CVPR 2024 Papers — Page 26
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
UFORecon: Generalizable Sparse-View Surface Reconstruction from Arbitrary and Unfavorable Sets
Youngju Na (KAIST), Sung-Eui Yoon (KAIST)
TransformerImage
🎯 What it does: A universal model called UFORecon is proposed for 3D surface reconstruction from arbitrary, non-overlapping multi-view image sets.
ULIP-2: Towards Scalable Multimodal Pre-training for 3D Understanding
Le Xue (Salesforce AI Research), Silvio Savarese (Stanford University)
GenerationRepresentation LearningTransformerContrastive LearningTextMultimodalityPoint Cloud
🎯 What it does: Designed and implemented the ULIP-2 framework, which automatically generates fine-grained language descriptions of 3D shapes through large multimodal models, constructs an unannotated tri-modal (point cloud, image, text) dataset, and conducts pre-training to enhance 3D representation learning effectiveness.
UltrAvatar: A Realistic Animatable 3D Avatar Diffusion Model with Authenticity Guided Textures
Mingyuan Zhou (OPPO US Research Center), Guojun Qi (Westlake University)
GenerationDiffusion modelImageMesh
🎯 What it does: UltrAvatar has been developed to generate animatable 3D avatars with realistic lighting editable PBR textures from text prompts or a single facial image.
Unbiased Estimator for Distorted Conics in Camera Calibration
Chaehyeon Song (Seoul National University), Ayoung Kim (Seoul National University)
Pose EstimationOptimizationImage
🎯 What it does: Proposed and implemented an unbiased estimation method using moments for precise center point estimation of distorted ellipses (circular calibration patterns), thereby achieving camera calibration based on circular calibration patterns.
Unbiased Faster R-CNN for Single-source Domain Generalized Object Detection
Yajing Liu (Chinese Academy of Sciences), Jiandong Tian (Nanjing University of Posts and Telecommunications)
Object DetectionDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an Unbiased Faster R‑CNN model for single-source domain generalization in object detection, aiming to eliminate data bias, attention bias, and prototype bias, thereby improving detection performance in unseen domains (different weather scenarios).
Uncertainty Visualization via Low-Dimensional Posterior Projections
Omer Yair (Technion Israel Institute of Technology), Tomer Michaeli (Technion Israel Institute of Technology)
Image TranslationRestorationGenerationDiffusion modelImageBiomedical Data
🎯 What it does: A framework is proposed for estimating and visualizing posterior distributions in low-dimensional subspaces for image inversion problems.
Uncertainty-aware Action Decoupling Transformer for Action Anticipation
Hongji Guo (Rensselaer Polytechnic Institute), Qiang Ji (Rensselaer Polytechnic Institute)
RecognitionTransformerGaussian SplattingOptical FlowVideoMultimodality
🎯 What it does: A motion prediction framework named UADT is proposed, which decouples motion prediction into two steps: predicting verbs and nouns, and quantifies and filters the uncertainty of each step through a probabilistic Transformer, achieving mutual assistance between the two models to enhance prediction performance.
Uncertainty-Aware Source-Free Adaptive Image Super-Resolution with Wavelet Augmentation Transformer
Yuang Ai (Institute of Automation, Chinese Academy of Sciences), Ran He (Institute of Automation, Chinese Academy of Sciences)
RestorationSuper ResolutionDomain AdaptationKnowledge DistillationTransformerGenerative Adversarial NetworkImage
🎯 What it does: A source-free domain adaptive image super-resolution framework SODA-SR is proposed, utilizing teacher-student learning and pseudo-labels to achieve domain transfer without source data.
Uncertainty-Guided Never-Ending Learning to Drive
Lei Lai (Boston University), John Seon Keun Yi (Boston University)
Autonomous DrivingReinforcement LearningVideo
🎯 What it does: This paper proposes a self-supervised perpetual learning framework named ∞-Driver, which can learn and update end-to-end driving strategies in real-time from online video streams without relying on labeled data.
Uncovering What Why and How: A Comprehensive Benchmark for Causation Understanding of Video Anomaly
Hang Du (Beijing University of Posts and Telecommunications), Xiaofeng Tao (Beijing University of Posts and Telecommunications)
Anomaly DetectionTransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: This study proposes a new benchmark for video anomaly causal understanding, CUVA, which includes three-dimensional descriptions of 'What happened', 'Why it happened', and 'How severe it is'. It also designs the MMEval evaluation framework and the prompt-based VLM method A-Guardian for baseline evaluation.
Understanding and Improving Source-free Domain Adaptation from a Theoretical Perspective
Yu Mitsuzumi (NTT Corporation), Hisashi Kashima (Kyoto University)
Domain AdaptationImage
🎯 What it does: This paper analyzes the role of discriminative and diversity loss in source-free domain adaptation (SFDA) from a theoretical perspective and proposes an improved SFDA method based on this analysis.
Understanding Video Transformers via Universal Concept Discovery
Matthew Kowal (York University), Pavel Tokmakov (Toyota Research Institute)
SegmentationExplainability and InterpretabilityRepresentation LearningTransformerVideo
🎯 What it does: The VTCD algorithm is proposed, which can discover interpretable spatiotemporal concepts from video Transformers under unsupervised conditions and assess their importance to model predictions.
