ICCV 2023 Papers — Page 21
IEEE/CVF International Conference on Computer Vision · 2156 papers
Understanding the Feature Norm for Out-of-Distribution Detection
Jaewoo Park (Yonsei University), Andrew Beng Jin Teoh (Yonsei University)
Anomaly DetectionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This study clarifies that the norm of the hidden layer feature vector is essentially equivalent to the confidence of the hidden classifier's maximum logit, revealing its ability to distinguish between ID and OOD in a class-agnostic manner. Based on this, a Negative Awareness Norm (NAN) is proposed to simultaneously capture activation and deactivation information, enhancing OOD detection.
Unfolding Framework with Prior of Convolution-Transformer Mixture and Uncertainty Estimation for Video Snapshot Compressive Imaging
Siming Zheng (Computer Network Information Center Chinese Academy of Science), Xin Yuan (Westlake University)
RestorationCompressionConvolutional Neural NetworkTransformerVideo
🎯 What it does: A Transformer + 3D-CNN deep unfolding network is proposed for the reconstruction of video snapshot compression imaging.
Uni-3D: A Universal Model for Panoptic 3D Scene Reconstruction
Xiang Zhang (University of California San Diego), Zhuowen Tu (University of California San Diego)
SegmentationDepth EstimationTransformerImagePoint Cloud
🎯 What it does: We propose Uni-3D, a unified model for complete 3D scene parsing and reconstruction from a single RGB image, capable of simultaneously outputting 3D semantic segmentation and geometric reconstruction of object instances and scene layout.
UniDexGrasp++: Improving Dexterous Grasping Policy Learning via Geometry-Aware Curriculum and Iterative Generalist-Specialist Learning
Weikang Wan (Peking University), He Wang (Peking University)
Robotic IntelligenceTransformerReinforcement LearningAgentic AIPoint CloudBenchmark
🎯 What it does: This paper proposes UniDexGrasp++, a general fingertip grasping strategy training pipeline based on geometric perception, curriculum learning, and iterative general-to-specific learning.
UniFace: Unified Cross-Entropy Loss for Deep Face Recognition
Jiancan Zhou (Shenzhen University), Jinming Duan (University of Birmingham)
RecognitionConvolutional Neural NetworkImage
🎯 What it does: A Unified Cross-Entropy (UCE) loss with a learnable unified threshold is designed, and the UniFace model is constructed for deep face recognition.
Unified Adversarial Patch for Cross-Modal Attacks in the Physical World
Xingxing Wei (Beihang University), Jie Yu (Beihang University)
Object DetectionAdversarial AttackImageMultimodality
🎯 What it does: A unified adversarial patch that works under both visible light and infrared perception modes is proposed, capable of interfering with two target detectors in the physical world at once;
Unified Coarse-to-Fine Alignment for Video-Text Retrieval
Ziyang Wang (University of North Carolina at Chapel Hill), Mohit Bansal (University of North Carolina at Chapel Hill)
RetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: A unified coarse-fine alignment model UCOFIA is proposed for video-text retrieval, accommodating multi-level alignments of video-sentence, frame-sentence, and patch-word.
Unified Data-Free Compression: Pruning and Quantization without Fine-Tuning
Shipeng Bai (Zhejiang University), Yong Liu (Zhejiang University)
CompressionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Unified Data-Free Compression framework (UDFC) that can simultaneously perform structured pruning and low-bit quantization on networks without using any data or fine-tuning.
Unified Out-Of-Distribution Detection: A Model-Specific Perspective
Reza Averly (Ohio State University), Wei-Lun Chao (Ohio State University)
Domain AdaptationAnomaly DetectionTransformerImage
🎯 What it does: A unified model-specific OOD detection framework MS-OOD is proposed, treating correctly predicted samples as acceptable and misclassified samples as needing rejection, covering ID, C-OOD, and S-OOD data.
Unified Pre-Training with Pseudo Texts for Text-To-Image Person Re-Identification
Zhiyin Shao (South China University of Technology), Jingdong Wang (Baidu VIS)
RetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: A unified pre-training pipeline called UniPT is proposed, which utilizes automatically generated pseudo-text to construct a large-scale image-text dataset LUPerson-T, addressing the issue of data and training inconsistency in text-to-image person retrieval.
Unified Visual Relationship Detection with Vision and Language Models
Long Zhao (Google Research), Ting Liu (Google Research)
Object DetectionTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Design and train a unified visual relationship detector (UniVRD) that can simultaneously perform object detection and relationship prediction in a joint label space across different datasets.
UniFormerV2: Unlocking the Potential of Image ViTs for Video Understanding
Kunchang Li (Shenzhen Institute of Advanced Technology), Yu Qiao (Shenzhen Institute of Advanced Technology)
RecognitionRepresentation LearningTransformerVideo
🎯 What it does: By inserting local temporal MHRA and global cross-attention units into the pre-trained image ViT, ViT is transformed into an efficient video understanding model called UniFormerV2.
UniFusion: Unified Multi-View Fusion Transformer for Spatial-Temporal Representation in Bird's-Eye-View
Zequn Qin (Zhejiang University), Xi Li (Zhejiang University)
SegmentationAutonomous DrivingTransformerImageVideo
🎯 What it does: This paper proposes UniFusion, a unified spatial-temporal fusion Transformer that combines multi-view spatial fusion and historical frame temporal fusion through virtual view mapping, achieving long-term, adaptive fusion in a bird's-eye view while maintaining information integrity.
