CVPR 2023 Papers — Page 18
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
Putting People in Their Place: Affordance-Aware Human Insertion Into Scenes
Sumith Kulal (Stanford University), Krishna Kumar Singh (Adobe Research)
Image TranslationGenerationPose EstimationDiffusion modelAuto EncoderVideo
🎯 What it does: This study investigates the insertion of people into scenes while considering affordance, proposing a self-supervised training diffusion model to achieve portrait insertion, pose prediction, and fusion.
PVO: Panoptic Visual Odometry
Weicai Ye (Zhejiang University), Guofeng Zhang (Zhejiang University)
SegmentationPose EstimationAutonomous DrivingOptimizationSimultaneous Localization and MappingOptical FlowImageVideo
🎯 What it does: A PVO (Panoptic Visual Odometry) framework is proposed, which combines Visual Odometry (VO) and Video Panoptic Segmentation (VPS), achieving simultaneous localization and 3D panoramic map construction through bidirectional cyclic optimization.
PVT-SSD: Single-Stage 3D Object Detector With Point-Voxel Transformer
Honghui Yang (Zhejiang University), Wanli Ouyang (Shanghai AI Laboratory)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: This paper proposes a single-stage 3D object detection framework PVT-SSD, which integrates voxel and point cloud representations and utilizes a point-voxel Transformer for efficient detection.
PyPose: A Library for Robot Learning With Physics-Based Optimization
Chen Wang (Carnegie Mellon University), Sebastian Scherer (Carnegie Mellon University)
OptimizationRobotic IntelligenceSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Developed PyPose - an open-source library based on PyTorch that integrates deep learning perception models with physics-based optimization, achieving end-to-end differentiable solutions for robotic learning tasks; and demonstrates its practical effects in various applications such as SLAM, planning, control, and inertial navigation through examples.
PyramidFlow: High-Resolution Defect Contrastive Localization Using Pyramid Normalizing Flow
Jiarui Lei (Zhejiang University), Dong Liu (Zhejiang University)
Object DetectionAnomaly DetectionFlow-based ModelContrastive LearningImage
🎯 What it does: This paper studies a fully normalized flow model called PyramidFlow for high-resolution defect contrast localization in industrial applications.
Q-DETR: An Efficient Low-Bit Quantized Detection Transformer
Sheng Xu (Beihang University), Baochang Zhang (Beihang University)
Object DetectionCompressionComputational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes a low-bit quantization method for DETR (Q-DETR) and corrects the information distortion of the quantized query vectors through Distributional Correction Distillation (DRD), significantly improving the detection performance of low-bit DETR.
Q: How To Specialize Large Vision-Language Models to Data-Scarce VQA Tasks? A: Self-Train on Unlabeled Images!
Zaid Khan (Northeastern University), Manmohan Chandraker (NEC Labs America)
Domain AdaptationData-Centric LearningTransformerSupervised Fine-TuningVision Language ModelImageMultimodality
🎯 What it does: A teacher model is constructed using a large-scale visual language model (VLM), leveraging unlabeled image generation question-answer pseudo-labels to enhance self-training on the target small-scale VQA dataset;
QPGesture: Quantization-Based and Phase-Guided Motion Matching for Natural Speech-Driven Gesture Generation
Sicheng Yang (Tsinghua University), Haolin Zhuang (Tsinghua University)
GenerationData SynthesisAuto EncoderTextMultimodalityAudio
🎯 What it does: This paper studies a motion matching framework based on quantization and phase guidance for generating natural and coherent gestures from speech and text.
Quality-Aware Pre-Trained Models for Blind Image Quality Assessment
Kai Zhao (Kuaishou Technology), Xing Wen (Kuaishou Technology)
RecognitionRestorationContrastive LearningImage
🎯 What it does: For the task of no-reference image quality assessment, this paper proposes a quality-aware self-supervised pre-training method called QPT.
QuantArt: Quantizing Image Style Transfer Towards High Visual Fidelity
Siyu Huang (Harvard University), Hanspeter Pfister (Harvard University)
Image TranslationGenerationAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: The QuantArt framework is proposed, which uses vector quantization and an art codebook to approximate the latent features of generated images to the distribution of real artworks, enhancing the visual realism of style transfer and achieving adjustable control over content, style, and visual quality.
Quantitative Manipulation of Custom Attributes on 3D-Aware Image Synthesis
Hoseok Do (LG Electronics), Jin Young Choi (Seoul National University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A custom attribute quantization and multi-attribute fine-tuning image editing model based on 3D GAN (EG3D) is proposed, which enables perspective-consistent editing of 3D images through user-defined attribute quantization values (0-1).
Quantum Multi-Model Fitting
Matteo Farina (University of Trento), Federica Arrigoni (Politecnico di Milano)
OptimizationTabularPhysics Related
🎯 What it does: The first quantum method QUMF for multi-model geometric fitting using decoherent quantum computers is proposed, along with a decomposable version DEQUMF to address large-scale problems.
Quantum-Inspired Spectral-Spatial Pyramid Network for Hyperspectral Image Classification
Jie Zhang (University of Macau), Yicong Zhou (University of Macau)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: A quantum-inspired spectral-spatial network (QSSN) and a pyramid framework based on it (QSSPN) are proposed for pixel-level classification of hyperspectral images (HSI).
