CVPR 2025 Papers — Page 21
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Proximal Algorithm Unrolling: Flexible and Efficient Reconstruction Networks for Single-Pixel Imaging
Ping Wang (Westlake University), Xin Yuan (Westlake University)
RestorationOptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a ProxUnroll method based on approximate gradients, utilizing proximal trajectory loss to train deep image restorers, enabling them to approach explicit proximal operators in single-pixel imaging (SPI), thus achieving fast and accurate reconstruction.
ProxyTransformation: Preshaping Point Cloud Manifold With Proxy Attention For 3D Visual Grounding
Qihang Peng (Tsinghua University), Gao Huang (Tsinghua University)
Object DetectionRepresentation LearningTransformerVision Language ModelTextPoint Cloud
🎯 What it does: This paper proposes the Proxy Transformation method for multimodal-guided enhancement of point cloud submanifolds in 3D visual localization tasks, significantly improving the geometric structure and semantic information of point clouds.
PS-Diffusion: Photorealistic Subject-Driven Image Editing with Disentangled Control and Attention
Weicheng Wang (Nankai University), Jufeng Yang (Nankai University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The PS-Diffusion framework is proposed to achieve subject-driven image editing with contextual interactions such as lighting, shadows, and reflections, generating realistic target images.
PS-EIP: Robust Photometric Stereo Based on Event Interval Profile
Kazuma Kitazawa (University of Tsukuba), Tsuyoshi Takatani (University of Tsukuba)
Depth EstimationMesh
🎯 What it does: The PS-EIP method is proposed, which utilizes the event interval time series (event interval profile) captured by an event camera under continuous light source movement for photometric stereo reconstruction, thereby achieving pixel-level surface normal estimation.
PSA-SSL: Pose and Size-aware Self-Supervised Learning on LiDAR Point Clouds
Barza Nisar (University of Toronto), Steven L. Waslander (University of Toronto)
Object DetectionSegmentationAutonomous DrivingRepresentation LearningContrastive LearningPoint Cloud
🎯 What it does: This paper proposes a point cloud representation learning framework PSA-SSL based on self-supervised contrastive learning, which significantly improves the performance of point cloud semantic segmentation and object detection by introducing a 3D bounding box regression pre-training task and LiDAR beam pattern augmentation.
PSBD: Prediction Shift Uncertainty Unlocks Backdoor Detection
Wei Li (Illinois Institute of Technology), Ren Wang (Illinois Institute of Technology)
ClassificationAnomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: A method for detecting backdoor samples in the training set using the prediction shift uncertainty (PSU) generated by Dropout during inference is proposed (PSBD).
Pseudo Visible Feature Fine-Grained Fusion for Thermal Object Detection
Ting Li (University of Electronic Science and Technology of China), Luping Ji (University of Electronic Science and Technology of China)
Object DetectionAutonomous DrivingGraph Neural NetworkGenerative Adversarial NetworkImageMultimodality
🎯 What it does: A pseudo-visible feature fine-grained fusion (PFGF) method is proposed for thermal image object detection, utilizing a prior thermal-to-visible (T2V) translation model to generate pseudo-visible features, which are fused with multi-scale thermal features through a graph neural network.
PSHuman: Photorealistic Single-image 3D Human Reconstruction using Cross-Scale Multiview Diffusion and Explicit Remeshing
Peng Li (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)
GenerationData SynthesisPose EstimationDiffusion modelImageMesh
🎯 What it does: We propose a framework for full-body 3D reconstruction from a single image based on diffusion models, called PSHuman. It utilizes cross-scale multi-view diffusion to generate color and normal images, and achieves high-quality, textured human meshes through explicit mesh sculpting initialized by SMPL-X.
PTDiffusion: Free Lunch for Generating Optical Illusion Hidden Pictures with Phase-Transferred Diffusion Model
Xiang Gao (Peking University), Jiaying Liu (Peking University)
Image TranslationGenerationDiffusion modelImageText
🎯 What it does: A text-guided image-to-image translation framework called PTDiffusion has been developed to generate optical illusion hidden images by integrating reference images into arbitrary scenes while preserving the semantic meaning of the text.
PUP 3D-GS: Principled Uncertainty Pruning for 3D Gaussian Splatting
Alex Hanson (University of Maryland), Tom Goldstein (University of Maryland)
CompressionOptimizationComputational EfficiencyGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes a post-processing pruning method called PUP 3D-GS, which can compress the number of Gaussians in the 3D Gaussian Splatting model by over 90% without altering the training process, while maintaining or even improving image quality.