Unify, Align and Refine: Multi-Level Semantic Alignment for Radiology Report Generation
Yaowei Li (Peking University), Yuexian Zou (Peking University)
GenerationTransformerAuto EncoderImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes the UAR framework, which achieves multi-layer cross-modal alignment to generate chest X-ray reports through three modules: Unification (Latent Space Unifier), Alignment (Cross-modal Representation Aligner), and Refinement (Text-to-Image Refiner).
UniKD: Universal Knowledge Distillation for Mimicking Homogeneous or Heterogeneous Object Detectors
Shanshan Lao (Tsinghua University), Yujiu Yang (Tsinghua University)
Object DetectionKnowledge DistillationImage
🎯 What it does: A universal knowledge distillation method named UniKD is proposed for transferring knowledge between homogeneous or heterogeneous object detectors.
Unilaterally Aggregated Contrastive Learning with Hierarchical Augmentation for Anomaly Detection
Guodong Wang (Beihang University), Di Huang (Beihang University)
Anomaly DetectionContrastive LearningImage
🎯 What it does: This paper proposes a contrastive learning-based anomaly detection framework called UniCon-HA, which focuses on normal samples while dispersing virtual anomalies through unidirectional aggregation.
UniSeg: A Unified Multi-Modal LiDAR Segmentation Network and the OpenPCSeg Codebase
Youquan Liu (Shanghai AI Laboratory), Yuenan Hou (Shanghai AI Laboratory)
SegmentationAutonomous DrivingMultimodalityPoint Cloud
🎯 What it does: This paper proposes UniSeg, a unified multi-modal LiDAR segmentation network that integrates voxel, range, and point views of RGB images and point clouds, achieving both semantic segmentation and panoptic segmentation.
UniT3D: A Unified Transformer for 3D Dense Captioning and Visual Grounding
Zhenyu Chen, Angel X. Chang (Simon Fraser University)
Object DetectionGenerationTransformerVision Language ModelTextPoint Cloud
🎯 What it does: This paper proposes UniT3D, a unified Transformer for 3D visual localization and dense description.
UnitedHuman: Harnessing Multi-Source Data for High-Resolution Human Generation
Jianglin Fu (Shanghai AI Laboratory), Ziwei Liu (Nanyang Technological University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Proposes the UnitedHuman framework, which utilizes multi-source spatial transformers and continuous GANs to map local part data to full-body space and generate high-quality full-body portraits at any resolution.
UniTR: A Unified and Efficient Multi-Modal Transformer for Bird's-Eye-View Representation
Haiyang Wang (Peking University), Liwei Wang (Peking University)
Object DetectionSegmentationAutonomous DrivingTransformerImageMultimodalityPoint Cloud
🎯 What it does: A unified multimodal Transformer backbone called UniTR is proposed, which can simultaneously and in parallel process camera images and LiDAR point clouds, and output a unified bird's-eye view representation for 3D detection and BEV segmentation.
Universal Domain Adaptation via Compressive Attention Matching
Didi Zhu (Zhejiang University), Chao Wu (Zhejiang University)
Domain AdaptationTransformerImage
🎯 What it does: The UniAM framework is proposed, utilizing the self-attention mechanism of visual Transformers and Compressed Attention Matching (CAM) to achieve universal domain adaptation.
UniverSeg: Universal Medical Image Segmentation
Victor Ion Butoi (Massachusetts Institute of Technology), Adrian V. Dalca (Massachusetts Institute of Technology)
SegmentationConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: A single universal medical image segmentation model named UniverSeg is proposed, which can directly complete new segmentation tasks using a support set (image-label pairs) without any training or fine-tuning.
UniVTG: Towards Unified Video-Language Temporal Grounding
Kevin Qinghong Lin (National University of Singapore), Mike Zheng Shou (National University of Singapore)
RecognitionRetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: A unified video-language temporal localization framework, UniVTG, is proposed, defining unified temporal labels (foreground, boundary, saliency) and developing a multimodal decoder that utilizes pseudo-labels for large-scale pre-training, supporting multiple tasks (moment retrieval, highlight detection, video summarization).
Unleashing Text-to-Image Diffusion Models for Visual Perception
Wenliang Zhao (Tsinghua University), Jiwen Lu (Tsinghua University)
SegmentationDepth EstimationDiffusion modelImage
🎯 What it does: The paper proposes the VPD framework, which utilizes a pre-trained text-to-image diffusion model for visual perception tasks.
Unleashing the Potential of Spiking Neural Networks with Dynamic Confidence
Chen Li (Manchester University), Steve Furber (Manchester University)
ClassificationComputational EfficiencySpiking Neural NetworkImage
🎯 What it does: A decision-making mechanism based on dynamic confidence is proposed, which determines when to terminate early during SNN inference based on the time-evolving confidence, reducing latency without sacrificing accuracy.
Unleashing the Power of Gradient Signal-to-Noise Ratio for Zero-Shot NAS
Zihao Sun (Chinese Academy of Sciences), Yu Hu (Chinese Academy of Sciences)
Neural Architecture Search
🎯 What it does: This paper proposes a new zero-shot neural architecture search (NAS) proxy called ξ-based Gradient Signal-to-Noise Ratio (ξ-GSNR) for predicting network accuracy at initialization without any training.