Query-Centric Trajectory Prediction
Zikang Zhou (City University of Hong Kong), Yu-Kai Huang (Carnegie Mellon University)
Autonomous DrivingTransformerTime Series
🎯 What it does: The QCNet framework is proposed to address the challenges of efficient online inference and multi-modal long-range prediction in autonomous driving trajectory prediction.
Query-Dependent Video Representation for Moment Retrieval and Highlight Detection
WonJun Moon (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
RetrievalTransformerContrastive LearningVideoText
🎯 What it does: This paper proposes a Query-Dependent DETR (QD-DETR) based on Transformer for video temporal segment retrieval and highlight detection.
R2Former: Unified Retrieval and Reranking Transformer for Place Recognition
Sijie Zhu (ByteDance), Heng Wang (Center for Research in Computer Vision, University of Central Florida)
RecognitionRetrievalTransformerImage
🎯 What it does: A unified visual place recognition framework R2Former is proposed, which uses Transformer for end-to-end global retrieval and re-ranking.
RA-CLIP: Retrieval Augmented Contrastive Language-Image Pre-Training
Chen-Wei Xie (Alibaba Group), Jingren Zhou (Alibaba Group)
ClassificationRetrievalTransformerContrastive LearningImageTextMultimodality
🎯 What it does: Enhance CLIP's image representation by retrieving relevant image-text pairs from a reference set during training and inference, and perform contrastive learning based on this.
RaBit: Parametric Modeling of 3D Biped Cartoon Characters With a Topological-Consistent Dataset
Zhongjin Luo (Chinese University of Hong Kong Shenzhen), Xiaoguang Han (Huawei Technologies)
GenerationData SynthesisPose EstimationConvolutional Neural NetworkGenerative Adversarial NetworkMesh
🎯 What it does: This paper first constructs a large 3D biped cartoon character dataset called 3DBiCar, and based on this dataset, proposes the RaBit parametric model to simultaneously model the shape, pose, and texture of characters. Subsequently, a baseline method called BiCarNet is designed for applications such as single-view reconstruction, sketch modeling, and animation transfer, demonstrating its effectiveness.
Randomized Adversarial Training via Taylor Expansion
Gaojie Jin (Institute of Software, Chinese Academy of Sciences), Xiaowei Huang (University of Liverpool)
OptimizationAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a method that adds small Gaussian noise to model weights and utilizes Taylor expansion to achieve randomized adversarial training, aiming to enhance both the model's adversarial robustness and clean data accuracy.
Range-Nullspace Video Frame Interpolation With Focalized Motion Estimation
Zhiyang Yu (Harbin Institute of Technology), Shunqing Ren (Harbin Institute of Technology)
RestorationData SynthesisOptical FlowVideo
🎯 What it does: A multi-frame video frame interpolation framework utilizing focused trajectory fitting and range-null space synthesis is proposed.
RangeViT: Towards Vision Transformers for 3D Semantic Segmentation in Autonomous Driving
Angelika Ando (Valeo), Renaud Marlet
SegmentationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: In this work, the authors propose a LiDAR point cloud semantic segmentation framework based on Vision Transformer (ViT) called RangeViT, which maps 3D point clouds to 2D images using range projection before segmentation.
Ranking Regularization for Critical Rare Classes: Minimizing False Positives at a High True Positive Rate
Kiarash Mohammadi (Mila, Université de Montréal), Frederick Tung (Borealis AI)
ClassificationAnomaly DetectionImageBiomedical Data
🎯 What it does: This paper proposes a ranking-based regularization method (RankReg) to significantly reduce the false positive rate (FPR) at a high true positive rate (TPR).
RankMix: Data Augmentation for Weakly Supervised Learning of Classifying Whole Slide Images With Diverse Sizes and Imbalanced Categories
Yuan-Chih Chen (Academia Sinica), Chun-Shien Lu (Academia Sinica)
ClassificationKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningImageBiomedical Data
🎯 What it does: Proposes the RankMix data augmentation method, which enhances model robustness in weakly supervised WSI classification by mixing features of different sizes through pseudo-labels and feature ranking.
Rate Gradient Approximation Attack Threats Deep Spiking Neural Networks
Tong Bu (Peking University), Zhaofei Yu (Peking University)
Adversarial AttackSpiking Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper studies the vulnerability of Spiking Neural Networks (SNN) to adversarial attacks and proposes a new attack method called Rate Gradient Approximation Attack (RGA).
Raw Image Reconstruction With Learned Compact Metadata
Yufei Wang (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
RestorationCompressionAuto EncoderImage
🎯 What it does: An end-to-end learning framework for raw image reconstruction is proposed, achieving high-quality raw image recovery by learning compressed metadata in the latent space.
Rawgment: Noise-Accounted RAW Augmentation Enables Recognition in a Wide Variety of Environments
Masakazu Yoshimura (Sony Group Corporation), Takeshi Ohashi (Sony Group Corporation)
RecognitionObject DetectionImage
🎯 What it does: A noise correction method for RAW data enhancement before ISP is proposed, combining color jittering, blurring, and a sensor noise model.