Unexplored Faces of Robustness and Out-of-Distribution: Covariate Shifts in Environment and Sensor Domains
Eunsu Baek (Seoul National University), Hyung-Sin Kim (Seoul National University)
Domain AdaptationAnomaly DetectionImage
🎯 What it does: This paper constructs a controllable testing platform, ES-Studio, using real cameras to capture 202k images under different lighting conditions and camera sensor parameters, creating the ImageNet-ES dataset, and conducting experiments on OOD detection, domain generalization, and camera sensor control on this dataset.
Ungeneralizable Examples
Jingwen Ye (National University of Singapore), Xinchao Wang (National University of Singapore)
ClassificationKnowledge DistillationAdversarial AttackContrastive LearningImage
🎯 What it does: A framework of Ungeneralizable Examples (UGEs) is proposed, allowing data to be learnable on authorized networks while being unlearnable on adversarial networks;
UniBind: LLM-Augmented Unified and Balanced Representation Space to Bind Them All
Yuanhuiyi Lyu (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
Representation LearningTransformerLarge Language ModelContrastive LearningImageVideoMultimodalityPoint CloudAudio
🎯 What it does: This paper proposes UniBind, which constructs modality-independent embedding centers using a knowledge base generated by large language models, and learns a unified and balanced representation space by aligning features from all seven modalities.
UniDepth: Universal Monocular Metric Depth Estimation
Luigi Piccinelli (ETH Zurich INSAIT), Fisher Yu (ETH Zurich INSAIT)
Depth EstimationAutonomous DrivingTransformerImage
🎯 What it does: A method called UniDepth is proposed for predicting complete 3D point full-scale depth estimation using only a single RGB image without any camera information.
Unified Entropy Optimization for Open-Set Test-Time Adaptation
Zhengqing Gao (Institute of Automation, Chinese Academy of Sciences), Cheng-Lin Liu (Institute of Automation, Chinese Academy of Sciences)
Domain AdaptationOptimizationImage
🎯 What it does: A unified entropy optimization framework (UniEnt/UniEnt+) is proposed for adaptive and unknown category detection of pre-trained models in an open-set testing environment (Open-Set TTA).
Unified Language-driven Zero-shot Domain Adaptation
Senqiao Yang (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
SegmentationDomain AdaptationAutonomous DrivingContrastive LearningImage
🎯 What it does: A unified language-driven zero-shot domain adaptation (ULDA) framework is proposed, enabling a single model to adapt to multiple target domains without using target domain images or domain IDs.
Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision Language Audio and Action
Jiasen Lu (Allen Institute for AI), Aniruddha Kembhavi (University of Washington)
GenerationData SynthesisRobotic IntelligenceTransformerVision Language ModelVision-Language-Action ModelGenerative Adversarial NetworkImageVideoTextMultimodalityAudio
🎯 What it does: A self-regressive multimodal model called Unified-IO 2 is proposed, which can simultaneously process images, text, audio, and actions.
Unifying Automatic and Interactive Matting with Pretrained ViTs
Zixuan Ye (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
SegmentationTransformerImage
🎯 What it does: A unified image matting framework called Smart Matting (SMat) is proposed, which accommodates both prompt-free automatic matting and prompt-based interactive matting.
Unifying Correspondence Pose and NeRF for Generalized Pose-Free Novel View Synthesis
Sunghwan Hong (Korea University), Chong Luo (Microsoft Research Asia)
Image TranslationData SynthesisPose EstimationConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: A unified framework called CoPoNeRF is proposed, which simultaneously performs 2D correspondence, camera pose estimation, and NeRF rendering on pose-free stereo image pairs, directly synthesizing new perspective images.
Unifying Top-down and Bottom-up Scanpath Prediction Using Transformers
Zhibo Yang (Stony Brook University), Dimitris Samaras (Stony Brook University)
TransformerImage
🎯 What it does: A unified Human Attention Transformer (HAT) model is proposed to simultaneously predict goal-directed (top-down) and free viewing (bottom-up) scan paths;
UniGarmentManip: A Unified Framework for Category-Level Garment Manipulation via Dense Visual Correspondence
Ruihai Wu (Peking University), Hao Dong (Peking University)
Domain AdaptationRobotic IntelligenceContrastive LearningPoint Cloud
🎯 What it does: The research proposes a unified dense visual correspondence framework (UniGarmentManip) that obtains the topological and functional correspondence of clothing at the category level through self-supervised learning, enabling the completion of various clothing manipulation tasks (unfolding, folding, hanging) with a single or few demonstrations.
UniGS: Unified Representation for Image Generation and Segmentation
Lu Qi (University of California), Ming-Hsuan Yang (University of California)
SegmentationGenerationDiffusion modelImage
🎯 What it does: A unified diffusion model called UniGS is proposed, which can generate images and produce instance-level segmentation masks, using a colormap to maintain a consistent representation of the masks with the images.