Unleashing Vanilla Vision Transformer with Masked Image Modeling for Object Detection
Yuxin Fang (Huazhong University of Science and Technology), Xinggang Wang (Huazhong University of Science and Technology)
Object DetectionTransformerImage
🎯 What it does: Proposes an efficient adaptation method for vanilla Vision Transformer (ViT) in object detection tasks based on Masked Image Modeling pre-training.
UnLoc: A Unified Framework for Video Localization Tasks
Shen Yan (Google Research), Cordelia Schmid (Google Research)
RecognitionSegmentationRetrievalTransformerVision Language ModelContrastive LearningVideoText
🎯 What it does: A unified one-stage framework called UnLoc is designed, utilizing the CLIP dual-tower pre-trained model and a video-text fusion module to construct a feature pyramid for instant retrieval, temporal action localization, and action segmentation tasks.
Unmasked Teacher: Towards Training-Efficient Video Foundation Models
Kunchang Li (Shenzhen Institute of Advanced Technology), Yu Qiao (Shenzhen Institute of Advanced Technology)
RetrievalComputational EfficiencyKnowledge DistillationTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes a method that utilizes an image foundation model (CLIP) as an unmasked teacher. By aligning unmasked visual tokens with teacher features and combining semantic masking and spatiotemporal attention, it achieves efficient unmasked video pre-training. Subsequently, cross-modal tasks are incorporated through progressive pre-training, ultimately resulting in a video foundation model capable of simultaneously handling scene, temporal, and video-text tasks.
Unmasking Anomalies in Road-Scene Segmentation
Shyam Nandan Rai (Politecnico di Torino), Barbara Caputo (Politecnico di Torino)
SegmentationAnomaly DetectionAutonomous DrivingTransformerContrastive LearningImage
🎯 What it does: The anomaly segmentation task is transformed from pixel-wise classification to mask-based classification, proposing the Mask2Anomaly framework for anomaly detection.
Unpaired Multi-domain Attribute Translation of 3D Facial Shapes with a Square and Symmetric Geometric Map
Zhenfeng Fan (Institute of Computing Technology Chinese Academy of Sciences), Shihong Xia (Soochow University)
Image TranslationGenerationData SynthesisGenerative Adversarial NetworkMesh
🎯 What it does: A framework for 3D facial shape attribute translation based on unpaired multi-domain is proposed, capable of generating high-quality 3D facial models with different expressions, ages, and genders in one go.
Unsupervised 3D Perception with 2D Vision-Language Distillation for Autonomous Driving
Mahyar Najibi (Waymo), Dragomir Anguelov (Waymo)
Object DetectionObject TrackingAutonomous DrivingKnowledge DistillationVision Language ModelMultimodalityPoint Cloud
🎯 What it does: A multimodal automatic labeling pipeline has been designed and implemented, utilizing LiDAR motion information and a pre-trained vision-language model to generate open vocabulary 3D boxes and trajectories for unlabeled data, and to train a 3D detector that can assign categories based on text queries during inference.
Unsupervised Accuracy Estimation of Deep Visual Models using Domain-Adaptive Adversarial Perturbation without Source Samples
JoonHo Lee (Samsung SDS), Kwonho Lee (Samsung SDS)
Domain AdaptationAdversarial AttackImage
🎯 What it does: A method for unsupervised estimation of the accuracy of deep visual models in the target domain under the condition of no source samples is proposed, called SF-DAP.
Unsupervised Compositional Concepts Discovery with Text-to-Image Generative Models
Nan Liu (University of Illinois at Urbana-Champaign), Antonio Torralba (Massachusetts Institute of Technology)
GenerationRepresentation LearningDiffusion modelImage
🎯 What it does: Using a pre-trained diffusion model and based on energy model explanations, an unsupervised method is proposed to automatically decompose combinable generative concepts from unlabeled images.
Unsupervised Domain Adaptation for Training Event-Based Networks Using Contrastive Learning and Uncorrelated Conditioning
Dayuan Jian (University of Southern California), Mohammad Rostami (University of Southern California)
Domain AdaptationRepresentation LearningConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised domain adaptation framework that transfers labeled frame images to unlabeled event data, enhancing representation learning through contrastive learning and uncorrelated conditional constraints.
Unsupervised Domain Adaptive Detection with Network Stability Analysis
Wenzhang Zhou (Institute of Software Chinese Academy of Sciences), Libo Zhang (Institute of Software Chinese Academy of Sciences)
Object DetectionDomain AdaptationAutonomous DrivingKnowledge DistillationImage
🎯 What it does: A new unsupervised domain adaptation detection framework is proposed, which improves the generalization ability of detectors learned from labeled source domains to unlabeled target domains through network stability analysis.
Unsupervised Facial Performance Editing via Vector-Quantized StyleGAN Representations
Berkay Kicanaoglu (Flawless AI), Gaurav Bharaj (Flawless AI)
SegmentationGenerationGenerative Adversarial NetworkImageVideo
🎯 What it does: This paper proposes an unsupervised video segmentation prior using StyleGAN feature vector quantization, combined with 3DMM and user-defined masks, to achieve local semantic editing of high-resolution facial videos (such as the insides of eyes and mouths).