Re-Basin via Implicit Sinkhorn Differentiation
Fidel A. Guerrero Peña (ETS Montreal), Marco Pedersoli (ETS Montreal)
OptimizationImage
🎯 What it does: This paper proposes a differentiable re-basin method based on the Sinkhorn operator and implicit differentiation to find function-equivalent rearrangements in the parameter space of neural networks, enabling model transfer, fusion, and incremental learning.
Re-GAN: Data-Efficient GANs Training via Architectural Reconfiguration
Divya Saxena (Hong Kong Polytechnic University), Tarun Kulshrestha (Hong Kong Polytechnic University)
GenerationData SynthesisComputational EfficiencyGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes Re-GAN, a GAN training framework that dynamically prunes and regrows during the training process, enabling the exploration of different subnetwork structures under limited data, thus achieving more efficient training.
Re-IQA: Unsupervised Learning for Image Quality Assessment in the Wild
Avinab Saha (University of Texas at Austin), Alan C. Bovik (University of Texas at Austin)
Representation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised mixed expert framework called Re-IQA, which learns content features and low-level quality features through dual encoders and utilizes a single-layer regression to achieve no-reference image quality prediction.
Re-Thinking Federated Active Learning Based on Inter-Class Diversity
SangMook Kim (KAIST AI), Se-Young Yun (KAIST AI)
Federated LearningImage
🎯 What it does: The LoGo algorithm is proposed within the framework of federated active learning, combining global and local models for a two-step clustering query, aiming to address both local heterogeneity and global imbalance issues.
Re-Thinking Model Inversion Attacks Against Deep Neural Networks
Ngoc-Bao Nguyen (Singapore University of Technology and Design), Ngai-Man Cheung (Singapore University of Technology and Design)
RecognitionKnowledge DistillationAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: An improved method for model inversion attacks on deep networks is proposed.
Re2TAL: Rewiring Pretrained Video Backbones for Reversible Temporal Action Localization
Chen Zhao (King Abdullah University of Science and Technology), Bernard Ghanem (King Abdullah University of Science and Technology)
RecognitionObject DetectionComputational EfficiencyTransformerVideo
🎯 What it does: This paper proposes a reversible network reconnection method (Re 2 TAL) that transforms existing pre-trained video backbone networks into reversible networks, achieving end-to-end temporal action localization, significantly reducing memory usage and improving performance.
Real-Time 6K Image Rescaling With Rate-Distortion Optimization
Chenyang Qi (Hong Kong University of Science and Technology), Qifeng Chen (Hong Kong University of Science and Technology)
Super ResolutionCompressionOptimizationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: The HyperThumbnail framework is proposed to compress high-resolution images into low-resolution JPEG thumbnails and to restore 6K high-quality images in real-time on the client side.
Real-Time Controllable Denoising for Image and Video
Zhaoyang Zhang (Chinese University of Hong Kong), Jinwei Gu (Chinese University of Hong Kong)
RestorationConvolutional Neural NetworkTransformerImageVideo
🎯 What it does: A real-time controllable denoising framework RCD is proposed, allowing for different denoising intensities and detail adjustments through noise editing after a single forward inference.
Real-Time Evaluation in Online Continual Learning: A New Hope
Yasir Ghunaim (King Abdullah University of Science and Technology), Bernard Ghanem (King Abdullah University of Science and Technology)
Computational EfficiencyConvolutional Neural NetworkReinforcement LearningImageTime Series
🎯 What it does: This paper proposes a real-time evaluation framework based on training complexity to assess the performance of online continuous learning (OCL) methods in high-speed data streams, and compares various existing methods on the large-scale timestamped image dataset CLOC.
Real-Time Multi-Person Eyeblink Detection in the Wild for Untrimmed Video
Wenzheng Zeng (Huazhong University of Science and Technology), Joey Tianyi Zhou (Agency for Science Technology and Research)
RecognitionObject DetectionObject TrackingTransformerVideo
🎯 What it does: The first unedited, field video dataset for multi-person eye blink detection, MPEblink, is proposed, and the InstBlink model is introduced based on a single-stage query Transformer, achieving simultaneous face detection, tracking, and instance-level blink detection.
Real-Time Neural Light Field on Mobile Devices
Junli Cao (Snap Inc), Jian Ren (Snap Inc)
GenerationData SynthesisComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: To achieve real-time neural light field rendering on mobile devices, the MobileR2L model is proposed, which can achieve quality comparable to NeRF under low storage and low latency conditions.
RealFusion: 360deg Reconstruction of Any Object From a Single Image
Luke Melas-Kyriazi (University of Oxford), Andrea Vedaldi (University of Oxford)
GenerationOptimizationDiffusion modelNeural Radiance FieldImagePoint Cloud
🎯 What it does: Reconstructing a complete 360° 3D model of an object from a single image, utilizing a pre-trained 2D diffusion model and single-image text inversion technology to generate multi-view priors, combined with NeRF for optimization.
