ICCV 2023 Papers — Page 3
IEEE/CVF International Conference on Computer Vision · 2156 papers
Bootstrap Motion Forecasting With Self-Consistent Constraints
Maosheng Ye (Hong Kong University of Science and Technology), Qifeng Chen (Hong Kong University of Science and Technology)
Autonomous DrivingKnowledge DistillationConvolutional Neural NetworkTime Series
🎯 What it does: A dual consistency framework based on self-supervised consistency constraints (MISC) is proposed for vehicle motion trajectory prediction.
Borrowing Knowledge From Pre-trained Language Model: A New Data-efficient Visual Learning Paradigm
Wenxuan Ma (Beijing Institute of Technology), Gao Huang (Tsinghua University)
ClassificationDomain AdaptationData-Centric LearningTransformerLarge Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes a method called BorLan that utilizes knowledge from pre-trained language models to enhance the learning efficiency of visual models in data-scarce tasks.
Both Diverse and Realism Matter: Physical Attribute and Style Alignment for Rainy Image Generation
Changfeng Yu (Huazhong University of Science and Technology), Luxin Yan (Fudan University)
RestorationGenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a Physical Alignment and Controllable Generation Network (PCGNet) that achieves controllable manipulation of raindrop attributes and generates diverse, realistic rainy day images.
Boundary-Aware Divide and Conquer: A Diffusion-Based Solution for Unsupervised Shadow Removal
Lanqing Guo (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
RestorationDiffusion modelImageVideo
🎯 What it does: An unsupervised shadow removal method based on diffusion models is proposed, utilizing the distinction between shadow, boundary, and non-shadow regions, combined with illumination consistency constraints and reflection preservation techniques, to achieve seamless shadow removal from a single shadow image.
Box-based Refinement for Weakly Supervised and Unsupervised Localization Tasks
Eyal Gomel (Tel Aviv University), Lior Wolf (Tel Aviv University)
RecognitionObject DetectionTransformerContrastive LearningImageMultimodality
🎯 What it does: By training a detector on the outputs of original localization networks (such as DINO and CLIP visual encoders) and using the bounding boxes generated by the detector to refine the original network, improvements in localization are achieved under unsupervised and weakly supervised conditions.
BoxDiff: Text-to-Image Synthesis with Training-Free Box-Constrained Diffusion
Jinheng Xie (National University of Singapore), Mike Zheng Shou (National University of Singapore)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: A training-free method named BoxDiff is proposed for synthesizing images based on spatial conditions provided by users (such as boxes or doodles).
BoxSnake: Polygonal Instance Segmentation with Box Supervision
Rui Yang (Tsinghua University), Xiu Li (Tsinghua University)
Object DetectionSegmentationTransformerImage
🎯 What it does: A BoxSnake method is proposed, which achieves end-to-end polygon instance segmentation using only box annotations;
Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of Synthetic and Compositional Images
Nitzan Bitton-Guetta (Ben Gurion University of the Negev), Roy Schwartz (Hebrew University of Jerusalem)
GenerationData SynthesisAnomaly DetectionTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: A synthetic image dataset called WHOOPS! was constructed, which is artificially designed and violates common sense, and four visual language tasks were proposed for it (anomaly explanation generation, image description, cross-modal matching, visual question answering).
Breaking Temporal Consistency: Generating Video Universal Adversarial Perturbations Using Image Models
Hee-Seon Kim (Korea Advanced Institute of Science and Technology), Changick Kim (Korea Advanced Institute of Science and Technology)
Adversarial AttackConvolutional Neural NetworkImageVideo
🎯 What it does: This paper proposes a framework for generating universal adversarial perturbations for videos based on image models and image data (BTC-UAP), utilizing image classification models for adversarial optimization on each frame, and attacking video models by minimizing the feature similarity between adjacent frames.
Breaking The Limits of Text-conditioned 3D Motion Synthesis with Elaborative Descriptions
Yijun Qian (META AI), Jungdam Won (Seoul National University)
GenerationData SynthesisTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: Proposes a two-stage text-based 3D motion synthesis model that can generate long and complex natural motion sequences based on detailed natural language descriptions.
Bridging Cross-task Protocol Inconsistency for Distillation in Dense Object Detection
Longrong Yang (Zhejiang University), Xi Li (Zhejiang University)
Object DetectionKnowledge DistillationImage
🎯 What it does: A knowledge distillation framework for dense object detection is proposed to address the issue of inconsistent cross-task protocols, with a binary classification distillation loss and an IoU localization distillation loss designed separately.
Bridging Vision and Language Encoders: Parameter-Efficient Tuning for Referring Image Segmentation
Zunnan Xu (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
Object DetectionSegmentationTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This study proposes a parameter-efficient tuning framework based on a dual-stream vision-language model—Bridger—for reference image segmentation tasks.
Bring Clipart to Life
Nanxuan Zhao (Adobe Research), Nan Cao (Tongji University)
Image TranslationGenerationDomain AdaptationGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: ClipFaceShop proposes a clipart-based facial photo editing method that can accurately transfer abstract clipart facial attributes (such as hairstyle, expression, beard, etc.) to real photos while maintaining the identity of the portrait.