Unsupervised Feature Representation Learning for Domain-generalized Cross-domain Image Retrieval
Conghui Hu (National University of Singapore), Gim Hee Lee (National University of Singapore)
RetrievalRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A domain generalization unsupervised cross-domain image retrieval framework DG-UCDIR is proposed, which utilizes unlabeled data from seen domains to train a universal feature extractor, enabling retrieval between any two unseen domains.
Unsupervised Image Denoising in Real-World Scenarios via Self-Collaboration Parallel Generative Adversarial Branches
Xin Lin (Sichuan University), Yinjie Lei (Sichuan University)
RestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an unsupervised image denoising network called SCPGabNet, which is based on self-coherent parallel generative adversarial branches and can improve denoising performance without increasing inference complexity.
Unsupervised Learning of Object-Centric Embeddings for Cell Instance Segmentation in Microscopy Images
Steffen Wolf (MRC Laboratory of Molecular Biology), Jan Funke (HHMI Janelia Research Campus)
SegmentationRepresentation LearningConvolutional Neural NetworkContrastive LearningImageMultimodalityBiomedical Data
🎯 What it does: This paper proposes an unsupervised cell instance segmentation method called CELLULUS, which learns representations by predicting spatial offsets between image patches using Object Center Embedding (OCE), and performs cell instance segmentation based on this.
Unsupervised Manifold Linearizing and Clustering
Tianjiao Ding (Johns Hopkins University), Benjamin D. Haeffele (Johns Hopkins University)
Representation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Designed and implemented an algorithm for simultaneous data clustering and learning low-dimensional linear representations in an unsupervised manner—Manifold Linearizing and Clustering (MLC).
Unsupervised Object Localization with Representer Point Selection
Yeonghwan Song (Gwangju Institute of Science and Technology), Jeany Son (Gwangju Institute of Science and Technology)
Object DetectionExplainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A method for unsupervised object localization based on representative point selection is proposed, which generates a foreground predictor using a self-supervised pre-trained model and provides interpretable localization results without the need for additional fine-tuning or supervision.
Unsupervised Open-Vocabulary Object Localization in Videos
Ke Fan (Fudan University), Tong He (Amazon Web Services)
Object DetectionSegmentationContrastive LearningVideo
🎯 What it does: A completely unsupervised spatiotemporal object localization and naming framework is proposed, combining Slot Attention with CLIP to achieve spatiotemporal consistent segmentation and text label assignment of objects in videos.
Unsupervised Prompt Tuning for Text-Driven Object Detection
Weizhen He (Zhejiang University), Yueting Zhuang (Zhejiang University)
Object DetectionPrompt EngineeringImage
🎯 What it does: This paper studies an unsupervised prompt tuning method aimed at improving the performance of GLIP-based text-driven object detection models in downstream tasks without using any labeled data.
Unsupervised Self-Driving Attention Prediction via Uncertainty Mining and Knowledge Embedding
Pengfei Zhu (Beijing University of Posts and Telecommunications), Huadong Ma (University of Rochester)
Autonomous DrivingImage
🎯 What it does: A novel unsupervised self-driving attention prediction model is proposed, which utilizes a natural scene pre-trained model to generate pseudo-labels and achieves adaptive training through uncertainty mining and knowledge embedding.
Unsupervised Surface Anomaly Detection with Diffusion Probabilistic Model
Xinyi Zhang (Tsinghua University), Shu-Tao Xia (Peng Cheng Laboratory)
Anomaly DetectionDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes an unsupervised surface defect detection method called DiffAD based on latent diffusion models, which utilizes noise conditional embedding and interpolation channels to enhance reconstruction quality and localization accuracy.
Unsupervised Video Deraining with An Event Camera
Jin Wang (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationContrastive LearningVideo
🎯 What it does: This paper proposes an unsupervised video de-raining network that combines event cameras, and designs heterogeneous separation modules and cross-modal fusion modules.
Unsupervised Video Object Segmentation with Online Adversarial Self-Tuning
Tiankang Su (Nanjing University of Information Science and Technology), Qingshan Liu (Nanjing University of Posts and Telecommunications)
Object DetectionSegmentationTransformerGenerative Adversarial NetworkOptical FlowVideo
🎯 What it does: This paper proposes a UVOS framework called OAST, which combines offline semi-supervised adversarial training with online self-supervised adversarial fine-tuning, significantly improving unsupervised video object segmentation performance by adaptively updating the model during testing.
UpCycling: Semi-supervised 3D Object Detection without Sharing Raw-level Unlabeled Scenes
Sunwook Hwang (Seoul National University), Hyung-Sin Kim (Seoul National University)
Object DetectionDomain AdaptationAutonomous DrivingSafty and PrivacySupervised Fine-TuningPoint Cloud
🎯 What it does: A semi-supervised learning framework named UpCycling is proposed for updating 3D object detection models using de-identified intermediate features generated by autonomous vehicles, without sharing the original point clouds.
Urban Radiance Field Representation with Deformable Neural Mesh Primitives
Fan Lu (Tongji University), Changjun Jiang (Tongji University)
Autonomous DrivingRepresentation LearningNeural Radiance FieldAuto EncoderPoint CloudMesh
🎯 What it does: This paper proposes and implements Deformable Neural Mesh Primitives (DNMP), which achieves local modeling of scene geometry and radiance information by voxelizing large-scale urban point clouds and assigning DNMP to each voxel.