RealImpact: A Dataset of Impact Sound Fields for Real Objects
Samuel Clarke (Adobe Research), Jiajun Wu (Adobe Research)
BenchmarkAudio
🎯 What it does: The REALIMPACT dataset has been constructed, containing 150,000 impact audio recordings of 50 everyday objects, with complete annotations including impact location, microphone position, impact force, material labels, and RGBD images.
Realistic Saliency Guided Image Enhancement
S. Mahdi H. Miangoleh (Simon Fraser University), Yağiz Aksoy (Adobe Research)
Image TranslationImage HarmonizationRestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Enhance or weaken the target area through traditional image editing operations (exposure, saturation, color curves, white balance) to guide viewer attention while maintaining the realism of the photo.
ReasonNet: End-to-End Driving With Temporal and Global Reasoning
Hao Shao (Chinese University of Hong Kong), Yu Liu (Shanghai Artificial Intelligence Laboratory)
Autonomous DrivingRecurrent Neural NetworkGraph Neural NetworkTransformerPoint CloudBenchmark
🎯 What it does: An end-to-end driving framework called ReasonNet is proposed, integrating temporal reasoning and global reasoning modules to model historical and global information in driving scenarios.
Rebalancing Batch Normalization for Exemplar-Based Class-Incremental Learning
Sungmin Cha (Seoul National University), Taesup Moon (Seoul National University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: A normalization layer named Task-Balanced Batch Normalization (TBBN) has been designed and implemented, specifically improved for the BN bias problem in sample storage-based offline class incremental learning (CIL) scenarios.
REC-MV: REconstructing 3D Dynamic Cloth From Monocular Videos
Lingteng Qiu (Chinese University of Hong Kong, Shenzhen), Xiaoguang Han (Chinese University of Hong Kong, Shenzhen)
GenerationOptimizationVideoMesh
🎯 What it does: By jointly optimizing explicit feature curves and implicit SDF from monocular video, we extract temporally coherent dynamic 3D garment meshes with open boundaries.
ReCo: Region-Controlled Text-to-Image Generation
Zhengyuan Yang (Microsoft), Lijuan Wang (Microsoft)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper studies a text-to-image generation model called ReCo that achieves regional control by incorporating position tokens.
Recognizability Embedding Enhancement for Very Low-Resolution Face Recognition and Quality Estimation
Jacky Chen Long Chai (Yonsei University), Andrew Beng Jin Teoh (Yonsei University)
RecognitionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A recognizable index (RI)-based embedding learning framework is proposed to enhance very low-resolution face recognition performance and achieve face quality assessment.
Recognizing Rigid Patterns of Unlabeled Point Clouds by Complete and Continuous Isometry Invariants With No False Negatives and No False Positives
Daniel Widdowson (University of Liverpool), Vitaliy Kurlin (University of Liverpool)
RecognitionComputational EfficiencyDrug DiscoveryPoint Cloud
🎯 What it does: This paper proposes two new complete and continuous rigid body equivariant invariants (Simplexwise Distance Distribution and Simplexwise Centered Distribution) for isometric determination of unlabeled point clouds, and provides a polynomial time algorithm.
Reconstructing Animatable Categories From Videos
Gengshan Yang (Carnegie Mellon University), Deva Ramanan (Carnegie Mellon University)
GenerationPose EstimationNeural Radiance FieldOptical FlowVideo
🎯 What it does: A method for animatable category-level 3D model reconstruction based on monocular video is proposed, which can simultaneously separate instance morphology and temporal motion and achieve cross-instance motion transfer.
Reconstructing Signing Avatars From Video Using Linguistic Priors
Maria-Paola Forte (Max Planck Institute for Intelligent Systems), Michael J. Black (Max Planck Institute for Intelligent Systems)
RecognitionPose EstimationVideo
🎯 What it does: This paper presents SGNify, a system capable of automatically reconstructing the 3D body shape and posture (including hands, face, and body) of sign language speakers from monocular videos.
Recovering 3D Hand Mesh Sequence From a Single Blurry Image: A New Dataset and Temporal Unfolding
Yeonguk Oh (Seoul National University), Kyoung Mu Lee (Seoul National University)
RestorationPose EstimationConvolutional Neural NetworkTransformerImageVideoMesh
🎯 What it does: This paper presents the BlurHand dataset and the BlurHandNet model for recovering 3D hand mesh sequences from a single blurred hand image.
Recurrence Without Recurrence: Stable Video Landmark Detection With Deep Equilibrium Models
Paul Micaelli (University of Edinburgh), Pavlo Molchanov (NVIDIA)
Pose EstimationComputational EfficiencyImageVideo
🎯 What it does: The deep equilibrium model (DEQ) is transferred to the task of facial keypoint detection, proposing the Landmark DEQ (LDEQ) model, which uses a one-shot root solver to directly solve the heatmap equilibrium points during forward propagation, significantly reducing memory and computational costs; and introduces a temporal consistency constraint through a 'non-recursive recursion' (RwR) mechanism during video inference to suppress keypoint jitter.
Recurrent Homography Estimation Using Homography-Guided Image Warping and Focus Transformer
Si-Yuan Cao (Zhejiang University), Hui-Liang Shen (Zhejiang University)
RecognitionImage TranslationOptimizationRecurrent Neural NetworkTransformerImage
🎯 What it does: This paper proposes a recursive homography estimation framework (RHWF) that combines homography-guided image pre-warping with the FocusFormer Transformer to achieve more accurate transformation estimation.