UniHuman: A Unified Model For Editing Human Images in the Wild
Nannan Li (Boston University), Zhe Lin (Adobe)
Image TranslationGenerationPose EstimationDiffusion modelImageText
🎯 What it does: We propose UniHuman, a unified model capable of simultaneously performing human pose reconstruction, virtual try-on, and text-based clothing editing in natural scenes.
UniMix: Towards Domain Adaptive and Generalizable LiDAR Semantic Segmentation in Adverse Weather
Haimei Zhao (University of Sydney), Dacheng Tao (Nanyang Technological University)
SegmentationDomain AdaptationPoint Cloud
🎯 What it does: The UniMix method is proposed, utilizing bridging domains and general mixing operations to enhance the domain adaptation and domain generalization performance of LiDAR semantic segmentation under adverse weather conditions.
UniMODE: Unified Monocular 3D Object Detection
Zhuoling Li, Hengshuang Zhao (Hong Kong University)
Object DetectionDomain AdaptationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A unified monocular 3D object detection framework called UniMODE is proposed, compatible with both indoor and outdoor scenes.
UnionFormer: Unified-Learning Transformer with Multi-View Representation for Image Manipulation Detection and Localization
Shuaibo Li (Beijing University of Technology), Xiaopeng Zhang (Institute of Automation Chinese Academy of Sciences)
Anomaly DetectionTransformerContrastive LearningImage
🎯 What it does: This paper studies a unified learning Transformer framework called UnionFormer, which integrates RGB, noise, and object views for image tampering detection and localization.
UniPAD: A Universal Pre-training Paradigm for Autonomous Driving
Honghui Yang (Zhejiang University), Wanli Ouyang (Zhejiang University)
Object DetectionSegmentationAutonomous DrivingNeural Radiance FieldMultimodalityPoint Cloud
🎯 What it does: A unified self-supervised pre-training paradigm called UniPAD is designed, which utilizes 3D differentiable rendering to map point clouds or multi-view images into voxel space and learns 3D features by reconstructing missing information through rendering.
UniPT: Universal Parallel Tuning for Transfer Learning with Efficient Parameter and Memory
Haiwen Diao (Dalian University of Technology), Long Chen (Hong Kong University of Science and Technology)
Convolutional Neural NetworkTransformerMultimodality
🎯 What it does: A new parameter-efficient transfer learning method called UniPT is proposed, which achieves transfer without the need for backpropagation through the main network by adding a lightweight parallel network alongside the pre-trained model.
UniPTS: A Unified Framework for Proficient Post-Training Sparsity
Jingjing Xie (Xiamen University), Rongrong Ji (Xiamen University)
ClassificationObject DetectionConvolutional Neural NetworkImage
🎯 What it does: A unified post-training sparse framework called UniPTS is proposed to achieve high sparsity while maintaining model performance with only a small amount of calibration data.
UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition
Xiaohan Ding (Tencent AI Lab), Ying Shan (Tencent AI Lab)
ClassificationRecognitionObject DetectionSegmentationConvolutional Neural NetworkImageVideoMultimodalityPoint CloudTime SeriesAudio
🎯 What it does: A universal large-kernel convolutional network called UniRepLKNet is proposed, which can achieve efficient recognition, segmentation, and detection in image tasks, as well as perform excellently in multimodal tasks such as audio, video, point clouds, and time series.
Universal Novelty Detection Through Adaptive Contrastive Learning
Hossein Mirzaei (Sharif University of Technology), Mohammad Hossein Rohban (Okinawa Institute of Science and Technology)
Anomaly DetectionContrastive LearningImage
🎯 What it does: The UNODE method is proposed, which implements universal anomaly detection using adaptive contrastive learning and probabilistic negative sample generation (AutoAugOOD).
Universal Robustness via Median Randomized Smoothing for Real-World Super-Resolution
Zakariya Chaouai (Universite Paris-Saclay), Mohamed Tamaazousti (Universite Paris-Saclay)
RestorationSuper ResolutionGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a robust super-resolution model CertSR implemented using Median Random Smoothing (MRS);
Universal Segmentation at Arbitrary Granularity with Language Instruction
Yong Liu (Tsinghua University), Yansong Tang (Tsinghua University)
SegmentationTransformerVision Language ModelImageMultimodality
🎯 What it does: A general segmentation model called UniLSeg based on language instructions is proposed, capable of completing segmentation tasks at any semantic granularity.
Universal Semi-Supervised Domain Adaptation by Mitigating Common-Class Bias
Wenyu Zhang (Institute for Infocomm Research Agency for Science Technology and Research), Chuan-Sheng Foo (Institute for Infocomm Research Agency for Science Technology and Research)
Domain AdaptationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new transfer learning setting called Universal Semi-Supervised Domain Adaptation (UniSSDA), which addresses the issues of partial labeling in the target domain and incomplete label space matching.
UniVS: Unified and Universal Video Segmentation with Prompts as Queries
Minghan Li (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
SegmentationConvolutional Neural NetworkTransformerPrompt EngineeringVideo
🎯 What it does: A unified video segmentation framework called UniVS is proposed, which utilizes prompts (visual or textual) as queries to uniformly handle all category-specific and prompt-specific video segmentation tasks.
