CVPR 2023 Papers — Page 24
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
vMAP: Vectorised Object Mapping for Neural Field SLAM
Xin Kong (Imperial College London), Andrew J. Davison (Imperial College London)
Object DetectionSegmentationOptimizationRobotic IntelligenceSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A real-time object-level neural field SLAM system called vMAP is proposed, which models each object individually using a small MLP and achieves efficient parallel optimization through vectorized training.
VNE: An Effective Method for Improving Deep Representation by Manipulating Eigenvalue Distribution
Jaeill Kim (Seoul National University), Wonjong Rhee (Seoul National University)
Domain AdaptationRepresentation LearningMeta LearningGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a regularization method based on Von Neumann entropy (VNE) to directly control the eigenvalue distribution of the autocorrelation matrix of deep model representations, thereby improving representation quality.
VolRecon: Volume Rendering of Signed Ray Distance Functions for Generalizable Multi-View Reconstruction
Yufan Ren (EPFL), Sabine Süsstrunk (ETH Zurich)
Depth EstimationTransformerNeural Radiance FieldImage
🎯 What it does: A cross-scene implicit reconstruction method called VolRecon based on Signed Ray Distance Function is proposed, utilizing projection features and global voxel features, and achieving detail-rich surface reconstruction through view Transformer and ray Transformer.
VoP: Text-Video Co-Operative Prompt Tuning for Cross-Modal Retrieval
Siteng Huang (Westlake University), Donglin Wang (Westlake University)
RetrievalTransformerPrompt EngineeringContrastive LearningVideoTextMultimodality
🎯 What it does: In the text-video retrieval task, parameter-efficient prompt tuning of the pre-trained CLIP model is performed, and three types of video-specific prompts (position, context, function) are added to the visual encoder, forming the VoP framework.
VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking
Yukang Chen (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
Object DetectionObject TrackingAutonomous DrivingPoint Cloud
🎯 What it does: A fully sparse voxel network VoxelNeXt is proposed for LiDAR 3D object detection and tracking, directly predicting objects based on sparse voxel features without the need for anchors or center proxies.
VoxFormer: Sparse Voxel Transformer for Camera-Based 3D Semantic Scene Completion
Yiming Li (New York University), Anima Anandkumar (California Institute of Technology)
SegmentationAutonomous DrivingTransformerPoint CloudBenchmark
🎯 What it does: This paper proposes VoxFormer, a two-stage camera-based 3D semantic scene completion framework that first generates sparse voxel queries using depth estimation and occupancy prediction, and then completes the entire 3D voxel grid through a sparse-to-dense Transformer similar to MAE.
VQACL: A Novel Visual Question Answering Continual Learning Setting
Xi Zhang (Chinese Academy of Sciences), Changsheng Xu (Tianjin University of Technology)
Representation LearningTransformerImageVideo
🎯 What it does: A new framework called VQACL (Visual Question Answering Continual Learning) is proposed, defining a dual-layer task sequence and introducing combinatorial testing; at the same time, a representation learning method based on sample-specific (SS) and sample-invariant (SI) features is designed, significantly enhancing the continual learning and combinatorial reasoning capabilities of VQA.
Watch or Listen: Robust Audio-Visual Speech Recognition With Visual Corruption Modeling and Reliability Scoring
Joanna Hong (Korea Advanced Institute of Science and Technology), Yong Man Ro (Korea Advanced Institute of Science and Technology)
RecognitionConvolutional Neural NetworkMultimodalityAudio
🎯 What it does: This paper proposes a robust multimodal speech recognition framework that can maintain robustness even when audio and video inputs are simultaneously affected by noise, occlusion, and other interferences;
Wavelet Diffusion Models Are Fast and Scalable Image Generators
Hao Phung (VinAI Research), Anh Tran (VinAI Research)
GenerationData SynthesisComputational EfficiencyDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A diffusion model based on discrete wavelet transform (Wavelet Diffusion) is designed, which performs low and high-frequency subband decomposition in both image and feature layers, and conducts reverse diffusion sampling on this basis, significantly improving training and inference speed while maintaining generation quality.
Weak-Shot Object Detection Through Mutual Knowledge Transfer
Xuanyi Du (Tencent), Chen Li (Tencent)
Object DetectionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A mutual knowledge transfer framework for weakly supervised object detection is designed, jointly training the Proposal Generator and the Multiple Instance Learning (MIL) module, and achieving bidirectional knowledge transfer through Knowledge Transfer Loss (KT loss) and Consistency Filtering (CF) to enhance the detection performance of the target dataset.