Recurrent Vision Transformers for Object Detection With Event Cameras
Mathias Gehrig (University of Zurich), Davide Scaramuzza (University of Zurich)
Object DetectionComputational EfficiencyTransformerImage
🎯 What it does: This paper proposes a novel Recursive Visual Transformer (RVT) backbone network for event camera object detection.
ReDirTrans: Latent-to-Latent Translation for Gaze and Head Redirection
Shiwei Jin (University of California San Diego), Truong Nguyen (University of California San Diego)
Image TranslationData SynthesisPose EstimationGenerative Adversarial NetworkImage
🎯 What it does: A ReDirTrans network is proposed, capable of redirecting gaze and head pose on high-resolution full-face images.
Reducing the Label Bias for Timestamp Supervised Temporal Action Segmentation
Kaiyuan Liu (Dalian University of Technology), Zihang Shao (Dalian University of Technology)
SegmentationTransformerVideo
🎯 What it does: This paper proposes a debiased framework D-TSTAS for the task of timestamp-supervised temporal action segmentation (TSTAS) to alleviate performance degradation caused by label bias.
Ref-NPR: Reference-Based Non-Photorealistic Radiance Fields for Controllable Scene Stylization
Yuechen Zhang (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
Image TranslationGenerationNeural Radiance FieldImage
🎯 What it does: Using a single stylized 2D view, combined with reference ray registration and template correspondence, to construct pseudo-ray supervision and semantic correspondence, achieving controllable non-photorealistic stylization of 3D scenes.
RefCLIP: A Universal Teacher for Weakly Supervised Referring Expression Comprehension
Lei Jin (Xiamen University), Rongrong Ji (Contemporary Amperex Technology Co. Limited)
RecognitionObject DetectionRecurrent Neural NetworkContrastive LearningImage
🎯 What it does: A weakly supervised reference expression understanding model called RefCLIP is proposed, which achieves weakly supervised training through anchor-text matching and anchor-based contrastive learning. At the same time, a general teacher-student pseudo-label training scheme is designed without model modifications, which can enhance the performance of existing REC models.
Referring Image Matting
Jizhizi Li (University of Sydney), Dacheng Tao (University of Sydney)
Image TranslationSegmentationTransformerVision Language ModelImageBenchmark
🎯 What it does: Proposes the Reference Image Matting (RIM) task and designs the benchmark method CLIPMat.
Referring Multi-Object Tracking
Dongming Wu (Beijing Institute of Technology), Jianbing Shen (University of Macau)
Object TrackingAutonomous DrivingTransformerVideoTextMultimodalityBenchmark
🎯 What it does: This paper proposes a language expression-based multi-object tracking task (RMOT) and constructs a new benchmark dataset, Refer-KITTI. Subsequently, an end-to-end Transformer model, TransRMOT, is designed and implemented to accomplish this task.
RefSR-NeRF: Towards High Fidelity and Super Resolution View Synthesis
Xudong Huang (Huawei Noah's Ark Lab), Yunhe Wang (Huawei Noah's Ark Lab)
GenerationData SynthesisSuper ResolutionNeural Radiance FieldImage
🎯 What it does: Proposes the RefSR-NeRF framework, which first trains a low-resolution NeRF representation and then uses a lightweight reference-guided super-resolution network to recover high-frequency details, achieving high-resolution and realistic view synthesis.
RefTeacher: A Strong Baseline for Semi-Supervised Referring Expression Comprehension
Jiamu Sun (Xiamen University), Rongrong Ji (Shenzhen Research Institute of Xiamen University)
RecognitionTransformerContrastive LearningImage
🎯 What it does: This paper proposes RefTeacher, a semi-supervised learning framework for referring expression comprehension (REC);
Region-Aware Pretraining for Open-Vocabulary Object Detection With Vision Transformers
Dahun Kim (Google Research), Weicheng Kuo (Google Research)
Object DetectionRetrievalTransformerContrastive LearningImageText
🎯 What it does: This paper proposes a Region-Aware Vision Transformer (RO-ViT) that enhances the performance of ViT in open vocabulary object detection by randomly cropping and scaling positional embeddings during the image-text contrastive pretraining phase, and by using focal loss, followed by fine-tuning with an improved object proposal scheme.
Regularization of Polynomial Networks for Image Recognition
Grigorios G. Chrysos, Volkan Cevher (École Polytechnique Fédérale de Lausanne)
ClassificationRecognitionConvolutional Neural NetworkImageAudio
🎯 What it does: This paper proposes a strong regularization approach using polynomial networks (R-PolyNets and D-PolyNets) to achieve image and audio recognition performance comparable to or even better than ResNet18.
Regularize Implicit Neural Representation by Itself
Zhemin Li (National University of Defense Technology), Deyu Meng (Xi'an Jiaotong University)
RestorationImage
🎯 What it does: This paper proposes a regularization method called Implicit Neural Representation Regularizer (INRR) to enhance the generalization performance of implicit neural representations (INR) under non-uniform sampling conditions.