BT^2: Backward-compatible Training with Basis Transformation
Yifei Zhou (University of California), Ser-Nam Lim (Meta AI)
ClassificationRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningImageMultimodality
🎯 What it does: The BT 2 method is proposed, which adds necessary dimensions through learnable basis transformations in backward-compatible training, maintaining the performance of the new model while being compatible with the old model.
Building a Winning Team: Selecting Source Model Ensembles using a Submodular Transferability Estimation Approach
Vimal K B (Indian Institute of Technology Hyderabad), Vineeth N Balasubramanian (Indian Institute of Technology Hyderabad)
ClassificationSegmentationDomain AdaptationOptimizationConvolutional Neural NetworkTabular
🎯 What it does: This paper proposes a transferability estimation metric called OSBORN for selecting a source model set to assist in constructing the optimal model ensemble for the target task.
Building Bridge Across the Time: Disruption and Restoration of Murals In the Wild
Huiyang Shao (University of Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
RestorationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a mural damage simulation framework based on Blender and the corresponding Attention Diffusion Framework (ADF) for the automatic detection and repair of mural damage.
Building Vision Transformers with Hierarchy Aware Feature Aggregation
Yongjie Chen (University of Science and Technology Beijing), Bin Fan (University of Science and Technology Beijing)
ClassificationObject DetectionSegmentationTransformerSupervised Fine-TuningImage
🎯 What it does: A Hierarchical Aware Feature Aggregation (HAFA) framework is proposed, which enhances local detail capture using a Local Adaptive Aggregation (LAA) module in the shallow layers, and improves global relationship modeling through semantic clustering and merging using a Semantic Information Aggregation (SIA) module in the deep layers.
Building3D: A Urban-Scale Dataset and Benchmarks for Learning Roof Structures from Point Clouds
Ruisheng Wang (University of Calgary), Hongxin Yang (University of Calgary)
Object DetectionSegmentationGenerationTransformerSupervised Fine-TuningPoint CloudMeshBenchmark
🎯 What it does: A city-level LiDAR point cloud dataset called Building3D has been constructed, which includes 16 Estonian cities and approximately 160,000 buildings, along with corresponding mesh and wireframe models.
BUS: Efficient and Effective Vision-Language Pre-Training with Bottom-Up Patch Summarization.
Chaoya Jiang (Peking University), Songfang Huang (Alibaba Group)
RetrievalCompressionTransformerVision Language ModelImageTextMultimodality
🎯 What it does: A Bottom-Up Patch Summarization (BUS) framework is proposed for efficiently and effectively compressing long visual token sequences in Vision-Language Pre-Training (VLP), achieving high performance on multimodal tasks.
C2F2NeUS: Cascade Cost Frustum Fusion for High Fidelity and Generalizable Neural Surface Reconstruction
Luoyuan Xu (Huazhong University of Science and Technology), Wei Yang (Huazhong University of Science and Technology)
Neural Radiance FieldPoint CloudMesh
🎯 What it does: Combining Multi-View Stereo (MVS) with Neural Implicit Surfaces (NIS) to achieve end-to-end high-precision 3D reconstruction for sparse views.
C2ST: Cross-Modal Contextualized Sequence Transduction for Continuous Sign Language Recognition
Huaiwen Zhang (Inner Mongolia University), De Hu (Inner Mongolia University)
RecognitionTransformerLarge Language ModelVideoTextMultimodality
🎯 What it does: A cross-modal contextual sequence transduction framework (C²ST) is designed for continuous sign language recognition, enhancing the ability to capture the semantics and context of sign language by embedding a sign language gloss language model into the video feature extraction and sequence decoding process.
CAD-Estate: Large-scale CAD Model Annotation in RGB Videos
Kevis-Kokitsi Maninis (Google Research), Vittorio Ferrari (Google Research)
Object DetectionObject TrackingPose EstimationRetrievalOptimizationSimultaneous Localization and MappingVideoMesh
🎯 What it does: This paper proposes a semi-automated workflow that combines RGB videos with a CAD model database to generate globally consistent 9-DoF CAD model pose annotations for multi-object scenes in videos, constructing the CAD-Estate dataset with a scale of 20k videos, 101k instances, and 12k independent CAD models.
CAFA: Class-Aware Feature Alignment for Test-Time Adaptation
Sanghun Jung (University of Washington), Jaegul Choo (KAIST AI)
Domain AdaptationContrastive LearningImage
🎯 What it does: A test-time adaptive method CAFA based on class-aware feature alignment is proposed, which utilizes the pre-computed source class Gaussian distribution and Mahalanobis distance to align features by simultaneously reducing intra-class distance and increasing inter-class distance under unsupervised data conditions.
Calibrating Panoramic Depth Estimation for Practical Localization and Mapping
Junho Kim (Seoul National University), Young Min Kim (Seoul National University)
Depth EstimationDomain AdaptationRobotic IntelligenceConvolutional Neural NetworkSimultaneous Localization and MappingPoint Cloud
🎯 What it does: In response to domain shift in panoramic depth estimation models in deployment environments, a geometry-consistent test-time self-supervised calibration method is proposed, which can quickly optimize the network both online and offline.