Weakly Supervised Class-Agnostic Motion Prediction for Autonomous Driving
Ruibo Li (Nanyang Technological University), Guosheng Lin (Nanyang Technological University)
Autonomous DrivingPoint Cloud
🎯 What it does: This paper proposes a weakly supervised class-agnostic motion prediction method based on foreground/background binary classification masks, utilizing a two-stage training approach to learn dynamic motion from a small number of annotated masks.
Weakly Supervised Monocular 3D Object Detection Using Multi-View Projection and Direction Consistency
Runzhou Tao (Beijing Institute of Technology), Jianbing Shen (University of Macau)
Object DetectionAutonomous DrivingImageBenchmark
🎯 What it does: A weakly supervised monocular 3D object detection method is proposed, which trains the model using only 2D bounding boxes and directional annotations to achieve 3D bounding box prediction.
Weakly Supervised Posture Mining for Fine-Grained Classification
Zhenchao Tang (Sun Yat-sen University), Calvin Yu-Chian Chen
ClassificationPose EstimationGraph Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A weakly supervised Pose Mining and Reverse Cross Entropy (PMRC) framework is proposed, which generates discriminative regions through a deep navigator, constructs a graph, and aggregates pose information using message passing for fine-grained classification.
Weakly Supervised Segmentation With Point Annotations for Histopathology Images via Contrast-Based Variational Model
Hongrun Zhang (University of Liverpool), Yalin Zheng (University of Liverpool)
SegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A variational model based on contrast mapping is proposed for weakly supervised segmentation of pathological images using only sparse point annotations, serving as supplementary supervision for deep segmentation networks.
Weakly Supervised Semantic Segmentation via Adversarial Learning of Classifier and Reconstructor
Hyeokjun Kweon (KAIST), Kuk-Jin Yoon (KAIST)
SegmentationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a weakly supervised semantic segmentation framework based on adversarial learning between a classifier and a reconstructor (ACR). It improves the quality of pseudo-labels by training the classifier to generate more accurate Class Activation Maps (CAM) and allowing the reconstructor to utilize residual information to reconstruct missing areas.
Weakly Supervised Temporal Sentence Grounding With Uncertainty-Guided Self-Training
Yifei Huang (University of Tokyo), Yoichi Sato (University of Tokyo)
RecognitionKnowledge DistillationData-Centric LearningVideoText
🎯 What it does: A weakly supervised temporal sentence localization method based on self-training is proposed, utilizing teacher-student cyclical consistency and uncertainty estimation to enhance proposal quality.
Weakly Supervised Video Emotion Detection and Prediction via Cross-Modal Temporal Erasing Network
Zhicheng Zhang (Nankai University), Jufeng Yang (Nankai University)
ClassificationRecognitionConvolutional Neural NetworkVideoMultimodalityAudio
🎯 What it does: This paper proposes a weakly supervised Cross-Modal Temporal Erasing Network to locate key frames and their contextual information in videos with only video-level emotion labels, thereby predicting the emotion categories of user-generated videos (UGV) more accurately.
Weakly Supervised Video Representation Learning With Unaligned Text for Sequential Videos
Sixun Dong (ShanghaiTech University), Shenghua Gao (Meituan)
RetrievalRepresentation LearningTransformerContrastive LearningVideoTextSequential
🎯 What it does: A weakly supervised sequential video representation learning framework is proposed, utilizing unaligned textual information for feature learning of videos.
Weakly-Supervised Domain Adaptive Semantic Segmentation With Prototypical Contrastive Learning
Anurag Das (MPI for Informatics), Bernt Schiele (MPI for Informatics)
SegmentationDomain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a general framework that utilizes weak labels from the target domain (image-level, point-level, rough annotations) along with source domain annotations for weakly supervised domain adaptive semantic segmentation, and reduces the domain gap through prototype alignment.
Weakly-Supervised Single-View Image Relighting
Renjiao Yi (National University of Defense Technology), Kai Xu (National University of Defense Technology)
Image TranslationRestorationImageVideo
🎯 What it does: An end-to-end method for weakly supervised single-view image inverse rendering and non-Lambertian reflection layers is proposed, achieving relighting and material editing of objects in a new scene from a single image.