Regularized Vector Quantization for Tokenized Image Synthesis
Jiahui Zhang (Nanyang Technological University), Shijian Lu (Max Planck Institute for Informatics)
GenerationData SynthesisGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A regularized vector quantization framework is proposed, integrating deterministic and randomized quantization to achieve high-quality tokenized image synthesis.
Regularizing Second-Order Influences for Continual Learning
Zhicheng Sun (Peking University), Gang Hua (Wormpex AI Research)
Image
🎯 What it does: A new consensus subset selection method is studied for the experience replay buffer in continual learning, utilizing influence functions and second-order influence regularization to enhance performance.
Reinforcement Learning-Based Black-Box Model Inversion Attacks
Gyojin Han (Korea Advanced Institute of Science and Technology), Junmo Kim (Korea Advanced Institute of Science and Technology)
Adversarial AttackReinforcement LearningGenerative Adversarial NetworkImage
🎯 What it does: Using reinforcement learning methods, this paper addresses black-box model reverse attacks by proposing the reconstruction of private training data through latent space search of a GAN generator.
Relational Context Learning for Human-Object Interaction Detection
Sanghyun Kim (Pohang University of Science and Technology), Minsu Cho (Pohang University of Science and Technology)
RecognitionObject DetectionTransformerImage
🎯 What it does: This paper proposes a three-branch Transformer structure for HOI detection framework MUREN, which utilizes multiple relationship embeddings (single, dual, and triple relationships) and attention fusion to achieve cross-branch context exchange, thereby enhancing the recognition performance of human-object interactions.
Relational Space-Time Query in Long-Form Videos
Xitong Yang (Meta AI), Du Tran (Meta AI)
RecognitionObject DetectionRetrievalVideoMultimodalityBenchmark
🎯 What it does: This paper proposes the Relational Space-Time Query (ReST) framework, which utilizes templated visual queries to evaluate activities, objects, and their interactions in long videos.
Reliability in Semantic Segmentation: Are We on the Right Track?
Pau de Jorge (University of Oxford), Grégory Rogez (NAVER LABS Europe)
SegmentationDomain AdaptationAnomaly DetectionConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: This paper systematically evaluates the robustness and uncertainty estimation of various modern semantic segmentation models (including ResNet, ConvNeXt, SETR, SegFormer, Segmenter, etc.) under natural domain shifts, exploring the impact of calibration methods (temperature scaling, cluster calibration, local temperature scaling) on calibration, misclassification detection, and OOD detection.
Reliable and Interpretable Personalized Federated Learning
Zixuan Qin (Tianjin University), Qinghua Hu (Tianjin University)
Federated LearningExplainability and InterpretabilityImage
🎯 What it does: This paper proposes a reliable and interpretable personalized federated learning framework called RIPFL, which quantifies client uncertainty using evidence theory and fully utilizes collective knowledge through reliable client selection and interpretable aggregation strategies.
ReLight My NeRF: A Dataset for Novel View Synthesis and Relighting of Real World Objects
Marco Toschi (Eyecan AI), Samuele Salti (University of Bologna)
GenerationData SynthesisNeural Radiance FieldImageBenchmark
🎯 What it does: A new dataset called ReNe is proposed, and improvements to NeRF are made on this dataset to achieve real-time view synthesis and relighting under unknown lighting conditions.
Relightable Neural Human Assets From Multi-View Gradient Illuminations
Taotao Zhou (ShanghaiTech University), Jingyi Yu (ShanghaiTech University)
RestorationGenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: This paper presents the UltraStage dataset and constructs illuminable neural human assets based on multi-view gradient lighting data, achieving high-quality geometric reconstruction and image synthesis from arbitrary viewpoints/lighting.
RelightableHands: Efficient Neural Relighting of Articulated Hand Models
Shun Iwase (Carnegie Mellon University), Jason Saragih (Reality Labs Research)
GenerationOptimizationComputational EfficiencyConvolutional Neural NetworkNeural Radiance FieldMesh
🎯 What it does: A teacher-student framework for neural rendering has been developed, capable of rendering highly realistic animated hand models in real-time under arbitrary lighting conditions.
Removing Objects From Neural Radiance Fields
Silvan Weder (Niantic), Sara Vicente (Niantic)
RestorationObject DetectionNeural Radiance FieldImage
🎯 What it does: A method for object removal in NeRF using 2D image inpainting and confidence-based view selection is proposed.
Renderable Neural Radiance Map for Visual Navigation
Obin Kwon (Seoul National University), Songhwai Oh (Seoul National University)
Robotic IntelligenceReinforcement LearningNeural Radiance FieldImage
🎯 What it does: This paper proposes a renderable neural radiance map (RNR-Map) that embeds environmental visual information into a grid map using latent codes, achieving image localization and goal-oriented navigation.
RenderDiffusion: Image Diffusion for 3D Reconstruction, Inpainting and Generation
Titas Anciukevičius (University of Edinburgh), Paul Guerrero (Adobe Research)
RestorationGenerationDiffusion modelScore-based ModelImage
🎯 What it does: A 3D-aware diffusion model named RenderDiffusion has been developed, capable of achieving 3D reconstruction, 3D-aware inpainting, and unconditional 3D generation using only single-view 2D images for training.