Calibrating Uncertainty for Semi-Supervised Crowd Counting
Chen LI, Chao Chen (Stony Brook University)
RecognitionObject DetectionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a supervised uncertainty calibration framework to select reliable pseudo-labels for semi-supervised crowd counting tasks. By directly supervising the uncertainty branch on labeled images and combining the accumulated spatial matching distance (ASM) based on point set matching as a proxy function, it achieves precise uncertainty assessment for each image block; subsequently, the low uncertainty prediction results are used as pseudo-labels, and the teacher model (EMA weights) continuously enhances the counting performance of the student model.
CAME: Contrastive Automated Model Evaluation
Ru Peng (Zhejiang University), Junbo Zhao (Zhejiang University)
ClassificationData SynthesisDomain AdaptationContrastive LearningImage
🎯 What it does: Proposes to estimate the classification accuracy of the model on the test set by calculating the contrastive loss without a training set.
Camera-Driven Representation Learning for Unsupervised Domain Adaptive Person Re-identification
Geon Lee (Yonsei University), Bumsub Ham (Yonsei University)
RecognitionDomain AdaptationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposes a camera-driven curriculum learning and camera diversity loss-based unsupervised domain adaptation method for pedestrian re-identification.
Can Language Models Learn to Listen?
Evonne Ng (University of California), Shiry Ginosar (University of California)
GenerationTransformerLarge Language ModelAuto EncoderVideoTextMultimodality
🎯 What it does: Based on a large language model (GPT-2) and VQ-VAE, this system converts the text of the speaker in a conversation into a three-dimensional facial motion sequence of the listener, achieving the generation of non-verbal feedback from the listener based solely on text.
CancerUniT: Towards a Single Unified Model for Effective Detection, Segmentation, and Diagnosis of Eight Major Cancers Using a Large Collection of CT Scans
Jieneng Chen (DAMO Academy Alibaba Group), Ling Zhang (DAMO Academy Alibaba Group)
ClassificationObject DetectionSegmentationTransformerImageBiomedical DataComputed Tomography
🎯 What it does: A single unified Transformer model called CancerUniT is proposed, capable of simultaneously detecting, segmenting, and diagnosing eight major types of cancer and non-cancerous tumors in CT scans.
Candidate-aware Selective Disambiguation Based On Normalized Entropy for Instance-dependent Partial-label Learning
Shuo He (University of Electronic Science and Technology of China), Lei Feng (Nanyang Technological University)
ClassificationImage
🎯 What it does: A two-stage Instance-Dependent Partial Label Learning (ID-PLL) framework is proposed, which first selects partially well-identified samples through normalized entropy and uses complementary label supervision to train the network; subsequently, it dynamically filters the remaining samples using class-aware thresholds and adaptive thresholds, incorporating additional identified and partially reliable unrecognized samples into the training.
Canonical Factors for Hybrid Neural Fields
Brent Yi (University of California Berkeley), Yi Ma (University of California Berkeley)
OptimizationRepresentation LearningNeural Radiance FieldImage
🎯 What it does: This paper proposes and implements a tomographic feature volume architecture (TILTED) that eliminates biases caused by axis alignment through learned transformations, and applies it to hybrid neural field models.
CaPhy: Capturing Physical Properties for Animatable Human Avatars
Zhaoqi Su (Tsinghua University), Yebin Liu (Tsinghua University)
GenerationOptimizationRecurrent Neural NetworkMesh
🎯 What it does: Using limited 3D scanning data, combined with physical constraints and 3D supervision, a neural cloth deformation model is trained to generalize to new poses, achieving animatable digital avatars of the human body.
Cascade-DETR: Delving into High-Quality Universal Object Detection
Mingqiao Ye (ETH Zurich), Fisher Yu (ETH Zurich)
Object DetectionDomain AdaptationTransformerImageBenchmark
🎯 What it does: This paper proposes Cascade-DETR, a model that achieves high-quality cross-domain object detection through cascaded attention and IoU prediction recalibration based on DETR.
CASSPR: Cross Attention Single Scan Place Recognition
Yan Xia (Technical University of Munich), Daniel Cremers (Technical University of Munich)
RecognitionRetrievalTransformerSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper proposes CASSPR, a cross-attention transformer that combines point clouds and sparse voxels for scene recognition from single-frame LiDAR point clouds.
Category-aware Allocation Transformer for Weakly Supervised Object Localization
Zhiwei Chen (Xiamen University), Rongrong Ji (Xiamen University)
Object DetectionTransformerImage
🎯 What it does: Utilizing the self-attention mechanism of Transformer to achieve weakly supervised object localization, the CATR framework is proposed to enhance localization accuracy through category awareness.
Causal-DFQ: Causality Guided Data-Free Network Quantization
Yuzhang Shang (Illinois Institute of Technology), Yan Yan (Illinois Institute of Technology)
Object DetectionContrastive LearningImage
🎯 What it does: Proposes a data-free network quantization method based on causal reasoning called Causal-DFQ.
CauSSL: Causality-inspired Semi-supervised Learning for Medical Image Segmentation
Juzheng Miao (Chinese University of Hong Kong), Pheng-Ann Heng (Chinese University of Hong Kong)
SegmentationConvolutional Neural NetworkImageBiomedical DataComputed Tomography
🎯 What it does: A semi-supervised segmentation method for medical imaging based on causal graphs, CauSSL, is proposed, which enhances model performance through algorithm independence.