Unknown Prompt the only Lacuna: Unveiling CLIP's Potential for Open Domain Generalization
Mainak Singha (Aisin Corporation), Biplab Banerjee (Indian Institute of Technology Bombay)
ClassificationDomain AdaptationPrompt EngineeringDiffusion modelContrastive LearningImageMultimodality
🎯 What it does: We propose ODG-CLIP, a multi-class classification framework that utilizes CLIP for open-domain generalization, capable of recognizing both known and unknown class samples.
Unleashing Channel Potential: Space-Frequency Selection Convolution for SAR Object Detection
Ke Li (Xidian University), Quan Wang (Xidian University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: A lightweight SAR target detection network based on Spatial-Frequency Selective Convolution (SFS-Conv) called SFS-CNet is proposed, which can extract multi-scale spatial features and multi-directional frequency features within a single layer of convolution and perform parameter-independent feature selection.
Unleashing Network Potentials for Semantic Scene Completion
Fengyun Wang (Nanjing University of Science and Technology), Jinhui Tang (Nanjing University of Science and Technology)
RecognitionSegmentationConvolutional Neural NetworkGenerative Adversarial NetworkMultimodality
🎯 What it does: The AMMNet framework is proposed, utilizing cross-modal modulation and adversarial training to enhance the performance of single-view RGB-D semantic scene reconstruction.
Unleashing the Potential of SAM for Medical Adaptation via Hierarchical Decoding
Zhiheng Cheng (East China Normal University), Yuyin Zhou (University of California Santa Cruz)
SegmentationDomain AdaptationTransformerSupervised Fine-TuningImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: We propose H-SAM, a prompt-free version of the Segment Anything Model, which achieves efficient fine-tuning and fine-grained segmentation of medical images through a two-stage hierarchical decoder.
Unleashing Unlabeled Data: A Paradigm for Cross-View Geo-Localization
Guopeng Li (Wuhan University), Gui-Song Xia (Wuhan University)
RetrievalDomain AdaptationGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes a completely unsupervised and semi-supervised cross-view geographic localization framework that utilizes unlabeled data to achieve retrieval from ground images to satellite images.
Unlocking Pre-trained Image Backbones for Semantic Image Synthesis
Tariq Berrada Ifriqi, Karteek Alahari (Univ. Grenoble Alpes)
SegmentationGenerationData SynthesisConvolutional Neural NetworkTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A GAN-based semantic image synthesis framework DP-SIMS is proposed, which introduces a frozen pre-trained feature backbone in the discriminator and uses cross-scale attention to inject noise in the generator, achieving high-quality, diverse, and segmentation-consistent image generation.
Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors
Ziqin Zhou (University of Adelaide), Lingqiao Liu (University of Adelaide)
SegmentationTransformerPrompt EngineeringImage
🎯 What it does: A few-shot semantic segmentation framework assisted by Relation Descriptor (RD) is proposed by leveraging the semantic grouping capability of pre-trained visual Transformers.
Unlocking the Potential of Prompt-Tuning in Bridging Generalized and Personalized Federated Learning
Wenlong Deng (University of British Columbia), Xiaoxiao Li (University of British Columbia)
Federated LearningTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes SGPT, a visual prompt tuning framework that combines shared prompts and grouped prompts in federated learning, enabling the global model to adapt to the data distribution of different clients without local fine-tuning.
Unmixing Before Fusion: A Generalized Paradigm for Multi-Source-based Hyperspectral Image Synthesis
Yang Yu (Wuhan University), Jiayi Ma (Wuhan University)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImageMultimodality
🎯 What it does: A 'decompose-then-fuse' paradigm is proposed, which maps multi-source (HSI and RGB) data to a low-dimensional dilution space, using generative models to generate diluted images and reconstruct high-quality high-dimensional HSI.
Unmixing Diffusion for Self-Supervised Hyperspectral Image Denoising
Haijin Zeng (IMEC-UGent), Wilfried Philips (IMEC-UGent)
RestorationTransformerDiffusion modelImage
🎯 What it does: A self-supervised hyperspectral image denoising framework called Diff-Unmix is proposed, based on spectral decomposition and diffusion models.
UnO: Unsupervised Occupancy Fields for Perception and Forecasting
Ben Agro (Waabi), Raquel Urtasun (Waabi)
Autonomous DrivingPoint Cloud
🎯 What it does: This paper proposes UNO, an unsupervised 4D occupancy world model that utilizes the implicit occupancy information from future LiDAR point clouds for self-supervised learning, capable of predicting the 3D spatial occupancy states in a time series; it also achieves transfer learning for point cloud prediction and BEV semantic occupancy.
Unraveling Instance Associations: A Closer Look for Audio-Visual Segmentation
Yuanhong Chen (Australian Institute for Machine Learning), Gustavo Carneiro (Centre for Vision, Speech and Signal Processing)
SegmentationConvolutional Neural NetworkContrastive LearningImageMultimodalityAudio
🎯 What it does: This paper proposes a cost-effective visual post-processing (VPO) dataset construction scheme and a supervised contrastive learning-based audio-visual segmentation method (CAVP) to improve the audio-visual segmentation task.