PURA: Parameter Update-Recovery Test-Time Adaption for RGB-T Tracking
Zekai Shao (University of Science and Technology Beijing), Hongmin Liu (University of Science and Technology Beijing)
Object TrackingDomain AdaptationImageVideo
🎯 What it does: The PURA framework is proposed, which quickly updates BatchNorm parameters using statistical information during testing, and recovers the main direction parameters through singular value decomposition of pseudo-gradients, achieving online domain shift adaptation for RGB-T tracking.
Pursuing Temporal-Consistent Video Virtual Try-On via Dynamic Pose Interaction
Dong Li (HiDream.ai Inc.), Tao Mei (HiDream.ai Inc.)
GenerationData SynthesisPose EstimationDiffusion modelImageVideo
🎯 What it does: A video virtual try-on framework DPIDM based on diffusion models is proposed, achieving human-clothing pose interaction through a skeleton pose adapter and hierarchical attention module.
PVC: Progressive Visual Token Compression for Unified Image and Video Processing in Large Vision-Language Models
Chenyu Yang (Tsinghua University), Jifeng Dai (Tsinghua University)
CompressionTransformerVision Language ModelImageVideoText
🎯 What it does: Proposes Progressive Visual Token Compression (PVC), treating images as 'static videos' to unify the hierarchical compression and encoding of images and videos in the Vision-Language Model.
PyTorchGeoNodes: Enabling Differentiable Shape Programs for 3D Shape Reconstruction
Sinisa Stekovic (École des Ponts et Chaussées), Friedrich Fraundorfer (Graz University of Technology)
OptimizationGaussian SplattingPoint CloudMesh
🎯 What it does: Proposes the PyTorchGeoNodes framework, which compiles Blender shape programs into differentiable PyTorch code, and combines genetic algorithms with gradient optimization to achieve differentiable reconstruction of object geometry and parameters in RGB-D scans.
Q-Bench-Video: Benchmark the Video Quality Understanding of LMMs
Zicheng Zhang (Shanghai Jiaotong University), Guangtao Zhai (Nanyang Technological University)
Large Language ModelVideoMultimodalityBenchmark
🎯 What it does: A Q-Bench-Video benchmark has been established for the systematic evaluation of large multimodal models in video quality perception.
Q-DiT: Accurate Post-Training Quantization for Diffusion Transformers
Lei Chen (Tsinghua University), Wenwu Zhu (Tsinghua University)
GenerationData SynthesisTransformerDiffusion modelImageVideo
🎯 What it does: This paper proposes Q-DiT, a post-training quantization method for Diffusion Transformers, addressing significant variance in weights/activations across input channels and variations in activation over time steps.
Q-Eval-100K: Evaluating Visual Quality and Alignment Level for Text-to-Vision Content
Zicheng Zhang (Shanghai Jiao Tong University), Guangtao Zhai (Meituan)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageVideoMultimodalityBenchmark
🎯 What it does: A large text-visual content evaluation dataset, Q-Eval-100K, has been proposed, and a unified evaluation model, Q-Eval-Score, has been trained based on this dataset, which can provide separate scores for visual quality and alignment.
Q-PART: Quasi-Periodic Adaptive Regression with Test-time Training for Pediatric Left Ventricular Ejection Fraction Regression
Jie Liu (City University of Hong Kong), Haoliang Li (Technische Universität München)
VideoBiomedical DataUltrasound
🎯 What it does: Proposes the Q-PART framework, which performs regression on pediatric left ventricular ejection fraction (LVEF) during the training phase at testing time, and divides echocardiograms into periodic and non-periodic components through latent space decomposition;
QMambaBSR: Burst Image Super-Resolution with Query State Space Model
Xin Di (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RestorationSuper ResolutionConvolutional Neural NetworkTransformerImage
🎯 What it does: A new burst image super-resolution network QMambaBSR has been designed and implemented to extract sub-pixel details from low-resolution burst image sequences and generate high-quality high-resolution images.
Quad-Pixel Image Defocus Deblurring: A New Benchmark and Model
Hang Chen (OMNIVISION), Chengming Liu (OMNIVISION)
RestorationConvolutional Neural NetworkImageBenchmark
🎯 What it does: This paper proposes a depth-of-field defocus deblurring method based on a four-pixel (QP) camera and constructs a new QPDD dataset, covering 4,935 pairs of RAW and sRGB images;
Quaffure: Real-Time Quasi-Static Neural Hair Simulation
Tuur Stuyck (Meta Reality Labs), Doug Roble (Meta Reality Labs)
GenerationOptimizationComputational EfficiencyConvolutional Neural NetworkAuto EncoderImageComputed Tomography
🎯 What it does: A real-time quasi-static neural hair simulation method is proposed, capable of generating naturally falling hair based on head and body shapes and postures within milliseconds.