CBA: Improving Online Continual Learning via Continual Bias Adaptor
Quanziang Wang (Xi'an Jiaotong University), Deyu Meng (Macau University of Science and Technology)
ClassificationOptimizationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: The Continual Bias Adaptor (CBA) module is proposed, which dynamically compensates for catastrophic distribution drift through augmented classifiers and dual-layer optimization in online continual learning, thereby alleviating forgetting.
CC3D: Layout-Conditioned Generation of Compositional 3D Scenes
Sherwin Bahmani (University of Toronto), Andrea Tagliasacchi (University of Toronto)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A 3D Generative Adversarial Network (CC3D) based on 2D semantic layout conditions is proposed, capable of generating perspective-consistent 3D renderings of multi-object scenes.
CDAC: Cross-domain Attention Consistency in Transformer for Domain Adaptive Semantic Segmentation
Kaihong Wang (Boston University), Margrit Betke (Boston University)
SegmentationDomain AdaptationTransformerImage
🎯 What it does: A Cross-Domain Attention Consistency (CDAC) method is proposed, utilizing the self-attention and cross-domain attention of Transformers for domain-adaptive semantic segmentation.
CDFSL-V: Cross-Domain Few-Shot Learning for Videos
Sarinda Samarasinghe (University of Central Florida), Mubarak Shah (University of Central Florida)
ClassificationRecognitionDomain AdaptationAuto EncoderContrastive LearningVideo
🎯 What it does: A cross-domain few-shot video action recognition method is proposed, which achieves feature balance between the source domain and the target domain by combining self-supervised pre-training and curriculum learning.
CDUL: CLIP-Driven Unsupervised Learning for Multi-Label Image Classification
Rabab Abdelfattah (University of Southern Mississippi), Song Wang (University of South Carolina)
ClassificationContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised multi-label image classification framework based on CLIP, which first generates pseudo-labels through global and local similarity aggregation, and then trains the classifier using a gradient alignment method.
Center-Based Decoupled Point-cloud Registration for 6D Object Pose Estimation
Haobo Jiang (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
Pose EstimationConvolutional Neural NetworkPoint Cloud
🎯 What it does: A decoupled point cloud registration framework based on object centers is proposed, estimating translation through center regression and predicting rotation using center-aligned point clouds, achieving 6D object pose estimation.
CFCG: Semi-Supervised Semantic Segmentation via Cross-Fusion and Contour Guidance Supervision
Shuo Li (Baidu Inc), Jingdong Wang (Baidu Inc)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: A semi-supervised semantic segmentation framework named CFCG is proposed, which combines cross-fusion supervision and adaptive contour guidance techniques.
CGBA: Curvature-aware Geometric Black-box Attack
Md Farhamdur Reza (North Carolina State University), Huaiyu Dai (North Carolina State University)
Adversarial AttackConvolutional Neural NetworkTransformerGaussian SplattingImage
🎯 What it does: Two decision boundary black-box attack methods, CGBA and CGBA-H, are proposed. They efficiently generate adversarial samples with a low query budget by searching for decision boundary points along a semicircular trajectory on a two-dimensional constrained plane, and provide a better initial boundary point selection scheme.
CHAMPAGNE: Learning Real-world Conversation from Large-Scale Web Videos
Seungju Han (Seoul National University), Youngjae Yu (Yonsei University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVideoTextMultimodality
🎯 What it does: A large-scale dataset of 18 million video dialogues, YTD-18M, was constructed, and a multimodal dialogue generation model, CHAMPAGNE, capable of integrating video, text, and titles was trained;
Chaotic World: A Large and Challenging Benchmark for Human Behavior Understanding in Chaotic Events
Kian Eng Ong (Singapore University of Technology and Design), Jun Liu (Singapore University of Technology and Design)
RecognitionObject DetectionSegmentationVideoTextMultimodalityBenchmarkAudio
🎯 What it does: The first large-scale multimodal dataset, Chaotic World, has been constructed to analyze human behavior in chaotic events, and a unified multitask model, IntelliCare, has been proposed.
ChartReader: A Unified Framework for Chart Derendering and Comprehension without Heuristic Rules
Zhi-Qi Cheng (Carnegie Mellon University), Alexander G. Hauptmann (Carnegie Mellon University)
TransformerVision Language ModelTabular
🎯 What it does: A unified framework called ChartReader is proposed, integrating three major tasks: chart rendering (Chart-to-Table) and chart understanding (ChartQA, Chart-to-Text), forming an end-to-end unstructured learning process.
Chasing Clouds: Differentiable Volumetric Rasterisation of Point Clouds as a Highly Efficient and Accurate Loss for Large-Scale Deformable 3D Registration
Mattias P. Heinrich (University of Luebeck), Lasse Hansen (EchoScout GmbH)
OptimizationGraph Neural NetworkPoint CloudComputed Tomography
🎯 What it does: Unsupervised and self-supervised registration of highly deformable 3D point clouds is performed, proposing a differentiable voxel rasterization loss that addresses the gradient sparsity and non-differentiability issues of traditional Chamfer/EMD in high-resolution and complex geometries.