WeatherStream: Light Transport Automation of Single Image Deweathering
Howard Zhang (University of California), Achuta Kadambi (University of California)
RestorationOptical FlowImageVideo
🎯 What it does: A time-sequenced multi-sampling dataset called WeatherStream has been constructed based on automated light transmission for training single-image de-weathering denoising models.
What Can Human Sketches Do for Object Detection?
Pinaki Nath Chowdhury (University of Surrey), Yi-Zhe Song (University of Surrey)
Object DetectionTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: A zero-shot object detection framework based on human-drawn sketches is proposed, capable of locating, recognizing, and focusing on specific instances or parts through user-drawn sketches.
What Happened 3 Seconds Ago? Inferring the Past With Thermal Imaging
Zitian Tang (Tsinghua University), Hang Zhao (Tsinghua University)
Pose EstimationConvolutional Neural NetworkVideoMultimodality
🎯 What it does: A method for inferring past human actions based on thermal pixels is proposed, and the first indoor thermal-visible synchronous dataset, Thermal-IM, is constructed.
What You Can Reconstruct From a Shadow
Ruoshi Liu (Columbia University), Carl Vondrick (Columbia University)
GenerationPose EstimationOptimizationGenerative Adversarial NetworkPoint Cloud
🎯 What it does: Utilize shadows to reconstruct the 3D shape of occluded objects, and jointly infer the light source position, pose, and shape through differentiable shadow rendering and a pre-trained generative model.
Where Is My Spot? Few-Shot Image Generation via Latent Subspace Optimization
Chenxi Zheng (South China University of Technology), Shengfeng He (Singapore Management University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Under the condition of very few samples, a latent subspace optimization method is proposed using the continuity and interpretability of the StyleGAN latent space, achieving high-quality and diverse image generation for unseen categories.
Where Is My Wallet? Modeling Object Proposal Sets for Egocentric Visual Query Localization
Mengmeng Xu (Meta AI), Juan-Manuel Pérez-Rúa (Meta AI)
Object DetectionObject TrackingTransformerVision Language ModelVideoTextMultimodality
🎯 What it does: A novel training strategy addressing viewpoint bias and task bias is proposed, and a Conditional Context Transformer (CocoFormer) is designed to achieve query-based object detection and localization, forming a complete visual query localization system.
Where We Are and What We're Looking At: Query Based Worldwide Image Geo-Localization Using Hierarchies and Scenes
Brandon Clark (University of Central Florida), Mubarak Shah (University of Central Florida)
TransformerImageBenchmark
🎯 What it does: A query-based multi-level geographic localization framework based on a Transformer decoder is proposed, which learns query vectors for different geographic levels and scenarios to directly predict S2 fine-grained locations.
Why Is the Winner the Best?
Matthias Eisenmann (German Cancer Research Center), Lena Maier-Hein (German Cancer Research Center)
Object DetectionSegmentationBiomedical DataReview/Survey PaperBenchmark
🎯 What it does: Through a questionnaire survey and statistical analysis of participants, organizers, and award-winning teams from 80 competitions at the 2021 ISBI and MICCAI conferences, this study systematically reviews competition practices, research progress, and award strategies.
Wide-Angle Rectification via Content-Aware Conformal Mapping
Qi Zhang (Tencent AI Lab), Qing Wang (Northwestern Polytechnical University)
RestorationOptimizationConvolutional Neural NetworkImage
🎯 What it does: A content-aware wide-angle image de-distortion method based on polar coordinates and least squares conformal mapping is proposed, which utilizes deep networks to automatically detect curve lines and salient regions, optimizing the mesh to maintain straight lines and local shapes while preserving the original ultra-wide perspective.
WildLight: In-the-Wild Inverse Rendering With a Flashlight
Ziang Cheng (Australian National University), Hongdong Li (Australian National University)
RestorationData SynthesisNeural Radiance FieldImage
🎯 What it does: Utilizing the separation of smartphone flash and ambient light, combined with neural light fields and physical BRDF, to achieve inverse rendering of indoor object geometry and reflective properties.
WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation
Jongheon Jeong (KAIST), Onkar Dabeer (AWS AI Labs)
ClassificationSegmentationAnomaly DetectionPrompt EngineeringContrastive LearningImage
🎯 What it does: This paper proposes a CLIP-based zero/few-shot industrial defect detection framework, WinCLIP/WinCLIP+, which can perform defect classification and segmentation under conditions of no annotations or only a few normal images.