RepMode: Learning to Re-Parameterize Diverse Experts for Subcellular Structure Prediction
Donghao Zhou (Chinese Academy of Sciences), Pheng-Ann Heng (Chinese Academy of Sciences)
SegmentationConvolutional Neural NetworkMixture of ExpertsImage
🎯 What it does: This paper proposes a network called RepMode, which is used to predict fluorescence images of various subcellular structures from 3D transmitted light images, addressing the challenges of partial labeling and multi-scale.
Representation Learning for Visual Object Tracking by Masked Appearance Transfer
Haojie Zhao (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
Object TrackingRepresentation LearningTransformerAuto EncoderVideo
🎯 What it does: A visual object tracking-specific representation learning method named Masked Appearance Transfer (MAT) is proposed, which jointly encodes the template and search area using an encoder-decoder architecture, and establishes target correspondence through non-trivial reconstruction objectives.
Representing Volumetric Videos As Dynamic MLP Maps
Sida Peng (Zhejiang University), Xiaowei Zhou (Zhejiang University)
Data SynthesisCompressionComputational EfficiencyConvolutional Neural NetworkNeural Radiance FieldVideo
🎯 What it does: This paper studies a volumetric video representation method for real-time dynamic scene view synthesis, namely dynamic MLP (Multi-Layer Perceptron) networks.
Reproducible Scaling Laws for Contrastive Language-Image Learning
Mehdi Cherti (Juelich Supercomputing Center Research Center Juelich), Jenia Jitsev (Juelich Supercomputing Center Research Center Juelich)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Systematically conducted large-scale learning experiments on CLIP to explore the impact of model size, data scale, and the number of training samples on downstream task performance, forming publicly reproducible scaling laws.
ResFormer: Scaling ViTs With Multi-Resolution Training
Rui Tian (Fudan University), Yu-Gang Jiang (Fudan University)
Object DetectionSegmentationKnowledge DistillationTransformerImageVideo
🎯 What it does: This paper proposes ResFormer, which achieves good adaptation to different resolutions by using multi-scale inputs, scale consistency loss, and global-local position encoding during the training phase.
Residual Degradation Learning Unfolding Framework With Mixing Priors Across Spectral and Spatial for Compressive Spectral Imaging
Yubo Dong (Xidian University), Guangming Shi (Xidian University)
RestorationTransformerImage
🎯 What it does: To address the problem of spectral image reconstruction under the CASSI system, a Residual Degradation Learning Unfolding Framework (RDLUF) is proposed, incorporating the Mix S2 Transformer to form an end-to-end trainable RDLUF-MixS2 model for recovering 3D spectral cubes.
Resource-Efficient RGBD Aerial Tracking
Jinyu Yang (Southern University of Science and Technology), Aleš Leonardis (University of Birmingham)
Object TrackingComputational EfficiencyConvolutional Neural NetworkVideoMultimodality
🎯 What it does: This paper proposes a new task for drone aerial trajectory tracking—RGBD aerial tracking—and designs an efficient multimodal tracker EMT;
Restoration of Hand-Drawn Architectural Drawings Using Latent Space Mapping With Degradation Generator
Nakkwan Choi (Ulsan National Institute of Science and Technology), Seungjoon Yang (Ulsan National Institute of Science and Technology)
RestorationData SynthesisAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a method for restoring hand-drawn architectural drawings based on Vector Quantized Variational Autoencoder (VQ-VAE). By learning the latent space of clean drawings and mapping the latent vectors of noisy drawings, it achieves denoising and restoration of hand-drawn drawings of wooden structures.
Rethinking Domain Generalization for Face Anti-Spoofing: Separability and Alignment
Yiyou Sun (Google Research), Wen-Sheng Chu (University of Wisconsin-Madison)
RecognitionDomain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This study proposes a cross-domain learning framework for facial anti-spoofing, SA-FAS, which enhances generalization ability by maintaining domain features that are separable and aligned.
Rethinking Feature-Based Knowledge Distillation for Face Recognition
Jingzhi Li (Samsung Research and Development Institute China), Sungjoo Suh (Samsung Advanced Institute of Technology)
RecognitionKnowledge DistillationImage
🎯 What it does: Research on identity-free supervised feature distillation methods, and propose a reverse distillation technique to narrow the intrinsic dimensionality gap between teacher and student;
Rethinking Federated Learning With Domain Shift: A Prototype View
Wenke Huang (Wuhan University), Bo Du (Wuhan University)
Domain AdaptationFederated LearningContrastive LearningImage
🎯 What it does: Designed and implemented the Federated Prototypes Learning (FPL) scheme, which utilizes clustering prototypes and unbiased prototypes to enhance model generalization and stability in a federated learning environment with domain shift.
Rethinking Few-Shot Medical Segmentation: A Vector Quantization View
Shiqi Huang (Beijing Institute of Technology), Jianan Li (Beijing Institute of Technology)
SegmentationImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a vector quantization-based learning framework for extracting and utilizing prototype vectors for segmentation in a small amount of labeled medical images.