UnSAMFlow: Unsupervised Optical Flow Guided by Segment Anything Model
Shuai Yuan (Meta), Denis Demandolx (Meta)
Autonomous DrivingOptical FlowImageVideo
🎯 What it does: This paper proposes UnSAMFlow, an unsupervised optical flow estimation network that utilizes object masks from the Segment Anything Model (SAM);
UnScene3D: Unsupervised 3D Instance Segmentation for Indoor Scenes
David Rozenberszki (Technical University of Munich), Angela Dai (Technical University of Munich)
Object DetectionSegmentationTransformerContrastive LearningPoint Cloud
🎯 What it does: A completely unsupervised 3D instance segmentation framework called UnScene3D is proposed, which utilizes self-supervised 2D/3D features and geometric coarse segmentation to generate pseudo-masks, and improves instance segmentation quality through self-training iterations.
Unsegment Anything by Simulating Deformation
Jiahao Lu (National University of Singapore), Xinchao Wang (National University of Singapore)
SegmentationAdversarial AttackOptical FlowImage
🎯 What it does: A high-transferable, prompt-free adversarial attack method UAD is proposed for prompt-based segmentation models (such as SAM), which can significantly disrupt segmentation results across various models and prompts.
Unsigned Orthogonal Distance Fields: An Accurate Neural Implicit Representation for Diverse 3D Shapes
Yujie Lu (Donghua University), Lin Gao (Institute of Computing Technology, Chinese Academy of Sciences)
GenerationRepresentation LearningPoint CloudMesh
🎯 What it does: A neural implicit representation based on the unsigned orthogonal distance field (UODF) is proposed for the precise reconstruction of various 3D shapes.
Unsupervised 3D Structure Inference from Category-Specific Image Collections
Weikang Wang (University of Bonn), Florian Bernard (University of Bonn)
Pose EstimationAuto EncoderImage
🎯 What it does: This paper proposes an unsupervised 3D keypoint inference method that can learn 3D keypoints and shape structures using only a category-specific collection of single-view images.
Unsupervised Blind Image Deblurring Based on Self-Enhancement
Lufei Chen (Sichuan University), Chao Ren (Sichuan University)
RestorationGenerationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a completely unsupervised blind image deblurring framework that utilizes generative adversarial networks to synthesize pseudo-paired clear and blurred images, and continuously improves the performance of the deblurring model through a self-enhancement strategy.
Unsupervised Deep Unrolling Networks for Phase Unwrapping
Zhile Chen (South China University of Technology), Hui Ji (National University of Singapore)
RestorationOptimizationKnowledge DistillationImage
🎯 What it does: An unsupervised end-to-end deep reordering network is proposed to recover the original phase map from noisy wrapped phase maps.
Unsupervised Feature Learning with Emergent Data-Driven Prototypicality
Yunhui Guo (University of Texas at Dallas), Stella X. Yu (University of Michigan)
Representation LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a framework for unsupervised feature learning in hyperbolic space called HACK, which utilizes spherical packing to generate uniform target points and maps images into the Poincaré ball through Hungarian matching, allowing prototype images to naturally cluster at the center of the ball while atypical images are distributed at the boundary, capturing visual similarity and encoding data-driven prototypicality.
Unsupervised Gaze Representation Learning from Multi-view Face Images
Yiwei Bao (Beihang University), Feng Lu (Beihang University)
Pose EstimationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a multi-view unsupervised learning framework called Multi-View Dual-Encoder (MV-DE), which learns pupil rotation (i.e., gaze) representations from unlabeled multi-view facial images and uses them for pupil estimation and redirection.
Unsupervised Keypoints from Pretrained Diffusion Models
Eric Hedlin (University of British Columbia), Kwang Moo Yi (University of British Columbia)
Object DetectionGenerationPose EstimationDiffusion modelImage
🎯 What it does: This paper proposes an unsupervised keypoint learning method that utilizes the cross-attention mechanism of a pre-trained text-to-image diffusion model (Stable Diffusion) to automatically locate semantic keypoints in images.
Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos
Leonhard Sommer (University of Freiburg), Adam Kortylewski (University of Freiburg)
Pose EstimationVideo
🎯 What it does: A completely unsupervised category-level 3D pose estimation framework is proposed, which first aligns object center videos to the same coordinate system through multi-view self-supervised alignment, and then trains a neural mesh model to predict the correspondence from image pixels to 3D template vertices, achieving single-image pose inference.
Unsupervised Occupancy Learning from Sparse Point Cloud
Amine Ouasfi (Inria), Adnane Boukhayma (Inria)
GenerationOptimizationRepresentation LearningPoint Cloud
🎯 What it does: This paper proposes an unsupervised learning sparse noise non-directional point cloud occupancy field, which achieves adaptive fitting of surface boundaries through root searching on the confidence boundary of the occupancy function and entropy regularization, thereby directly inferring the complete three-dimensional shape.
Unsupervised Salient Instance Detection
Xin Tian (Huawei Technologies), Rynson Lau (City University of Hong Kong)
Object DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: An unsupervised salient instance detection method called SCoCo is proposed, which utilizes self-supervised Transformer features to achieve segmentation and identification of salient instances.
Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling
Leon Sick (Ulm University), Timo Ropinski (Ulm University)
SegmentationTransformerContrastive LearningImage
🎯 What it does: A novel unsupervised semantic segmentation method is proposed, which enhances performance through deep-guided feature correlation and sampling.
Unsupervised Template-assisted Point Cloud Shape Correspondence Network
Jiacheng Deng (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
RecognitionSegmentationTransformerPoint Cloud
🎯 What it does: This paper proposes an unsupervised Template-Assisted Point Cloud Shape Correspondence Network (TANet), which includes a template generation module and a template assistance module to achieve dense correspondence between source and target point clouds.
Unsupervised Universal Image Segmentation
Dantong Niu (Berkeley AI Research UC Berkeley), Trevor Darrell (Berkeley AI Research UC Berkeley)
Object DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: A unified unsupervised image segmentation framework U2Seg is proposed, capable of simultaneously performing instance segmentation, semantic segmentation, and panoptic segmentation tasks.
Unsupervised Video Domain Adaptation with Masked Pre-Training and Collaborative Self-Training
Arun Reddy (Johns Hopkins University), Rama Chellappa (Johns Hopkins University)
Domain AdaptationTransformerContrastive LearningVideo
🎯 What it does: This paper proposes the UNITE framework, which utilizes CLIP as a spatial teacher for mask-free unsupervised pre-training and collaborative self-training on target domain videos, achieving unsupervised video domain adaptation.
Unveiling Parts Beyond Objects: Towards Finer-Granularity Referring Expression Segmentation
Wenxuan Wang (Institute of Automation, Chinese Academy of Sciences), Jing Liu (Institute of Automation, Chinese Academy of Sciences)
Object DetectionSegmentationTransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: This paper proposes a Multi-Granularity Referring Expression Segmentation task (MRES), constructs the RefCOCOm evaluation benchmark, releases the largest-scale 32M visual localization dataset, and introduces a unified UniRES model to achieve object and part-level referring segmentation.
Unveiling the Power of Audio-Visual Early Fusion Transformers with Dense Interactions through Masked Modeling
Shentong Mo (Carnegie Mellon University), Pedro Morgado (University of Wisconsin Madison)
ClassificationSegmentationTransformerAuto EncoderMultimodalityAudio
🎯 What it does: A deep fusion audio-visual Transformer is proposed, utilizing learnable fusion tokens to achieve early fusion through dense local interactions, and is pretrained under a self-supervised masked reconstruction framework.
Unveiling the Unknown: Unleashing the Power of Unknown to Known in Open-Set Source-Free Domain Adaptation
Fuli Wan (Xidian University), Cheng Deng (Xidian University)
Domain AdaptationContrastive LearningImage
🎯 What it does: This paper proposes an open-set source unsupervised domain adaptation framework in which source data is unavailable and the target domain contains unknown classes. It utilizes an unknown diffuser to actively mine unknown classes in the target domain from a source pre-trained model, thereby achieving a bidirectional enhancement of knowledge transfer for known classes and generalization for unknown classes.
Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution
Shangchen Zhou (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
RestorationSuper ResolutionDiffusion modelOptical FlowVideo
🎯 What it does: A text-guided latent diffusion model called Upscale-A-Video has been developed for achieving video super-resolution while maintaining detail reproduction and temporal consistency.
URHand: Universal Relightable Hands
Zhaoxi Chen (Meta), Shunsuke Saito (Meta)
RestorationGenerationGenerative Adversarial NetworkImage
🎯 What it does: A general relightable model has been trained to render hands in real-time from any perspective, pose, lighting, and identity, supporting quick personalization from mobile scans.
USE: Universal Segment Embeddings for Open-Vocabulary Image Segmentation
Xiaoqi Wang (Bosch Research North America), Liu Ren (Bosch Research North America)
SegmentationRetrievalLarge Language ModelContrastive LearningImageText
🎯 What it does: A vocabulary-free image segmentation framework based on Universal Sentence Embedding (USE) is proposed, which includes a data pipeline for automated multi-granularity image-text segment pair generation and a lightweight segment embedding model, capable of achieving zero-shot semantic/component classification for segmentation regions of any granularity.
Using Human Feedback to Fine-tune Diffusion Models without Any Reward Model
Kai Yang (Tsinghua University), Xiu Li (Tsinghua University)
GenerationData SynthesisCompressionReinforcement LearningDiffusion modelImage
🎯 What it does: A D3PO method is proposed, which directly utilizes human preferences to fine-tune diffusion models without a reward model.
Utility-Fairness Trade-Offs and How to Find Them
Sepehr Dehdashtian (Michigan State University), Vishnu Naresh Boddeti (Michigan State University)
ClassificationRepresentation LearningConvolutional Neural NetworkContrastive LearningImageTabular
🎯 What it does: This study investigates the inherent trade-off between fairness and effectiveness in machine learning models, proposing two new frameworks: Data Space Trade-off (DST) and Label Space Trade-off (LST). Subsequently, the U-FaTE method is designed to numerically estimate these two trade-offs from the data.