UrbanGIRAFFE: Representing Urban Scenes as Compositional Generative Neural Feature Fields
Yuanbo Yang (Zhejiang University), Yiyi Liao (Zhejiang University)
GenerationData SynthesisAutonomous DrivingConvolutional Neural NetworkNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: This paper presents UrbanGIRAFFE, a compositional generative model based on 3D priors that can synthesize realistic urban scenes with controllable camera angles, object positions, and scene decorations (stuff).
USAGE: A Unified Seed Area Generation Paradigm for Weakly Supervised Semantic Segmentation
Zelin Peng (Shanghai Jiao Tong University), Qi Tian (Huawei Inc.)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: A unified seed region generation paradigm, USAGE, is proposed to address the issues of overly weak or strong activations in CNNs and Transformers in weakly supervised semantic segmentation.
Using a Waffle Iron for Automotive Point Cloud Semantic Segmentation
Gilles Puy (Valeo), Renaud Marlet (Université Gustave Eiffel)
SegmentationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: WaffleIron is proposed, a novel 3D backbone network that achieves semantic segmentation of vehicle-mounted LiDAR point clouds using only MLP and dense 2D convolutions, without the need for sparse convolutions.
uSplit: Image Decomposition for Fluorescence Microscopy
Ashesh Ashesh (Human Technopole), Florian Jug (Human Technopole)
SegmentationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: Proposes the µ Split method, which efficiently performs image decomposition in fluorescence microscopy images using lateral context (LC), splitting the superimposed two-channel images into their respective channels.
V-FUSE: Volumetric Depth Map Fusion with Long-Range Constraints
Nathaniel Burgdorfer (Stevens Institute of Technology), Philippos Mordohai (Stevens Institute of Technology)
Depth EstimationConvolutional Neural NetworkPoint CloudBenchmark
🎯 What it does: An end-to-end deep depth fusion framework V-FUSE is proposed, which improves the depth and confidence maps generated by MVS using multi-view depth and confidence maps combined with learnable long-range voxel visibility constraints.
V3Det: Vast Vocabulary Visual Detection Dataset
Jiaqi Wang (Shanghai AI Laboratory), Dahua Lin (Chinese University of Hong Kong)
Object DetectionTransformerLarge Language ModelImage
🎯 What it does: The V3Det dataset has been constructed and made public, collecting 243k high-resolution images, annotating 13,204 categories with precise bounding boxes, and providing a hierarchical category tree and professional descriptions.
VAD: Vectorized Scene Representation for Efficient Autonomous Driving
Bo Jiang (Huazhong University of Science and Technology), Xinggang Wang (Huazhong University of Science and Technology)
Autonomous DrivingTransformerMultimodality
🎯 What it does: This paper proposes an end-to-end autonomous driving framework called VAD, which is completely based on vectorized scene representation (vectorized maps and vectorized motion) and directly outputs driving trajectories from multi-view cameras, eliminating the traditional rasterized perception and post-processing steps.
VADER: Video Alignment Differencing and Retrieval
Alexander Black (University of Surrey), John Collomosse (Adobe Research)
RetrievalConvolutional Neural NetworkTransformerContrastive LearningVideo
🎯 What it does: VADER is a set of spatiotemporal matching, alignment, and differential techniques for video clips, capable of retrieving original videos from massive video libraries and locating segments, followed by visualizing and annotating edited areas;
Vanishing Point Estimation in Uncalibrated Images with Prior Gravity Direction
Rémi Pautrat (ETH Zurich), Daniel Barath (Microsoft)
Pose EstimationOptimizationSimultaneous Localization and MappingImage
🎯 What it does: Two types of 2-line minimal solvers based on known gravity direction are proposed for estimating the Manhattan frame and focal length under uncalibrated cameras, along with the design of a non-minimal solver and a hybrid RANSAC framework.
VAPCNet: Viewpoint-Aware 3D Point Cloud Completion
Zhiheng Fu (University of Western Australia), Mohammed Bennamoun (University of Western Australia)
RestorationRepresentation LearningConvolutional Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: An unsupervised perspective representation learning and perspective-aware 3D point cloud completion network, VAPCNet, is proposed, capable of completing sparse point clouds from unknown viewpoints.
Variational Causal Inference Network for Explanatory Visual Question Answering
Dizhan Xue (Chinese Academy of Sciences), Changsheng Xu (Chinese Academy of Sciences)
GenerationExplainability and InterpretabilityTransformerVision Language ModelMultimodality
🎯 What it does: A Variational Causal Inference Network (VCIN) is proposed to achieve causal consistency and interpretability generation between visual question answering answers and multimodal explanations.
Variational Degeneration to Structural Refinement: A Unified Framework for Superimposed Image Decomposition
Wenyu Li (Tianjin University), Yue Lang (Hebei University of Technology)
RestorationAuto EncoderImage
🎯 What it does: A unified framework VDSR is proposed for the decomposition of single mixed images, and it is extended to tasks such as rain removal, reflection removal, and shadow removal.
Verbs in Action: Improving Verb Understanding in Video-Language Models
Liliane Momeni (University of Oxford), Cordelia Schmid (Google Research)
Large Language ModelVision Language ModelContrastive LearningVideoText
🎯 What it does: By incorporating hard negative samples targeting verbs and verb phrase alignment loss into video-text contrastive learning, the verb understanding ability of CLIP-like video language models is enhanced.