WINNER: Weakly-Supervised hIerarchical decompositioN and aligNment for Spatio-tEmporal Video gRounding
Mengze Li (Zhejiang University), Fei Wu (Shanghai Institute for Advanced Study of Zhejiang University)
Object DetectionSegmentationTransformerContrastive LearningVideoText
🎯 What it does: The WINNER framework is proposed under weak supervision conditions, utilizing hierarchical decomposition and alignment to achieve spatiotemporal localization of video and language.
WIRE: Wavelet Implicit Neural Representations
Vishwanath Saragadam (Rice University), Richard G. Baraniuk (Rice University)
RestorationSuper ResolutionImageBiomedical DataComputed Tomography
🎯 What it does: Proposed an implicit neural representation (WIRE) based on complex Gabor wavelet activation functions, achieving high precision, fast, and robust continuous function fitting.
X-Avatar: Expressive Human Avatars
Kaiyue Shen (ETH Zurich), Otmar Hilliges (ETH Zurich)
GenerationPose EstimationPoint CloudMesh
🎯 What it does: A unified implicit human avatar model, X-Avatar, has been developed that can learn from 3D scans or RGB-D data, capable of capturing body, hand, and facial poses as well as high-frequency appearance details.
X-Pruner: eXplainable Pruning for Vision Transformers
Lu Yu (James Cook University), Wei Xiang (La Trobe University)
CompressionExplainability and InterpretabilityTransformerImage
🎯 What it does: An interpretable structured pruning framework X-Pruner is proposed, which performs unit-level pruning on visual Transformers using interpretable sensitivity-aware masks.
X3KD: Knowledge Distillation Across Modalities, Tasks and Stages for Multi-Camera 3D Object Detection
Marvin Klingner (Qualcomm Technologies), Fatih Porikli (Qualcomm AI Research)
Object DetectionAutonomous DrivingKnowledge DistillationMultimodalityPoint Cloud
🎯 What it does: This paper proposes the X KD framework, which enhances multi-camera 3D object detection performance through cross-modal, cross-task, and cross-stage knowledge distillation.
YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors
Chien-Yao Wang (Academia Sinica), Hong-Yuan Mark Liao (Academia Sinica)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: The YOLOv7 architecture and a series of training techniques are proposed, setting new records in real-time object detection for both speed and accuracy.
You Are Catching My Attention: Are Vision Transformers Bad Learners Under Backdoor Attacks?
Zenghui Yuan (Huazhong University of Science and Technology), Yu Cheng (Microsoft Research)
ClassificationAdversarial AttackTransformerImage
🎯 What it does: A backdoor attack framework called BadViT targeting Vision Transformer (ViT) has been designed and implemented, and an invisible variant has been proposed, which uses a unified local trigger to induce the model's self-attention to focus on that location, achieving a high success rate for backdoor attacks.
You Can Ground Earlier Than See: An Effective and Efficient Pipeline for Temporal Sentence Grounding in Compressed Videos
Xiang Fang (Huazhong University of Science and Technology), Guoshun Nan (Beijing University of Posts and Telecommunications)
CompressionComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkOptical FlowVideoTextMultimodality
🎯 What it does: This paper studies the method of temporal sentence grounding (TSG) in the compressed video domain, directly utilizing I-frames, motion vectors, and residuals in compressed videos for target moment localization.
You Do Not Need Additional Priors or Regularizers in Retinex-Based Low-Light Image Enhancement
Huiyuan Fu (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
RestorationKnowledge DistillationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A prior-free and regularization-free Retinex decomposition and synthesis network (RFR) is proposed, which achieves low-light image enhancement through contrastive learning and self-supervised knowledge distillation.
You Need Multiple Exiting: Dynamic Early Exiting for Accelerating Unified Vision Language Model
Shengkun Tang (North Carolina State University), Dongkuan Xu (Northeastern University)
Computational EfficiencyTransformerMixture of ExpertsVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes the MuE multi-layer early exit strategy, which allows a unified visual language model to dynamically exit layers in both the encoder and decoder within a sequence-to-sequence framework, thereby improving inference efficiency.
You Only Segment Once: Towards Real-Time Panoptic Segmentation
Jie Hu (Xiamen University), Liujuan Cao (Xiamen University)
Object DetectionSegmentationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes the YOSO framework, which achieves semantic and instance panoptic segmentation with a single segmentation, suitable for real-time scenarios.