Rethinking Gradient Projection Continual Learning: Stability / Plasticity Feature Space Decoupling
Zhen Zhao (East China Normal University), Lizhuang Ma (East China Normal University)
ClassificationObject DetectionImage
🎯 What it does: This paper proposes the Space Decoupling (SD) algorithm, which divides the feature space into a stable subspace I and a plastic subspace R, combining gradient projection to achieve continuous learning and enhance the balance between stability and plasticity.
Rethinking Image Super Resolution From Long-Tailed Distribution Learning Perspective
Yuanbiao Gou (Sichuan University), Xi Peng (Sichuan University)
RestorationSuper ResolutionConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper explores the dual fitting problem in image super-resolution from the perspective of machine learning, proposing to reframe image super-resolution as a long-tail distribution learning problem, and designs a new solution to improve the fitting balance between low-frequency and high-frequency regions.
Rethinking Optical Flow From Geometric Matching Consistent Perspective
Qiaole Dong (Fudan University), Yanwei Fu (Fudan University)
Domain AdaptationTransformerOptical FlowImageVideo
🎯 What it does: This paper proposes a new optical flow estimation framework: first pre-training a feature matching extractor using Geometric Image Matching (GIM) data, and then fine-tuning it for the optical flow task, aiming to enhance feature representation and cross-dataset generalization.
Rethinking Out-of-Distribution (OOD) Detection: Masked Image Modeling Is All You Need
Jingyao Li (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
Anomaly DetectionTransformerSupervised Fine-TuningImage
🎯 What it does: A framework for OOD detection called MOOD is proposed, which is based on Masked Image Modeling (MIM). It learns the ID distribution features through a self-supervised reconstruction task and then uses the Mahalanobis distance for anomaly detection.
Rethinking the Approximation Error in 3D Surface Fitting for Point Cloud Normal Estimation
Hang Du (Hikvision Research Institute), Shiliang Pu (Hikvision Research Institute)
Graph Neural NetworkPoint Cloud
🎯 What it does: This paper analyzes the approximation error in n-jet surface fitting and proposes two design principles (Z-direction transformation and normal error estimation) to improve the accuracy of point cloud normal estimation.
Rethinking the Correlation in Few-Shot Segmentation: A Buoys View
Yuan Wang (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
SegmentationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A method is proposed to correct pixel-level error matching in few-shot segmentation by learning representative reference features (buoys) and employing an adaptive association ABC network.
Rethinking the Learning Paradigm for Dynamic Facial Expression Recognition
Hanyang Wang (Shanghai Institute of AI for Education), Aimin Zhou (Shanghai Institute of AI for Education)
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkVideo
🎯 What it does: View dynamic expression recognition as a weakly supervised multi-instance learning problem, proposing the M3DFEL framework: first, divide the video into 3D instances and use 3D-CNN to learn short-term temporal sequences; then use BiLSTM, attention mechanisms, and dynamic multi-instance normalization to aggregate instances, capturing long-term temporal sequences, ultimately obtaining expression predictions.
Rethinking Video ViTs: Sparse Video Tubes for Joint Image and Video Learning
AJ Piergiovanni (Google Research), Anelia Angelova (Google Research)
ClassificationRecognitionComputational EfficiencyTransformerImageVideo
🎯 What it does: By using Sparse Video Tubes in videos, the standard Vision Transformer (ViT) is transformed into an efficient model capable of processing both images and videos simultaneously.
REVEAL: Retrieval-Augmented Visual-Language Pre-Training With Multi-Source Multimodal Knowledge Memory
Ziniu Hu (University of California), Alireza Fathi (Google Research)
GenerationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: A retrievable visual language model REVEAL has been constructed, jointly training a retriever and a generator, utilizing multi-source multimodal knowledge memory to answer knowledge-intensive tasks.
Revealing the Dark Secrets of Masked Image Modeling
Zhenda Xie (Tsinghua University), Yue Cao (Microsoft Research)
ClassificationObject TrackingPose EstimationDepth EstimationTransformerSupervised Fine-TuningImageVideo
🎯 What it does: Compared the representation differences between Masked Image Modeling (MIM) and supervised pre-training models, revealing the locality bias and head diversity features of MIM in Transformers through visualization (attention distance, entropy, KL, CKA) and large-scale experiments.
ReVISE: Self-Supervised Speech Resynthesis With Visual Input for Universal and Generalized Speech Regeneration
Wei-Ning Hsu (Meta AI Research), Yossi Adi (Meta Reality Labs Research)
RestorationGenerationTransformerGenerative Adversarial NetworkVideoAudio
🎯 What it does: This paper proposes a unified audio-video speech regeneration framework called ReVISE, which utilizes self-supervised speech units to predict speech content from damaged audio or silent video and regenerate high-quality speech.
Revisiting Multimodal Representation in Contrastive Learning: From Patch and Token Embeddings to Finite Discrete Tokens
Yuxiao Chen (Rutgers University), Hongxia Yang (Zhejiang University)
RetrievalRepresentation LearningTransformerContrastive LearningImageTextMultimodality
🎯 What it does: A contrastive learning framework is proposed that uses finite discrete tokens (FDT) as a shared base to map images and text into a unified discrete space, achieving cross-modal alignment.