Quantization without Tears
Minghao Fu (Nanjing University), Jianxin Wu (Nanjing University)
ClassificationObject DetectionSegmentationGenerationTransformerLarge Language ModelDiffusion modelImageText
🎯 What it does: The Quantization without Tears (QwT) method is proposed, which incorporates a lightweight linear compensation module into the quantization network to eliminate information loss caused by quantization, thereby achieving a high-precision, low-resource quantized model.
QuartDepth: Post-Training Quantization for Real-Time Depth Estimation on the Edge
Xuan Shen (Northeastern University), Jiuxiang Gu (Adobe Research)
Depth EstimationComputational EfficiencyImage
🎯 What it does: This paper proposes the QuartDepth framework, which implements post-training quantization of 4-bit weights/activations on edge ASICs to support real-time monocular depth estimation.
QuCOOP: A Versatile Framework for Solving Composite and Binary-Parametrised Problems on Quantum Annealers
Natacha Kuete Meli (University of Siegen), Michael Moeller (MPI for Informatics)
OptimizationPoint CloudMesh
🎯 What it does: The QuCOOP framework is proposed, utilizing Adiabatic Quantum Computing (AQC) to iteratively construct local QUBO forms to solve composite and binary parameterized non-quadratic objective functions, thereby extending the applicability of traditional AQC to QUBO.
Query Efficient Black-Box Visual Prompting with Subspace Learning
Zhaogeng Liu (Jilin University), Yi Chang (Jilin University)
ClassificationOptimizationComputational EfficiencyTransformerPrompt EngineeringAuto EncoderImage
🎯 What it does: In the black-box visual prompt learning task, a query-efficient framework based on subspace learning is proposed, utilizing pre-trained models and low-dimensional subspaces to generate input-dependent visual prompts.
Question-Aware Gaussian Experts for Audio-Visual Question Answering
Hongyeob Kim (Sungkyunkwan University), Sungeun Hong (Sungkyunkwan University)
RecognitionTransformerMixture of ExpertsVideoMultimodalityAudio
🎯 What it does: The QA-TIGER framework is proposed for audio-visual question answering, which enhances the accuracy of question answering by explicitly injecting question information during the encoding phase and utilizing a mixture of Gaussian experts model for continuous time dynamic modeling.
R-SCoRe: Revisiting Scene Coordinate Regression for Robust Large-Scale Visual Localization
Xudong Jiang (ETH Zurich), Marc Pollefeys (Microsoft)
Pose EstimationRetrievalOptimizationGraph Neural NetworkSimultaneous Localization and MappingImage
🎯 What it does: A robust large-scale visual localization method R-SCoRe is proposed, achieving high-precision localization based on scene coordinate regression, with a map size of only several tens of megabytes.
R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning
Lijun Sheng (University of Science and Technology of China), Ran He (University of Science and Technology of China)
ClassificationAdversarial AttackTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: A robust testing prompt tuning method for CLIP, called R-TPT, is proposed to enhance the model's defense capability against adversarial attacks during inference.
R2C: Mapping Room to Chessboard to Unlock LLM As Low-Level Action Planner
Ziyi Bai (Chinese Academy of Sciences), Xilin Chen (Chinese Academy of Sciences)
Robotic IntelligenceTransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: The Room to Chessboard (R2C) framework is proposed, mapping the room environment to a chessboard, allowing LLMs to both decompose high-level tasks and generate precise low-level actions.
RaCFormer: Towards High-Quality 3D Object Detection via Query-based Radar-Camera Fusion
Xiaomeng Chu (University of Science and Technology of China), Yanyong Zhang (University of Science and Technology of China)
Object DetectionAutonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: High-quality 3D object detection achieved through query-based radar-camera fusion
RAD: Region-Aware Diffusion Models for Image Inpainting
Sora Kim (Hanyang University), Minsik Lee (Hanyang University)
RestorationDiffusion modelImage
🎯 What it does: This paper proposes Region-Aware Diffusion Models (RAD), which allows the diffusion model to locally and asynchronously complete image inpainting by setting different noise schedules for each pixel.
Radio Frequency Ray Tracing with Neural Object Representation for Enhanced RF Modeling
Xingyu Chen (University of California San Diego), Xinyu Zhang (University of California San Diego)
Neural Radiance FieldPoint CloudMesh
🎯 What it does: The RFScape framework is proposed, which combines SDF-based neural object representation with traditional ray tracing for accurate and editable radio frequency propagation modeling.