ZBS: Zero-Shot Background Subtraction via Instance-Level Background Modeling and Foreground Selection
Yongqi An (National Laboratory of Pattern Recognition Institute of Automation CAS), Jinqiao Wang (National Laboratory of Pattern Recognition Institute of Automation CAS)
Object DetectionObject TrackingSegmentationAnomaly DetectionContrastive LearningVideoBenchmark
🎯 What it does: A background subtraction framework based on zero-shot object detection (ZBS) is proposed, achieving instance-level background modeling and foreground selection.
ZegCLIP: Towards Adapting CLIP for Zero-Shot Semantic Segmentation
Ziqin Zhou (University of Adelaide), Yifan Liu (Sichuan University)
SegmentationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: We propose a one-stage zero-shot semantic segmentation framework ZegCLIP, which transfers CLIP from image-level zero-shot classification to pixel-level segmentation.
Zero-Shot Dual-Lens Super-Resolution
Ruikang Xu (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationSuper ResolutionGenerative Adversarial NetworkContrastive LearningOptical FlowImage
🎯 What it does: A zero-shot dual-camera super-resolution framework ZeDuSR is proposed, which learns scene-specific SR models using only a single pair of dual-camera images during testing.
Zero-Shot Everything Sketch-Based Image Retrieval, and in Explainable Style
Fengyin Lin (Beijing University of Posts and Telecommunications), Yonggang Qi (Beijing University of Posts and Telecommunications)
RetrievalExplainability and InterpretabilityTransformerImage
🎯 What it does: This paper addresses the zero-shot sketch-based image retrieval (ZS-SBIR) problem by proposing a unified cross-modal Transformer network that can perform retrieval in all scenarios, including cross-category, fine-grained, and cross-dataset, using a single model while providing interpretability.
Zero-Shot Generative Model Adaptation via Image-Specific Prompt Learning
Jiayi Guo (Tsinghua University), Gao Huang (Tsinghua University)
GenerationDomain AdaptationPrompt EngineeringContrastive LearningImage
🎯 What it does: A method of Image-Specific Prompt Learning (IPL) is proposed for domain adaptation in zero-shot generative models;
Zero-Shot Model Diagnosis
Jinqi Luo (Carnegie Mellon University), Fernando De la Torre (Carnegie Mellon University)
Data SynthesisAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A zero-shot model diagnosis framework called ZOOM is proposed, which uses CLIP-driven StyleGAN to generate semantic adversarial images for sensitivity analysis and robustness enhancement of deep visual models, completely without the need for manually labeled test sets.
Zero-Shot Noise2Noise: Efficient Image Denoising Without Any Data
Youssef Mansour (Technical University of Munich), Reinhard Heckel (Technical University of Munich)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A zero-shot image denoising method ZS-N2N is proposed, which does not require training data or noise models.
Zero-Shot Object Counting
Jingyi Xu (Stony Brook University), Dimitris Samaras (Stony Brook University)
Object DetectionConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: The zero-shot object counting (ZSC) task is proposed, which allows counting instances of a target category in an image using only the category name, avoiding the reliance on human-annotated instances found in traditional methods.
Zero-Shot Pose Transfer for Unrigged Stylized 3D Characters
Jiashun Wang (Carnegie Mellon University), Jan Kautz (NVIDIA)
Pose EstimationAuto EncoderMesh
🎯 What it does: A zero-shot pose transfer method is proposed, which can transfer the reference character's pose to unbound, stylized 3D characters.
Zero-Shot Referring Image Segmentation With Global-Local Context Features
Seonghoon Yu (Gwangju Institute of Science and Technology), Jeany Son (Gwangju Institute of Science and Technology)
Object DetectionSegmentationTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: A zero-shot reference image segmentation framework is proposed, utilizing a pre-trained CLIP model and global-local context features to achieve the matching of text descriptions with pixel-level segmentation without any supervised training.
Zero-Shot Text-to-Parameter Translation for Game Character Auto-Creation
Rui Zhao (Netease Fuxi AI Lab), Changjie Fan (Netease Fuxi AI Lab)
GenerationTransformerVision Language ModelText
🎯 What it does: A zero-shot text-to-parameter translation (T2P) method based on CLIP is proposed, enabling the automatic generation of skeletal-driven game characters without reference images.