UV-IDM: Identity-Conditioned Latent Diffusion Model for Face UV-Texture Generation
Hong Li (Beihang University), Baochang Zhang (Beihang University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a latent diffusion model based on identity conditions (UV-IDM) that can quickly generate high-fidelity, identity-consistent UV textures from a single wild image for 3D face reconstruction.
UVEB: A Large-scale Benchmark and Baseline Towards Real-World Underwater Video Enhancement
Yaofeng Xie (Ocean University of China), Bing Zheng (Ocean University of China)
RestorationConvolutional Neural NetworkVideoBenchmark
🎯 What it does: The first large-scale high-resolution underwater video enhancement benchmark UVEB has been constructed, and the first supervised underwater video enhancement network UVE-Net has been proposed.
V?: Guided Visual Search as a Core Mechanism in Multimodal LLMs
Penghao Wu (University of California San Diego), Saining Xie (New York University)
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Proposes the SEAL framework, embedding an LLM-guided visual search mechanism V* in multimodal LLMs to achieve active localization and fine-grained questioning of targets in high-resolution images.
VA3: Virtually Assured Amplification Attack on Probabilistic Copyright Protection for Text-to-Image Generative Models
Xiang Li (National University of Singapore), Kenji Kawaguchi (National University of Singapore)
GenerationAdversarial AttackPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper proposes an online 'Virtual Assurance Amplification Attack' (VA3), which significantly increases the probability of generating infringing images by text-to-image generation models under probabilistic copyright protection through multiple interactions, and provides an adversarial prompt optimization algorithm called Anti-NAF for NAF protection.
Validating Privacy-Preserving Face Recognition under a Minimum Assumption
Hui Zhang (Anhui University), Xuejun Li (Anhui University)
RecognitionSafty and PrivacyGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a privacy verification method called Map V based on minimal assumptions (1k1c), which utilizes deep image priors and zero-order gradient estimation to attack privacy-protecting facial recognition systems with only a limited number of queries.
Vanishing-Point-Guided Video Semantic Segmentation of Driving Scenes
Diandian Guo (ETH Zurich), Luc Van Gool (ETH Zurich)
SegmentationAutonomous DrivingTransformerVideo
🎯 What it does: A driving scene video semantic segmentation network called VPSeg is proposed, which is based on vanishing point (VP) priors. It achieves precise mining of cross-frame correspondences and long-range fine-grained features through two modules, MotionVP and DenseVP, within a context-detail framework.
VAREN: Very Accurate and Realistic Equine Network
Silvia Zuffi (IMATI-CNR), Michael J. Black (Max Planck Institute for Intelligent Systems)
GenerationData SynthesisPose EstimationPoint CloudMesh
🎯 What it does: A novel 3D horse posture parameter model called VAREN has been developed based on real 3D scans, capable of capturing muscle deformation.
VastGaussian: Vast 3D Gaussians for Large Scene Reconstruction
Jiaqi Lin (Tsinghua University), Wenming Yang (Tsinghua University)
Gaussian SplattingPoint Cloud
🎯 What it does: To achieve high-quality reconstruction and real-time rendering for large-scale scenes, a method called VastGaussian based on 3D Gaussian splatting is proposed.
VBench: Comprehensive Benchmark Suite for Video Generative Models
Ziqi Huang (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationData SynthesisTransformerPrompt EngineeringOptical FlowVideoBenchmark
🎯 What it does: This paper presents VBench, a comprehensive evaluation benchmark for video generation models, refined into 16 separable dimensions with dedicated prompts and evaluation methods.
VCoder: Versatile Vision Encoders for Multimodal Large Language Models
Jitesh Jain (SHI Labs), Humphrey Shi (Picsart AI Research)
Object DetectionSegmentationDepth EstimationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: To address the shortcomings of multimodal large language models (MLLM) in object perception and counting tasks, the VCoder controller is proposed, and the COCO Segmentation Text (COST) dataset is constructed for training and evaluating the object recognition, counting, and depth order perception capabilities of MLLMs.
VecFusion: Vector Font Generation with Diffusion
Vikas Thamizharasan (University of Massachusetts Amherst), Evangelos Kalogerakis (University of Massachusetts Amherst)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: This paper proposes a two-stage diffusion model (first generating low-resolution raster fonts and then generating vector paths) to automatically synthesize complete vector fonts, supporting tasks such as missing character completion, few-shot style transfer, and font interpolation.
Vector Graphics Generation via Mutually Impulsed Dual-domain Diffusion
Zhongyin Zhao (Shanghai Jiao Tong University), Bingbing Ni (Shanghai Jiao Tong University)
GenerationTransformerDiffusion modelImage
🎯 What it does: A dual-domain (vector-pixel) diffusion framework is proposed, which synchronously generates high-quality vector graphics in both domains by utilizing mutually stimulating conditional information.
Versatile Medical Image Segmentation Learned from Multi-Source Datasets via Model Self-Disambiguation
Xiaoyang Chen (University of Pennsylvania), Yong Fan (University of Pennsylvania)
SegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A weakly supervised medical image segmentation method is proposed, which trains a single general model using multi-source partial and sparse labeled data.