RADIOv2.5: Improved Baselines for Agglomerative Vision Foundation Models
Greg Heinrich (NVIDIA), Pavlo Molchanov (NVIDIA)
RecognitionObject DetectionSegmentationTransformerVision Language ModelContrastive LearningImage
🎯 What it does: This paper presents RADIOv2.5, a visual foundation model that maintains high quality at any resolution through multi-teacher fusion (such as CLIP, DINOv2, SAM, etc.), and further enhances performance using techniques like multi-resolution training, mosaic augmentation, PHI-S normalization, and Token Merging.
RAEncoder: A Label-Free Reversible Adversarial Examples Encoder for Dataset Intellectual Property Protection
Fan Xing (Hainan University), Xiaoyi Zhou (Hainan University)
Adversarial AttackData-Centric LearningConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A label-free reversible adversarial sample encoder, RAEncoder, is proposed to protect the intellectual property of datasets.
RainyGS: Efficient Rain Synthesis with Physically-Based Gaussian Splatting
Qiyu Dai (Peking University), Mengyu Chu (Peking University)
GenerationData SynthesisComputational EfficiencyGaussian SplattingImageVideoPhysics Related
🎯 What it does: Combining physical modeling with 3D Gaussian Splatting to achieve controllable and real-time rain effect synthesis in open-world scenes.
RandAR: Decoder-only Autoregressive Visual Generation in Random Orders
Ziqi Pang (University of Illinois at Urbana-Champaign), Yu-Xiong Wang (University of Illinois at Urbana-Champaign)
RestorationGenerationData SynthesisTransformerLarge Language ModelImage
🎯 What it does: We propose RandAR, a decoder-only Transformer capable of autoregressive image generation in arbitrary order; by inserting position instruction tokens before each image token to enable the model to learn to generate in random order; further achieving zero-shot capabilities such as parallel decoding, unsupervised inpainting, outpainting, resolution extrapolation, and bidirectional feature extraction.
Random Conditioning for Diffusion Model Compression with Distillation
Dohyun Kim (Korea University), Paul Hongsuck Seo (Korea University)
GenerationCompressionKnowledge DistillationDiffusion modelImageText
🎯 What it does: A technique called Random Conditioning is proposed, which efficiently compresses and distills knowledge from teacher diffusion models by pairing noise images with random text conditions in the absence of images, enabling the learning and generation of unseen concepts.
RANGE: Retrieval Augmented Neural Fields for Multi-Resolution Geo-Embeddings
Aayush Dhakal (Washington University in St. Louis), Nathan Jacobs (Washington University in St. Louis)
ClassificationRetrievalContrastive LearningImage
🎯 What it does: This paper proposes a retrieval-enhanced geographic location embedding method (RANGE), which utilizes high-resolution satellite image features of similar locations to supplement the high-frequency visual information lost in contrastive learning, thereby generating more expressive multi-scale geographic embeddings.
RAP: Retrieval-Augmented Personalization for Multimodal Large Language Models
Haoran Hao (Chinese University of Hong Kong), Xiangyu Yue (Beijing Institute of Technology)
GenerationRetrievalRecommendation SystemTransformerLarge Language ModelVision Language ModelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: This paper proposes the RAP (Retrieval-Augmented Personalization) framework, which utilizes external knowledge bases to store, retrieve, and inject personalized user information, enabling multimodal large language models (MLLMs) to retain their original capabilities while being able to memorize, retrieve, and generate content that includes user-specific concepts in real-time.
Rashomon Sets for Prototypical-Part Networks: Editing Interpretable Models in Real-Time
Jon Donnelly (Duke University), Cynthia Rudin (Duke University)
ClassificationExplainability and InterpretabilityComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Proposes the Proto-RSet framework, which can quickly generate and interactively use various equivalent performance ProtoPNet models (i.e., the Rashomon set), addressing the interactive bottleneck in model debugging, supporting real-time insertion/deletion of prototypes while ensuring accuracy.
RASP: Revisiting 3D Anamorphic Art for Shadow-Guided Packing of Irregular Objects
Soumyaratna Debnath (Indian Institute of Technology Gandhinagar), Shanmuganathan Raman (Indian Institute of Technology Gandhinagar)
OptimizationMesh
🎯 What it does: A framework based on differentiable rendering, RASP, is proposed for packing, component assembly, and multi-view artistic creation of arbitrary-shaped 3D objects guided by shadows/projections.