VeRi3D: Generative Vertex-based Radiance Fields for 3D Controllable Human Image Synthesis
Xinya Chen (Zhejiang University), Yiyi Liao (Zhejiang University)
GenerationData SynthesisPose EstimationNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: We propose VeRi3D, a generative radiance field based on SMPL vertices, capable of learning controllable 3D human image synthesis from single-view 2D images, supporting camera, pose, body shape, and part-level editing.
Versatile Diffusion: Text, Images and Variations All in One Diffusion Model
Xingqian Xu (SHI Labs), Humphrey Shi (SHI Labs)
GenerationData SynthesisDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: This paper proposes Versatile Diffusion (VD), a multi-stream multi-modal diffusion model capable of performing tasks such as text-to-image, image-to-text, and image variant generation within the same network, and supports dual/multi-context mixing and style/semantic unsupervised decoupling.
VertexSerum: Poisoning Graph Neural Networks for Link Inference
Ruyi Ding (Northeastern University), Yunsi Fei (Northeastern University)
Safty and PrivacyAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: A privacy leakage attack on graph neural networks called VertexSerum is proposed, which amplifies connection information and successfully steals the link relationships between nodes by applying lightweight adversarial perturbations to the node features during the training phase.
VI-Net: Boosting Category-level 6D Object Pose Estimation via Learning Decoupled Rotations on the Spherical Representations
Jiehong Lin (South China University of Technology), Kui Jia (South China University of Technology)
Object DetectionPose EstimationPoint Cloud
🎯 What it does: VI-Net is proposed to achieve high-precision category-level 6D object pose estimation by performing feature learning on the sphere and decoupling rotation into viewpoint rotation and in-plane rotation, utilizing V-Branch and I-Branch for binary classification and regression, respectively.
Video Action Recognition with Attentive Semantic Units
Yifei Chen (Huawei), Wei Peng (Huawei)
RecognitionRepresentation LearningTransformerVision Language ModelContrastive LearningVideo
🎯 What it does: This paper proposes the use of fine-grained semantic units (Semantic Units, SU) and Multi-Region Attention (MRA) to guide representation learning for video action recognition, forming a cross-modal decoder to generate spatiotemporal video representations.
Video Action Segmentation via Contextually Refined Temporal Keypoints
Borui Jiang (Peking University), Yadong Mu (Peking University)
RecognitionSegmentationGraph Neural NetworkTransformerVideo
🎯 What it does: A context-refined temporal keypoint (RTK) method is proposed, which represents actions using sparse keypoints and achieves video action segmentation through graph matching and rule reorganization.
Video Adverse-Weather-Component Suppression Network via Weather Messenger and Adversarial Backpropagation
Yijun Yang (Hong Kong University of Science and Technology), Lei Zhu (Hong Kong University of Science and Technology)
RestorationTransformerGenerative Adversarial NetworkVideo
🎯 What it does: A unified video multi-bad weather removal framework, ViWS-Net, has been designed to restore video frames contaminated by various weather conditions such as rain, fog, and snow all at once.
Video Anomaly Detection via Sequentially Learning Multiple Pretext Tasks
Chenrui Shi (Beijing Institute of Technology), Yunde Jia (Beijing Institute of Technology)
Anomaly DetectionContrastive LearningVideo
🎯 What it does: This paper proposes to learn multi-task pre-training tasks in ascending order of difficulty to improve the performance of unsupervised video anomaly detection.
Video Background Music Generation: Dataset, Method and Evaluation
Le Zhuo (Beihang University), Si Liu (Beihang University)
GenerationData SynthesisRetrievalTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: A complete video background music generation scheme is proposed, including datasets, models, and evaluation metrics;
Video Object Segmentation-aware Video Frame Interpolation
Jun-Sang Yoo (Korea University), Seung-Won Jung (Korea University)
Image TranslationObject TrackingSegmentationOptical FlowVideo
🎯 What it does: This paper proposes an auxiliary training framework VOS-VFI based on Video Object Segmentation (VOS) to enhance the clarity and accuracy of foreground object boundaries in Video Frame Interpolation (VFI) models.
Video OWL-ViT: Temporally-consistent Open-world Localization in Video
Georg Heigold (Google), Thomas Kipf (Google)
Object DetectionObject TrackingTransformerImageVideo
🎯 What it does: This paper proposes an end-to-end model, Video OWL-ViT, that enables open-world object localization and tracking in videos. It integrates the pre-trained open-world image detector OWL-ViT by adding a Transformer decoder and recursively applying it to video frames, achieving temporally consistent localization without relying on manual matching.
Video State-Changing Object Segmentation
Jiangwei Yu (University of Illinois at Urbana-Champaign), Yu-Xiong Wang (University of Illinois at Urbana-Champaign)
Object DetectionSegmentationContrastive LearningOptical FlowVideoBenchmark
🎯 What it does: This paper proposes the Video State Change Object Segmentation (VSCOS) task and constructs a benchmark dataset and evaluation metrics.