CheckerPose: Progressive Dense Keypoint Localization for Object Pose Estimation with Graph Neural Network
Ruyi Lian (Stony Brook University), Haibin Ling (Stony Brook University)
Pose EstimationGraph Neural NetworkImage
🎯 What it does: Aiming at six degrees of freedom (6-DoF) object pose estimation from a single RGB image, the CheckerPose method is proposed.
ChildPlay: A New Benchmark for Understanding Children's Gaze Behaviour
Samy Tafasca (Idiap Research Institute), Jean-Marc Odobez (Idiap Research Institute)
RecognitionObject DetectionOptimizationData-Centric LearningConvolutional Neural NetworkSupervised Fine-TuningVideoBenchmark
🎯 What it does: This paper proposes a gaze target prediction task specifically for children, constructs the ChildPlay dataset, and designs a 3DFoV prediction model that utilizes geometrically consistent depth information, significantly improving gaze target recognition performance for children compared to adults.
Chinese Text Recognition with A Pre-Trained CLIP-Like Model Through Image-IDS Aligning
Haiyang Yu (Fudan University), Xiangyang Xue (Fudan University)
RecognitionTransformerContrastive LearningImageText
🎯 What it does: A two-stage framework is designed: first, a CLIP pre-trained model aligned with printed character images and Chinese character IDS is used to learn the normative representation of Chinese characters, and then this representation is applied to a text recognition model to achieve zero-shot recognition of Chinese text.
Chop & Learn: Recognizing and Generating Object-State Compositions
Nirat Saini (University of Maryland), Abhinav Shrivastava (University of Maryland)
RecognitionGenerationTransformerDiffusion modelImageVideo
🎯 What it does: This paper constructs a novel dataset called Chop & Learn and proposes two tasks aimed at unseen combinations: composite image generation and composite action recognition, with the goal of evaluating the model's zero-shot capabilities in object-state transfer and generation.
CHORD: Category-level Hand-held Object Reconstruction via Shape Deformation
Kailin Li (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
Object DetectionGenerationDepth EstimationGenerative Adversarial NetworkImagePoint Cloud
🎯 What it does: Research on 3D reconstruction of handheld objects at the category level
Chordal Averaging on Flag Manifolds and Its Applications
Nathan Mankovich (Colorado State University), Tolga Birdal (Imperial College London)
ClassificationOptimizationImage
🎯 What it does: This paper proposes a new, provably convergent algorithm for computing flag means and flag medians on the flag manifold, and applies it to computer vision tasks such as subspace statistics, illumination alignment, handwritten digit classification, PCA weighted averaging, and rigid motion averaging.
CHORUS : Learning Canonicalized 3D Human-Object Spatial Relations from Unbounded Synthesized Images
Sookwan Han (Seoul National University), Hanbyul Joo (Seoul National University)
Object DetectionData SynthesisPose EstimationPrompt EngineeringDiffusion modelImage
🎯 What it does: Learn the three-dimensional human-object interaction spatial relationships using a self-supervised approach by generating a large number of multi-view synthetic images;
Chupa: Carving 3D Clothed Humans from Skinned Shape Priors using 2D Diffusion Probabilistic Models
Byungjun Kim (Seoul National University), Hanbyul Joo (Seoul National University)
GenerationDiffusion modelMesh
🎯 What it does: A 3D human generation pipeline named Chupa is proposed, which combines diffusion models and neural rendering techniques to generate diverse and realistic 3D human digital avatars.
CIRI: Curricular Inactivation for Residue-aware One-shot Video Inpainting
Weiying Zheng (South China University of Technology), Shengfeng He (Singapore Management University)
RestorationSegmentationVideo
🎯 What it does: This study investigates a one-shot video inpainting framework that converts traditional video inpainting models to single annotation (only the first frame mask), addressing the problem of filling in missing areas in dynamic scenes.
CiT: Curation in Training for Effective Vision-Language Data
Hu Xu (Meta AI), Christoph Feichtenhofer (Meta AI)
Computational EfficiencyRepresentation LearningData-Centric LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes an algorithm for dynamic data selection during the training process—CiT (Curation in Training). By utilizing a pre-trained text encoder, it matches the metadata of the target task with a vast number of image-text pairs for similarity, thereby automatically filtering out more relevant data in the training loop, significantly improving data utilization efficiency.
CiteTracker: Correlating Image and Text for Visual Tracking
Xin Li (Peng Cheng Laboratory), Ming-Hsuan Yang (University of California Merced)
Object TrackingTransformerPrompt EngineeringImageTextMultimodality
🎯 What it does: Utilizing CLIP to generate textual descriptions of target images and modeling the correlation between dynamic textual features and the visual features of the search images, thereby achieving a visual tracking method based on image-text association—CiteTracker.
CL-MVSNet: Unsupervised Multi-View Stereo with Dual-Level Contrastive Learning
Kaiqiang Xiong (Peking University), Ronggang Wang (Peng Cheng Laboratory)
Depth EstimationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A dual-layer contrastive learning-based unsupervised multi-view stereo reconstruction method CL-MVSNet is proposed, addressing the issues of low-texture areas and viewpoint dependency.
Class Prior-Free Positive-Unlabeled Learning with Taylor Variational Loss for Hyperspectral Remote Sensing Imagery
Hengwei Zhao (Wuhan University), Yanfei Zhong (Wuhan University)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a method for positive and negative label (PU) learning without class priors, using Taylor variational loss and self-correcting optimization for hyperspectral remote sensing images.