RaSS: Improving Denoising Diffusion Samplers with Reinforced Active Sampling Scheduler
Xin Ding (University of Science and Technology of China), Zhibo Chen (University of Science and Technology of China)
GenerationOptimizationReinforcement LearningDiffusion modelImage
🎯 What it does: Proposes an adaptive sampling scheduler RaSS based on reinforcement learning, dynamically planning the step size for each sampling process.
Rate-In: Information-Driven Adaptive Dropout Rates for Improved Inference-Time Uncertainty Estimation
Tal Zeevi (Yale University), John A. Onofrey (Yale University)
Biomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: During the inference phase, adaptive adjustment of Monte Carlo Dropout in neural networks is implemented to achieve more accurate predictions of uncertainty estimation.
RayFlow: Instance-Aware Diffusion Acceleration via Adaptive Flow Trajectories
Huiyang Shao (Tsinghua University), Xuefeng Xiao (ByteDance Inc.)
GenerationComputational EfficiencyDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: The RayFlow framework is proposed, which accelerates diffusion sampling by assigning a unique target mean and adaptive flow trajectory to each sample.
RC-AutoCalib: An End-to-End Radar-Camera Automatic Calibration Network
Van-Tin Luu (National Yang Ming Chiao Tung University), Ching-Chun Huang (National Yang Ming Chiao Tung University)
Autonomous DrivingConvolutional Neural NetworkRecurrent Neural NetworkSimultaneous Localization and MappingMultimodalityPoint Cloud
🎯 What it does: An end-to-end RC-AutoCalib network is designed to achieve online self-calibration of 3D millimeter-wave radar and cameras.
RCP-Bench: Benchmarking Robustness for Collaborative Perception Under Diverse Corruptions
Shihang Du (Tongji University), Guang Chen (Tongji University)
Object DetectionAutonomous DrivingImageBenchmark
🎯 What it does: This paper proposes RCP-Bench, a comprehensive benchmark for evaluating the robustness of multi-vehicle collaborative perception under various real-world distortions (such as weather, camera failures, and time mismatches), and presents two training strategies, RCP-Drop and RCP-Mix, to enhance robustness for this benchmark.
RDD: Robust Feature Detector and Descriptor using Deformable Transformer
Gonglin Chen (University of Southern California), Yajie Zhao (University of Southern California)
RecognitionObject DetectionKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: A two-branch network named RDD is proposed, which uses convolutional networks to detect key points and employs a deformable Transformer to generate robust descriptors, achieving robust feature detection and description.
Re-HOLD: Video Hand Object Interaction Reenactment via adaptive Layout-instructed Diffusion Model
Yingying Fan (Wuhan University), Jingdong Wang (Baidu Inc.)
RestorationGenerationDiffusion modelVideo
🎯 What it does: Proposes the Re-HOLD framework to achieve high-fidelity video reproduction based on hand-object interaction.
Re-thinking Temporal Search for Long-Form Video Understanding
Jinhui Ye (Stanford University), Manling Li (Stanford University)
Object DetectionRetrievalComputational EfficiencyVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: A lightweight framework T* is proposed to transform long video temporal search into spatial search, and the LV-HAYSTACK dataset is constructed to evaluate long video retrieval performance.
Real-IAD D3: A Real-World 2D/Pseudo-3D/3D Dataset for Industrial Anomaly Detection
Wenbing Zhu (Fudan University), Lizhuang Ma (Shanghai Jiao Tong University)
Anomaly DetectionContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: A Real-IAD D³ tri-modal industrial defect detection dataset (RGB, photometric stereo pseudo-3D, point cloud) is proposed, and based on this, a D³M multi-modal fusion model is designed to validate its effectiveness in industrial defect detection and localization tasks.
Real-time Free-view Human Rendering from Sparse-view RGB Videos using Double Unprojected Textures
Guoxing Sun (Max Planck Institute for Informatics), Marc Habermann (Max Planck Institute for Informatics)
SegmentationGenerationConvolutional Neural NetworkGaussian SplattingVideo
🎯 What it does: A method called Double Unprojected Textures (DUT) is proposed, which enables real-time generation of high-resolution (4K) portrait free-viewpoint rendering from sparse RGB videos.
Real-time High-fidelity Gaussian Human Avatars with Position-based Interpolation of Spatially Distributed MLPs
Youyi Zhan, Kun Zhou
GenerationPose EstimationComputational EfficiencyGaussian SplattingVideo
🎯 What it does: A Gaussian human head representation based on spatially distributed MLP is proposed, capable of real-time rendering of high-fidelity appearance dependent on pose.