Video Task Decathlon: Unifying Image and Video Tasks in Autonomous Driving
Thomas E. Huang (ETH Zurich), Fisher Yu (ETH Zurich)
Autonomous DrivingVideo
🎯 What it does: Proposed the Video Task Decathlon (VTD) and a unified network VTDNet, supporting ten different image and video tasks;
Video-FocalNets: Spatio-Temporal Focal Modulation for Video Action Recognition
Syed Talal Wasim (Mohamed bin Zayed University of AI), Fahad Shahbaz Khan (Australian National University)
RecognitionConvolutional Neural NetworkTransformerVideo
🎯 What it does: Proposes Video-FocalNet, a spatiotemporal video recognition architecture that combines convolution with focal modulation;
VideoFlow: Exploiting Temporal Cues for Multi-frame Optical Flow Estimation
Xiaoyu Shi (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
Autonomous DrivingTransformerOptical FlowVideo
🎯 What it does: Proposes the VideoFlow framework, which estimates bidirectional optical flow using three or more frames of video simultaneously.
VidStyleODE: Disentangled Video Editing via StyleGAN and NeuralODEs
Moayed Haji Ali, Aykut Erdem
GenerationData SynthesisGenerative Adversarial NetworkVideoOrdinary Differential Equation
🎯 What it does: We propose VidStyleODE, which utilizes StyleGAN2 and neural ODE to separate video content and motion, constructing a high-resolution video generation and editing framework that is consistent in duration and controllable by text.
View Consistent Purification for Accurate Cross-View Localization
Shan Wang (Australian National University), Hongdong Li (Australian National University)
Pose EstimationAutonomous DrivingConvolutional Neural NetworkSimultaneous Localization and MappingImage
🎯 What it does: A sparse visual cross-view localization method based on satellite images and multi-camera systems, named PureACL, is proposed, which can achieve precise 3-DoF (lateral, longitudinal, yaw) localization in dynamic environments through key points that are consistent in perspective and located on the ground.
Viewing Graph Solvability in Practice
Federica Arrigoni (Politecnico di Milano), Andrea Fusiello (University of Udine)
Graph
🎯 What it does: This paper studies the view graph solvability problem in structured light 3D reconstruction and proposes an efficient method to test finite solvability in large-scale uncalibrated graphs and extract the maximum solvable subgraph.
ViewRefer: Grasp the Multi-view Knowledge for 3D Visual Grounding
Zoey Guo (Shanghai Artificial Intelligence Laboratory), Xuelong Li (Shanghai Artificial Intelligence Laboratory)
RecognitionObject DetectionTransformerLarge Language ModelPoint Cloud
🎯 What it does: A multi-view framework called ViewRefer is proposed, which extends text using language models and achieves visual localization of 3D scenes by integrating multi-view 3D features through a fusion transformer.
Viewset Diffusion: (0-)Image-Conditioned 3D Generative Models from 2D Data
Stanislaw Szymanowicz (Visual Geometry Group University of Oxford), Andrea Vedaldi (Visual Geometry Group University of Oxford)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A viewset generation method based on diffusion models (Viewset Diffusion) is proposed, which learns to generate 3D objects using only multi-view 2D image data and supports 3D reconstruction and generation under zero-view, single-view, and multi-view conditions.
ViLLA: Fine-Grained Vision-Language Representation Learning from Real-World Data
Maya Varma (Stanford University), Curtis Langlotz (Stanford University)
Object DetectionRetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodalityMagnetic Resonance ImagingComputed TomographyPositron Emission TomographyUltrasound
🎯 What it does: This study investigates whether VLM trained on high 'pairwise complexity' datasets can learn fine-grained region-attribute relationships, and proposes ViLLA, which retrains VLM by generating region-attribute pairs through self-supervised mapping.
ViLTA: Enhancing Vision-Language Pre-training through Textual Augmentation
Weihan Wang (Tsinghua University), Yankui Sun (Tsinghua University)
RetrievalKnowledge DistillationRepresentation LearningTransformerVision Language ModelImageTextMultimodality
🎯 What it does: The ViLTA model is proposed to improve visual-language pre-training by combining cross distillation and language model-based synthetic hard negative samples to enhance robustness and convergence speed.
ViM: Vision Middleware for Unified Downstream Transferring
Yutong Feng (Alibaba Group), Jingren Zhou (Alibaba Group)
ClassificationObject DetectionSegmentationDepth EstimationTransformerMixture of ExpertsVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: A Vision Middleware (ViM) framework has been constructed, where a series of lightweight plugin modules (a total of 47) are trained by freezing the backbone on large models (such as CLIP ViT‑B/16), and these modules are aggregated as needed for multi-task unified transfer in downstream tasks.
ViperGPT: Visual Inference via Python Execution for Reasoning
Dídac Surís (Columbia University), Carl Vondrick (Columbia University)
Object DetectionTransformerLarge Language ModelVision Language ModelImageVideoMultimodality
🎯 What it does: The ViperGPT framework is proposed, utilizing large language models (such as Codex) to generate Python code that combines visual and language modules, enabling programmatic reasoning directly on images/videos without the need for additional training.
Virtual Try-On with Pose-Garment Keypoints Guided Inpainting
Zhi Li (Bytedance), Alex C. Kot (Nanyang Technological University)
Image TranslationGenerationGraph Neural NetworkDiffusion modelAuto EncoderImage
🎯 What it does: A posture-clothing keypoint guided virtual try-on method KGI has been designed and implemented, which can generate high-fidelity try-on effects that maintain the shape and pattern of clothing given a portrait and clothing image.