Class-Aware Patch Embedding Adaptation for Few-Shot Image Classification
Fusheng Hao (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), Jun Cheng (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences)
ClassificationRepresentation LearningTransformerSupervised Fine-TuningImage
🎯 What it does: In few-shot image classification, image representation and similarity computation are improved by making the patch embeddings of ViT category-related and defining a dense similarity matrix.
Class-incremental Continual Learning for Instance Segmentation with Image-level Weak Supervision
Yu-Hsing Hsieh (National Taiwan University), Chu-Song Chen (National Taiwan University)
Object DetectionSegmentationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A class-incremental continual learning instance segmentation framework CL4WSIS based on image-level weak labels has been developed, which can gradually learn new categories while maintaining instance segmentation capabilities for old categories using only image labels.
Class-Incremental Grouping Network for Continual Audio-Visual Learning
Shentong Mo (Carnegie Mellon University), Yapeng Tian (University of Texas at Dallas)
Knowledge DistillationRepresentation LearningConvolutional Neural NetworkTransformerMultimodalityAudio
🎯 What it does: A continuous learning framework named Class-Incremental Grouping Network (CIGN) is proposed for category-level semantic representation learning of audio and visual inputs in a multi-task environment, achieving audio-visual source classification.
Class-relation Knowledge Distillation for Novel Class Discovery
Peiyan Gu (ShanghaiTech University), Xuming He (ShanghaiTech University)
Knowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A category relationship knowledge distillation framework based on the predicted distribution of known categories is proposed for new category discovery in unlabeled samples.
CleanCLIP: Mitigating Data Poisoning Attacks in Multimodal Contrastive Learning
Hritik Bansal (University of California Los Angeles), Kai-Wei Chang (University of California Los Angeles)
Representation LearningAdversarial AttackData-Centric LearningTransformerSupervised Fine-TuningContrastive LearningImageTextMultimodality
🎯 What it does: A framework named CleanCLIP is proposed for unsupervised fine-tuning, which can eliminate backdoors generated by data poisoning attacks in multimodal contrastive learning models (such as CLIP);
ClimateNeRF: Extreme Weather Synthesis in Neural Radiance Field
Yuan Li (University of Illinois Urbana-Champaign), Shenlong Wang (University of Illinois Urbana-Champaign)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: Combining the NeRF model with physical simulation to generate realistic 3D visualizations of climate change scenarios (such as floods, snowfall, and haze).
CLIP-Cluster: CLIP-Guided Attribute Hallucination for Face Clustering
Shuai Shen (Tsinghua University), Jiwen Lu (Tsinghua University)
ClassificationRecognitionGraph Neural NetworkTransformerContrastive LearningImage
🎯 What it does: This paper proposes a CLIP-based attribute hallucination framework (CLIP-Cluster) that generates features corresponding to various attributes (age, pose, expression) guided by text, and utilizes a neighbor-aware generative model to fuse these features to reduce attribute differences within the same identity, thereby achieving more compact facial clustering.
CLIP-Driven Universal Model for Organ Segmentation and Tumor Detection
Jie Liu (City University of Hong Kong), Zongwei Zhou (Johns Hopkins University)
Object DetectionSegmentationConvolutional Neural NetworkVision Language ModelContrastive LearningImageBiomedical DataComputed Tomography
🎯 What it does: A general model based on CLIP has been proposed and trained, capable of segmenting and detecting 25 organs and 6 types of tumors across various abdominal CT datasets, and it can handle partially annotated data.
CLIP2Point: Transfer CLIP to Point Cloud Classification with Image-Depth Pre-Training
Tianyu Huang (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
ClassificationContrastive LearningImagePoint Cloud
🎯 What it does: The CLIP2Point method is proposed, utilizing image-depth contrastive learning to pre-train a deep encoder, transferring CLIP's visual-text knowledge to 3D point cloud classification tasks.
CLIPascene: Scene Sketching with Different Types and Levels of Abstraction
Yael Vinker (Tel Aviv University), Ariel Shamir (Reichman University)
SegmentationGenerationTransformerVision Language ModelContrastive LearningImage
🎯 What it does: A two-stage MLP network trained individually for each image based on CLIP is proposed, generating vector scene sketches with controllable fidelity and simplicity, supporting foreground and background separation.
CLIPN for Zero-Shot OOD Detection: Teaching CLIP to Say No
Hualiang Wang (Hong Kong University of Science and Technology), Xiaomeng Li (Hong Kong University of Science and Technology)
ClassificationAnomaly DetectionTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: Proposes the CLIPN method, injecting the 'negation' logic of CLIP into zero-shot OOD detection;
CLIPTER: Looking at the Bigger Picture in Scene Text Recognition
Aviad Aberdam (Amazon Web Services), Ron Litman (Amazon Web Services)
RecognitionTransformerVision Language ModelImage
🎯 What it does: This paper proposes the CLIPTER framework, which integrates global scene context into existing scene text recognizers by merging image-level features extracted from frozen vision-language models (such as CLIP/BLIP) with local features of cropped text images through gated cross-attention.