RealEdit: Reddit Edits As a Large-scale Empirical Dataset for Image Transformations
Peter Sushko (University of Washington), Ranjay Krishna (University of Washington)
Image TranslationRestorationSupervised Fine-TuningDiffusion modelImage
🎯 What it does: This paper creates a large-scale real-world image editing dataset named REALEDIT and fine-tunes a text-guided image editing model on it.
Realistic Test-Time Adaptation of Vision-Language Models
Maxime Zanella (UCLouvain), Ismail Ben Ayed
Domain AdaptationTransformerVision Language ModelImage
🎯 What it does: This paper proposes a real-time adaptation method for visual language models, Stat A, aimed at maintaining zero-shot performance in more realistic deployment scenarios.
Reanimating Images using Neural Representations of Dynamic Stimuli
Jacob Yeung (Carnegie Mellon University), Michael J. Tarr (Carnegie Mellon University)
GenerationData SynthesisDiffusion modelOptical FlowVideoMagnetic Resonance Imaging
🎯 What it does: By decoupling static image features from motion information, the fMRI signals of subjects are used to predict video optical flow, and the predicted optical flow is input into a motion-conditioned diffusion model to achieve image reanimation based on brain activity.
Reason-before-Retrieve: One-Stage Reflective Chain-of-Thoughts for Training-Free Zero-Shot Composed Image Retrieval
Yuanmin Tang (Institute of Information Engineering), Qi Wu (University of Adelaide)
RetrievalTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: This paper proposes a one-stage reflective chain reasoning method (OSrCIR) that directly utilizes multimodal large language models to complete the entire reasoning process from reference images and text descriptions to target image retrieval without training.
ReasonGrounder: LVLM-Guided Hierarchical Feature Splatting for Open-Vocabulary 3D Visual Grounding and Reasoning
Zhenyang Liu (Fudan University), Xiangyang Xue (Fudan University)
Object DetectionSegmentationTransformerLarge Language ModelContrastive LearningGaussian SplattingPoint Cloud
🎯 What it does: Proposes the ReasonGrounder framework, which utilizes large visual language model (LVLM)-guided hierarchical feature Gaussian rendering to achieve open vocabulary 3D visual localization and reasoning, supporting automatic localization of implicit instructions and occluded scenes.
Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach
Steeven Janny (Naver Labs Europe), Christian Wolf (Naver Labs Europe)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: Conduct large-scale experiments and analysis on end-to-end visual navigation agents based on reinforcement learning, studying their internal dynamics, memory, planning capabilities, and validating performance on real robots.
Reasoning Mamba: Hypergraph-Guided Region Relation Calculating for Weakly Supervised Affordance Grounding
Yuxuan Wang (Xidian University), Cheng Deng (Xidian University)
Object DetectionSegmentationGraph Neural NetworkImage
🎯 What it does: A weakly supervised object affordance localization framework R-Mamba is proposed, which achieves precise localization of affordance regions using hypergraphs and state space models.
Reasoning to Attend: Try to Understand How <SEG> Token Works
Rui Qian (Fudan University), Dejing Dou (Fudan University)
RecognitionSegmentationTransformerSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: Token Works
ReCap: Better Gaussian Relighting with Cross-Environment Captures
Jingzhi Li (University of Wurzburg), Radu Timofte (University of Wurzburg)
RestorationOptimizationGaussian SplattingPoint Cloud
🎯 What it does: The ReCap method is proposed, utilizing multi-task supervision captured across environments to jointly optimize shared material properties and multiple learnable environment maps, achieving 3D Gaussian lighting reconstruction and relighting.
ReCapture: Generative Video Camera Controls for User-Provided Videos using Masked Video Fine-Tuning
David Junhao Zhang (Google), Nataniel Ruiz (Google)
RestorationGenerationData SynthesisDiffusion modelVideoPoint CloudStochastic Differential Equation
🎯 What it does: Given a video shot by a user, a two-stage process is used to generate a new camera trajectory and produce a corresponding new video, while preserving the original scene motion and content.
Recognition-Synergistic Scene Text Editing
Zhengyao Fang (Harbin Institute of Technology), Wenjie Pei (Harbin Institute of Technology)
RecognitionImage TranslationTransformerDiffusion modelImage
🎯 What it does: A unified scene text editing method RS-STE is proposed, which utilizes the implicit style-content separation of the recognition model to achieve text content replacement while maintaining style.
ReCon: Enhancing True Correspondence Discrimination through Relation Consistency for Robust Noisy Correspondence Learning
Quanxing Zha (Huaqiao University), Nannan Wang (Xidian University)
RetrievalContrastive LearningImageTextMultimodality
🎯 What it does: A relationship consistency learning framework named ReCon is proposed to address the noisy correspondence problem in cross-modal retrieval and improve the recognition and retrieval performance of true correspondences.
