Baptist University), Pong C. Yuen (Chinese University of Hong Kong)
ClassificationSegmentationExplainability and InterpretabilityGraph Neural NetworkImageBiomedical Data
🎯 What it does: This study proposes the XFibrosis method, which explicitly transforms the blood vessels and fibers in liver biopsy images into a primal-dual graph structure, and achieves fibrosis staging through dual graph convolution.
XScale-NVS: Cross-Scale Novel View Synthesis with Hash Featurized Manifold
Guangyu Wang (Tsinghua University), Lu Fang (Tsinghua University)
GenerationData SynthesisNeural Radiance FieldGaussian SplattingImageBenchmark
🎯 What it does: A surface manifold representation based on hash coding is proposed, combined with delayed neural rendering, to achieve high-fidelity new view synthesis across scales.
YOLO-World: Real-Time Open-Vocabulary Object Detection
Tianheng Cheng (Tencent), Ying Shan (Tencent)
Object DetectionConvolutional Neural NetworkPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Developed YOLO-World, a real-time open vocabulary object detection framework that combines Vision-Language pre-training with the re-parameterized RepVL-PAN to achieve zero-shot open vocabulary detection.
YolOOD: Utilizing Object Detection Concepts for Multi-Label Out-of-Distribution Detection
Alon Zolfi (Ben Gurion University of the Negev), Asaf Shabtai (Ben Gurion University of the Negev)
Object DetectionAnomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes YolOOD, which utilizes the 'objectness' and classification scores of the YOLO object detection model to achieve out-of-distribution (OOD) detection for multi-label images.
You Only Need Less Attention at Each Stage in Vision Transformers
Shuoxi Zhang (Huazhong University of Science and Technology), Kun He (Microsoft Research Asia)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a new Vision Transformer structure—Less-Attention Vision Transformer (LaViT), which significantly reduces the computational complexity of self-attention and alleviates the attention saturation problem by calculating complete self-attention only in a few layers at each stage and then reusing the saved attention matrices through linear transformations.