CLIPTrans: Transferring Visual Knowledge with Pre-trained Models for Multimodal Machine Translation
Devaansh Gupta (Boston College), Donglai Wei (Boston College)
Image TranslationTransformerSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: The CLIPTrans framework is proposed, utilizing the pre-trained multimodal M-CLIP and multilingual mBART for two-stage training, transferring visual knowledge from image-text pre-training to text translation, supporting multimodal machine translation without image inference.
CLNeRF: Continual Learning Meets NeRF
Zhipeng Cai (Intel Labs), Matthias Müller
GenerationData SynthesisNeural Radiance FieldImageBenchmark
🎯 What it does: This paper proposes a continuous learning NeRF system called CLNeRF and creates a continuous learning dataset for evaluation named World Across Time (WAT).
Cloth2Body: Generating 3D Human Body Mesh from 2D Clothing
Lu Dai (Hong Kong University of Science and Technology), Hui Xiong (Hong Kong University of Science and Technology)
GenerationPose EstimationImageMesh
🎯 What it does: The Cloth2Body task is proposed: generating a matching 3D human mesh from a single 2D clothing image, supporting multiple poses and body types.
ClothesNet: An Information-Rich 3D Garment Model Repository with Simulated Clothes Environment
Bingyang Zhou (Harbin Institute of Technology), Lin Shao (Xi'an Jiaotong University)
ClassificationSegmentationRobotic IntelligenceConvolutional Neural NetworkReinforcement LearningPoint CloudMesh
🎯 What it does: This paper constructs the largest 3D clothing model dataset, ClothesNet (approximately 4,400 items), and provides rich annotations including categories, feature labels, boundary lines, and key points. It also implements a differentiable clothing simulation environment based on DiffClothAI for tasks such as robotic grasping, folding, hanging, and wearing.
ClothPose: A Real-world Benchmark for Visual Analysis of Garment Pose via An Indirect Recording Solution
Wenqiang Xu (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
Pose EstimationOptimizationVideoBenchmark
🎯 What it does: The GarmentTwin recording system and the ClothPose dataset are proposed to capture and label the dynamic poses of clothing in the real world through a method that combines physical simulation and optimization.
CLR: Channel-wise Lightweight Reprogramming for Continual Learning
Yunhao Ge (University of Southern California), Laurent Itti (Google Research)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Channel-wise Lightweight Reprogramming (CLR) method, which achieves task reshaping in continual learning by adding a minimal number of channel-level learnable convolutional kernels on a frozen shared CNN backbone, thereby avoiding catastrophic forgetting.
ClusT3: Information Invariant Test-Time Training
Gustavo A. Vargas Hakim (École de technologie supérieure), Christian Desrosiers (École de technologie supérieure)
Domain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised training method during testing based on information maximization, called ClusT3, to adapt models under domain shift.
Clusterformer: Cluster-based Transformer for 3D Object Detection in Point Clouds
Yu Pei (HikVision Research Institute), Shiliang Pu (HikVision Research Institute)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: We propose Clusterformer, a clustering-based Transformer query-style 3D object detector that directly executes the Transformer decoder on sparse voxel features without projecting onto the BEV plane.
Clustering based Point Cloud Representation Learning for 3D Analysis
Tuo Feng (University of Technology Sydney), Qinghua Zheng (Xi'an Jiaotong University)
Object DetectionSegmentationRepresentation LearningContrastive LearningPoint Cloud
🎯 What it does: A clustering-based supervised learning framework is proposed, utilizing online clustering within each category to discover potential subcategory patterns, which are used as auxiliary constraints to enhance point cloud representation learning.
Clutter Detection and Removal in 3D Scenes with View-Consistent Inpainting
Fangyin Wei (Princeton University), Szymon Rusinkiewicz (Princeton University)
RestorationObject DetectionSegmentationConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: An automated system is proposed that can detect and remove clutter from indoor RGB-D scans, and subsequently perform consistent geometric and texture inpainting from multiple viewpoints.
CMDA: Cross-Modality Domain Adaptation for Nighttime Semantic Segmentation
Ruihao Xia (East China University of Science and Technology), Yang Tang (East China University of Science and Technology)
SegmentationDomain AdaptationImageMultimodality
🎯 What it does: For nighttime semantic segmentation, an unsupervised cross-modal domain adaptation framework CMDA is proposed, utilizing information from images and event sensors to achieve the transfer of source domain daytime images to target domain nighttime images.
Co-Evolution of Pose and Mesh for 3D Human Body Estimation from Video
Yingxuan You (Peking University), Xia Li (ETH Zurich)
Pose EstimationRecurrent Neural NetworkTransformerVideoMesh
🎯 What it does: Using 2D pose sequences and image features from videos, we first estimate the 3D skeleton of the mid-frame, and then regress the 3D human mesh through an image-guided pose-mesh collaborative evolution network.
CO-Net: Learning Multiple Point Cloud Tasks at Once with A Cohesive Network
Tao Xie (Harbin Institute of Technology), Ruifeng Li (Harbin Institute of Technology)
ClassificationObject DetectionSegmentationNeural Architecture SearchPoint Cloud
🎯 What it does: Proposes CO-Net, which jointly learns multiple point cloud tasks (classification, segmentation, detection) to achieve parameter sharing and efficient storage.