Reconciling Stochastic and Deterministic Strategies for Zero-shot Image Restoration using Diffusion Model in Dual
Chong Wang (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: A framework called RDMD is proposed, which unifies the use of a single pre-trained diffusion model for deterministic regression and random sampling in zero-shot image restoration tasks.
ReconDreamer: Crafting World Models for Driving Scene Reconstruction via Online Restoration
Chaojun Ni (Peking University), Wenjun Mei (Peking University)
RestorationAutonomous DrivingDiffusion modelNeural Radiance FieldGaussian SplattingWorld ModelVideo
🎯 What it does: This paper proposes the ReconDreamer framework, which significantly improves the rendering quality of dynamic driving scenes during large lane changes (multi-lane movement) through online restoration and progressive data updates, addressing the artifacts and detail loss issues that traditional NeRF/3DGS encounter on new trajectories.
Reconstructing Animals and the Wild
Peter Kulits (Max Planck Institute for Intelligent Systems), Silvia Zuffi (IMATI-CNR)
GenerationData SynthesisTransformerLarge Language ModelImage
🎯 What it does: This paper proposes a single-image natural scene inverse image reconstruction method based on large language models (LLM), capable of simultaneously reconstructing animals and their surrounding environments, outputting editable and animatable 3D scene code.
Reconstructing Close Human Interaction with Appearance and Proxemics Reasoning
Buzhen Huang (National University of Singapore), Gim Hee Lee (National University of Singapore)
Pose EstimationOptimizationDiffusion modelGaussian SplattingVideo
🎯 What it does: A dual-branch optimization framework is proposed, utilizing human appearance, social proximity, and physical constraints to reconstruct close-range interactive actions and appearances of two people from monocular outdoor videos.
Reconstructing Humans with a Biomechanically Accurate Skeleton
Yan Xia (University of Texas at Austin), Georgios Pavlakos (Zhejiang University)
Pose EstimationTransformerImage
🎯 What it does: A method is proposed for 3D human reconstruction by training a Transformer to regress the parameters of the SKEL skeletal model from a single image.
Reconstructing In-the-Wild Open-Vocabulary Human-Object Interactions
Boran Wen (Shanghai Jiao Tong University), Yong-Lu Li (Shanghai Jiao Tong University)
Object DetectionPose EstimationDepth EstimationOptimizationGaussian SplattingImagePoint Cloud
🎯 What it does: Construct fine-grained 3D human-object interaction data from a single real scene image, and propose a Gaussian-based untrained optimizer for reconstructing three-dimensional interaction relationships.
Reconstructing People, Places, and Cameras
Lea Müller, Angjoo Kanazawa (University of California Berkeley)
Pose EstimationOptimizationTransformerPoint CloudMesh
🎯 What it does: This paper proposes the HSfM method, which jointly reconstructs human meshes, scene point clouds, and camera parameters from sparse uncalibrated multi-view images, achieving globally consistent metric scale reconstruction.
Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models
Jingfeng Yao (Huazhong University of Science and Technology), Xinggang Wang (Huazhong University of Science and Technology)
GenerationOptimizationTransformerDiffusion modelRectified FlowAuto EncoderImage
🎯 What it does: The VA-VAE introduces a VF loss aligned through the Vision Foundation model, addressing the optimization dilemma between reconstruction and generation in high-dimensional tokenizers, and proposes LightningDiT to accelerate training of the Diffusion Transformer.
Recover and Match: Open-Vocabulary Multi-Label Recognition through Knowledge-Constrained Optimal Transport
Hao Tan (Institute of Automation, Chinese Academy of Sciences), Zhen Lei (Institute of Automation, Chinese Academy of Sciences)
ClassificationRecognitionVision Language ModelContrastive LearningImage
🎯 What it does: Proposes the RAM framework, which recovers the local semantics of CLIP through the Ladder Local Adapter and uses Knowledge-Constrained Optimal Transport for precise matching of image regions and labels, achieving open vocabulary multi-label recognition.
Recovering Dynamic 3D Sketches from Videos
Jaeah Lee (Seoul National University), Jaesik Park (Seoul National University)
GenerationOptimizationVideoPoint Cloud
🎯 What it does: Recovering the 3D motion of objects from videos and abstracting and visualizing the motion in a sparse form using deformable 3D Bézier curves (3D strokes).
Rectification-specific Supervision and Constrained Estimator for Online Stereo Rectification
Rui Gong (Nanyang Technological University), Jun Cheng (Institute for Infocomm Research)
Pose EstimationDepth EstimationOptimizationOptical FlowImage
🎯 What it does: This paper proposes an online stereo disparity correction framework that combines a semi-dense matcher and a direct optimization estimator for correction constraints, without the need for training with real correspondences.