Visible-Infrared Person Re-Identification via Semantic Alignment and Affinity Inference
Xingye Fang (Beijing Institute of Technology), Ying Fu (Beijing Institute of Technology)
RecognitionRetrievalImageMultimodality
🎯 What it does: This paper proposes an end-to-end visible-infrared person re-identification framework called SAAI, which combines semantic alignment feature learning and an affinity reasoning module to achieve cross-modal person matching.
Vision Grid Transformer for Document Layout Analysis
Cheng Da (Alibaba Group), Cong Yao (Alibaba Group)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a Dual-Stream Vision Grid Transformer (VGT), which utilizes the Grid Transformer (GiT) for token-level and segment-level semantic pre-training on a two-dimensional document grid, and then combines it with ViT to perform document layout analysis.
Vision HGNN: An Image is More than a Graph of Nodes
Yan Han (University of Texas at Austin), Zhangyang Wang (University of Texas at Austin)
ClassificationObject DetectionSegmentationGraph Neural NetworkImage
🎯 What it does: Proposes the Vision HyperGraph Neural Network (ViHGNN), which divides images into patches as nodes to represent images as hypergraphs, and dynamically learns the hypergraph structure during training through Fuzzy C-Means.
Vision Relation Transformer for Unbiased Scene Graph Generation
Gopika Sudhakaran (Technical University of Darmstadt), Stefan Roth (Technical University of Darmstadt)
ClassificationObject DetectionGenerationTransformerMixture of ExpertsImageMultimodality
🎯 What it does: This paper proposes a Vision Relation Transformer (VETO) that enhances relation prediction through local-level entity patch generation and multimodal fusion, combined with a Mutually Exclusive ExperT (MEET) multi-expert learning strategy to achieve unbiased scene graph generation.
Vision Transformer Adapters for Generalizable Multitask Learning
Deblina Bhattacharjee (École Polytechnique Fédérale de Lausanne), Mathieu Salzmann (École Polytechnique Fédérale de Lausanne)
Domain AdaptationTransformerImage
🎯 What it does: Design scalable adapters on pre-trained visual Transformers to achieve multi-task learning under the premise of parameter efficiency, and enable zero-shot transfer to new tasks and new domains.
Visual Explanations via Iterated Integrated Attributions
Oren Barkan (Open University), Noam Koenigstein (Tel Aviv University)
SegmentationExplainability and InterpretabilityConvolutional Neural NetworkTransformerImage
🎯 What it does: A general visual model interpretation method called Iterated Integrated Attributions (IIA) is proposed, which generates precise explanation heatmaps through iterative integration of inputs, internal representations, and their gradients.
Visual Traffic Knowledge Graph Generation from Scene Images
Yunfei Guo (Institute of Automation of Chinese Academy of Sciences), Cheng-Lin Liu (Tencent Technology)
Object DetectionAutonomous DrivingGraph Neural NetworkImageGraph
🎯 What it does: This paper proposes the Visual Traffic Knowledge Graph Generation (VTKGG) task, which involves extracting information about roads, lanes, traffic signs, and their components from scene images, and constructing a traffic knowledge graph that includes various heterogeneous elements and complex relationships.
Visually-Prompted Language Model for Fine-Grained Scene Graph Generation in an Open World
Qifan Yu (Zhejiang University), Yueting Zhuang (Zhejiang University)
Object DetectionGenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes an automated data augmentation framework called CaCao, which extracts fine-grained predicates from large-scale pre-trained language models using a visual prompting language model to address the long-tail distribution problem in visual relation generation. It further introduces the Epic open-world predicate generation module to achieve zero-shot predicate prediction.
VL-Match: Enhancing Vision-Language Pretraining with Token-Level and Instance-Level Matching
Junyu Bi (Institute of Computing Technology, Chinese Academy of Sciences), Qi Zhang (Microsoft Corporation)
RetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: VL-Match is designed to introduce Vision-Language Replaced Token Detection (VL-RTD) at the token level using a generative-discriminative framework, and to generate fine-grained negative samples at the instance level through Fine-Grained Image-Text Matching (FG-ITM) and NegGen, enhancing the matching capability of vision-language pre-training.
VL-PET: Vision-and-Language Parameter-Efficient Tuning via Granularity Control
Zi-Yuan Hu (Chinese University of Hong Kong), Liwei Wang (Chinese University of Hong Kong)
TransformerVision Language ModelImageVideoTextMultimodality
🎯 What it does: A parameter-efficient fine-tuning framework for visual-language tasks, VL-PET, is proposed, which improves the adaptation effect of PLM through a granularity control mechanism, lightweight module design, and multi-head modular modifications.
VLN-PETL: Parameter-Efficient Transfer Learning for Vision-and-Language Navigation
Yanyuan Qiao (Australian Institute for Machine Learning, University of Adelaide), Qi Wu (Australian Institute for Machine Learning, University of Adelaide)
TransformerPrompt EngineeringMultimodality
🎯 What it does: The study applies the Parameter-Efficient Transfer Learning (PETL) method to Visual Language Navigation (VLN) and proposes the VLN-PETL framework, designing the Historical Interaction Boost (HIB) and Cross-Modal Interaction Boost (CIB) modules.
VLSlice: Interactive Vision-and-Language Slice Discovery
Eric Slyman (Oregon State University), Stefan Lee (Oregon State University)
Object DetectionRecommendation SystemTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper presents an interactive system called VL Slice, which helps researchers quickly discover and construct subsets (slices) that align with the bias dimensions of visual language models within unannotated image collections.