CO-PILOT: Dynamic Top-Down Point Cloud with Conditional Neighborhood Aggregation for Multi-Gigapixel Histopathology Image Representation
Ramin Nakhli (University of British Columbia), Ali Bashashati (University of British Columbia)
ClassificationRepresentation LearningGraph Neural NetworkSupervised Fine-TuningImagePoint CloudBiomedical Data
🎯 What it does: This study presents CO-PILOT, a dynamic point cloud graph neural network based on cellular graphs, used to predict patient survival rates from multi-giga pixel H&E tissue slices;
Coarse-to-Fine Amodal Segmentation with Shape Prior
Jianxiong Gao (Fudan University), Yanwei Fu (Fudan University)
Object DetectionSegmentationConvolutional Neural NetworkTransformerGenerative Adversarial NetworkImageVideo
🎯 What it does: A coarse-to-fine amodal segmentation method called C2F-Seg is proposed, which first generates a rough complete mask in the vector quantization latent space using a transformer, and then refines it to obtain a fine mask using convolution.
Coarse-to-Fine: Learning Compact Discriminative Representation for Single-Stage Image Retrieval
Yunquan Zhu (YouTu Lab Tencent), Xing Sun (YouTu Lab Tencent)
RetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A Coarse-to-Fine framework CFCD is proposed, achieving single-stage image retrieval through adaptive MadaCos loss and triplet loss.
COCO-O: A Benchmark for Object Detectors under Natural Distribution Shifts
Xiaofeng Mao (Alibaba Group), Hui Xue (Zhejiang University)
Object DetectionConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: The COCO-O test set is proposed to evaluate the robustness of object detectors under natural distribution shifts, and systematic experiments are conducted on over 100 modern detectors using this dataset.
Coherent Event Guided Low-Light Video Enhancement
Jinxiu Liang (Peking University), Boxin Shi (Peking University)
RestorationConvolutional Neural NetworkOptical FlowVideoMultimodality
🎯 What it does: A deep learning framework is proposed that utilizes a combination of event cameras and low-light video inputs to achieve denoising and exposure compensation for low-light videos.
CoIn: Contrastive Instance Feature Mining for Outdoor 3D Object Detection with Very Limited Annotations
Qiming Xia (Xiamen University), Cheng Wang (Texas A&M University)
Object DetectionAutonomous DrivingContrastive LearningPoint Cloud
🎯 What it does: Under the condition of a very small amount of labeling (only 2%), the CoIn framework is proposed, which combines multi-class contrastive learning, instance feature mining, and label-to-pseudo-label contrastive learning to significantly improve 3D object detection performance.
CoinSeg: Contrast Inter- and Intra- Class Representations for Incremental Segmentation
Zekang Zhang, Yunchao Wei
SegmentationComputational EfficiencyConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a new method to address specific computer vision tasks, aiming to improve the performance and efficiency of the model.
Collaborative Propagation on Multiple Instance Graphs for 3D Instance Segmentation with Single-point Supervision
Shichao Dong (Nanyang Technological University), Guosheng Lin (Nanyang Technological University)
Object DetectionSegmentationConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper proposes a weakly supervised 3D instance segmentation method called RWSeg, which only requires labeling a single point on each object to complete the instance segmentation task.
Collaborative Tracking Learning for Frame-Rate-Insensitive Multi-Object Tracking
Yiheng Liu (ByteDance Inc), Yi Fu (ByteDance Inc)
Object TrackingTransformerVideo
🎯 What it does: This paper proposes a Collaborative Tracking Learning (ColTrack) framework that enhances multi-object tracking performance in low frame rate videos by using multiple historical queries to jointly track the same target.
Collecting The Puzzle Pieces: Disentangled Self-Driven Human Pose Transfer by Permuting Textures
Nannan Li (Boston University), Bryan A. Plummer (Boston University)
Image TranslationRestorationGenerationPose EstimationTransformerGenerative Adversarial NetworkImage
🎯 What it does: Achieving human pose transfer using a self-supervised approach. By dividing the input image into small patches and randomly permuting them, the pose information is removed, and cross-attention sampling is performed across three networks (source pose, texture, target pose) to ultimately generate a high-quality portrait under the target pose, with a background restoration network used to obtain a complete image.
Combating Noisy Labels with Sample Selection by Mining High-Discrepancy Examples
Xiaobo Xia (University of Sydney), Tongliang Liu (University of Sydney)
ClassificationData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes the CoDis method, which handles deep learning tasks with noisy labels in a robust manner by using dual network collaboration and selecting samples with significant differences in predicted probabilities during training.
Communication-efficient Federated Learning with Single-Step Synthetic Features Compressor for Faster Convergence
Yuhao Zhou (Sichuan University), Jiancheng Lv (Sichuan University)
Federated LearningComputational EfficiencyImage
🎯 What it does: A single-step synthetic feature compressor (3SFC) is proposed to achieve communication efficiency in federated learning with an extremely low compression ratio.
Communication-Efficient Vertical Federated Learning with Limited Overlapping Samples
Jingwei Sun (Duke University), Holger R. Roth (NVIDIA)
Federated LearningSupervised Fine-TuningImageTabularFinance Related
🎯 What it does: A one-shot and few-shot vertical federated learning framework is proposed, utilizing semi-supervised learning to address the issues of high communication costs and limited overlapping samples.