Versatile Navigation Under Partial Observability via Value-guided Diffusion Policy
Gengyu Zhang (Illinois Institute of Technology), Yan Yan (Illinois Institute of Technology)
OptimizationRobotic IntelligenceReinforcement LearningDiffusion modelPoint Cloud
🎯 What it does: A value-guided diffusion strategy has been designed for long-term path planning in partially observable environments, capable of operating in both 2D mazes and 3D real indoor scenes.
VGGSfM: Visual Geometry Grounded Deep Structure From Motion
Jianyuan Wang (University of Oxford), David Novotny (Meta AI)
Pose EstimationOptimizationTransformerSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: A fully differentiable structure-from-motion (SfM) pipeline called VGGSfM is proposed and implemented, which can directly track points from unordered image collections, jointly predict camera poses and 3D point clouds, and further optimize through differentiable beam adjustment.
VicTR: Video-conditioned Text Representations for Activity Recognition
Kumara Kahatapitiya (Stony Brook University), Michael S. Ryoo (Google Research)
RecognitionTransformerVision Language ModelContrastive LearningVideoText
🎯 What it does: A video understanding model named VicTR is proposed, which learns video-specific text representations based on image-text pre-training models (such as CLIP) to achieve video action recognition.
vid-TLDR: Training Free Token Merging for Light-weight Video Transformer
Joonmyung Choi (Korea University), Hyunwoo J. Kim (Korea University)
RetrievalComputational EfficiencyTransformerVideoTextMultimodality
🎯 What it does: Without the need for additional training, vid-TLDR is proposed to perform early token merging on video Transformers to reduce computational costs and improve performance.
Video Frame Interpolation via Direct Synthesis with the Event-based Reference
Yuhan Liu (Beijing University of Technology), Zhen Yang (Beijing University of Technology)
Image TranslationRestorationData SynthesisTransformerOptical FlowVideo
🎯 What it does: This paper proposes a direct video frame interpolation framework based on event cameras: first, the structure reference of the intermediate frame is reconstructed using event data; then, key frame features are aligned through bidirectional E-PCD; finally, a Transformer is used to refine the coarse interpolation results.
Video Harmonization with Triplet Spatio-Temporal Variation Patterns
Zonghui Guo (Institute of Computing Technology, Chinese Academy of Sciences), Haiyong Zheng (Ocean University of China)
Image HarmonizationRestorationTransformerVideo
🎯 What it does: A Video Triplet Transformer (VTT) framework is proposed for video and video tasks (such as video fusion, video enhancement, and video de-moiré) to achieve visual consistency and temporal consistency in videos by modeling three types of spatiotemporal variation patterns (short-term spatial, long-term global, and long-term dynamic).
Video Interpolation with Diffusion Models
Siddhant Jain (Google Research), Janne Kontkanen (Google Research)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: A video frame interpolation method based on a cascaded diffusion model (VIDIM) is proposed, capable of generating high-quality intermediate frames and complete videos given the start and end frames.
Video Prediction by Modeling Videos as Continuous Multi-Dimensional Processes
Gaurav Shrivastava (University of Maryland), Abhinav Shrivastava (University of Maryland)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: A diffusion model is proposed that treats videos as a continuous multi-dimensional process to improve video prediction tasks;
Video ReCap: Recursive Captioning of Hour-Long Videos
Md Mohaiminul Islam (University of North Carolina at Chapel Hill), Gedas Bertasius (Meta AI)
GenerationTransformerLarge Language ModelVideoText
🎯 What it does: A recursive video subtitle generation model, Video ReCap, has been developed to generate multi-level subtitles for long videos ranging from seconds to hours.
Video Recognition in Portrait Mode
Mingfei Han (Bytedance), Heng Wang (Bytedance)
RecognitionConvolutional Neural NetworkTransformerVideoMultimodalityAudio
🎯 What it does: The first fine-grained video dataset focused on portrait mode, PortraitMode-400, has been constructed, and cross-mode evaluation has been conducted on it.
Video Super-Resolution Transformer with Masked Inter&Intra-Frame Attention
Xingyu Zhou (University of Electronic Science and Technology of China), Shuhang Gu (University of Electronic Science and Technology of China)
RestorationSuper ResolutionTransformerVideo
🎯 What it does: We propose MIA-VSR, a Transformer-based video super-resolution model that utilizes feature-level masking and inter-frame attention, capable of significantly reducing FLOPs and memory usage while maintaining or improving PSNR.
Video-Based Human Pose Regression via Decoupled Space-Time Aggregation
Jijie He (Zhejiang Gongshang University), Wenwu Yang (Zhejiang Gongshang University)
Pose EstimationConvolutional Neural NetworkTransformerVideo
🎯 What it does: This paper proposes a regression-based multi-frame human pose estimation framework called DSTA, which can directly regress keypoint coordinates from video frames.
Video-P2P: Video Editing with Cross-attention Control
Shaoteng Liu (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelVideo
🎯 What it does: Proposes a Video-P2P framework that implements text-driven editing of real videos using cross-attention control.