Rectified Diffusion Guidance for Conditional Generation
Mengfei Xia (Tsinghua University), Yong-Jin Liu (Tsinghua University)
GenerationData SynthesisDiffusion modelImageTextStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes an algorithm for correcting classifier-free guidance (ReCFG), which addresses the theoretical flaw of expected bias produced by traditional CFG in diffusion model sampling, achieving more accurate conditional sampling without the need for retraining.
Recurrence-Enhanced Vision-and-Language Transformers for Robust Multimodal Document Retrieval
Davide Caffagni (University of Modena and Reggio Emilia), Rita Cucchiara (Istituto Italiano di Tecnologia)
RetrievalRecurrent Neural NetworkTransformerVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: A cross-modal retrieval framework named ReT is proposed, supporting multi-modal queries and documents that include images and text, utilizing multi-layer visual and textual representations for retrieval.
Recurrent Feature Mining and Keypoint Mixup Padding for Category-Agnostic Pose Estimation
Junjie Chen (Jiangxi University of Finance and Economics), Yuming Fang (Jiangxi University of Finance and Economics)
Pose EstimationTransformerImage
🎯 What it does: A recursive fine-grained and structure-aware feature mining framework is proposed, introducing keypoint Mixup filling to address the category inconsistency in pose estimation tasks.
Redefining <Creative> in Dictionary: Towards an Enhanced Semantic Understanding of Creative Generation
Fu Feng (Southeast University), Xin Geng (Southeast University)
GenerationDiffusion modelText
🎯 What it does: Proposes redefining 'creative' as a new token <CreTok>, enabling the diffusion model to directly generate composite creative images in a zero-shot manner through image-free training.
ReDiffDet: Rotation-equivariant Diffusion Model for Oriented Object Detection
Jiaqi Zhao (China University of Mining and Technology), Rui Yao (China University of Mining and Technology)
Object DetectionConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A rotation-equivariant directed object detection framework called ReDiffDet is proposed, which uses a two-dimensional Gaussian distribution to represent directed boxes and performs denoising diffusion inference.
Reducing Class-wise Confusion for Incremental Learning with Disentangled Manifolds
Huitong Chen (Tianjin University), Qinghua Hu (Tianjin University)
ClassificationKnowledge DistillationRepresentation LearningConvolutional Neural NetworkAuto EncoderContrastive LearningImage
🎯 What it does: This paper proposes an incremental learning method (CREATE) that utilizes lightweight autoencoders to learn class-separating low-dimensional manifolds, aiming to reduce inter-class confusion and alleviate catastrophic forgetting.
Ref-GS: Directional Factorization for 2D Gaussian Splatting
Youjia Zhang (Huazhong University of Science and Technology), Wei Yang (Huazhong University of Science and Technology)
GenerationData SynthesisComputational EfficiencyGaussian SplattingImage
🎯 What it does: This paper proposes Ref-GS, a directional factorization method for 2D Gaussian spot rendering that accurately recovers scene geometry while maintaining high-quality view-dependent reflections.
Reference-Based 3D-Aware Image Editing with Triplanes
Bahri Batuhan Bilecen (Bilkent University), Aysegul Dundar (Bilkent University)
Image TranslationGenerationImage
🎯 What it does: A 3D perception image editing framework based on reference images is proposed, utilizing triplane space to achieve high-quality, 3D-consistent local attribute transfer and fusion.
RefPose: Leveraging Reference Geometric Correspondences for Accurate 6D Pose Estimation of Unseen Objects
Jaeguk Kim (Seoul National University), Nam Ik Cho (Seoul National University)
Pose EstimationOptical FlowImageBenchmark
🎯 What it does: A two-stage 6D pose estimation framework called RefPose is proposed, designed specifically for unseen objects based on reference images and geometric correspondences.
Relation-Rich Visual Document Generator for Visual Information Extraction
Zi-Han Jiang (National Taiwan University), Chu-Song Chen (National Taiwan University)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelImageText
🎯 What it does: This paper proposes RIDGE, a two-stage relationship-rich visual document generator that first uses LLM to generate text content with hierarchical relationships, and then generates diverse, layout-logical document images through self-supervised learning.
Relation3D : Enhancing Relation Modeling for Point Cloud Instance Segmentation
Jiahao Lu (University of Science and Technology of China), Jiacheng Deng (University of Science and Technology of China)
Object DetectionSegmentationTransformerContrastive LearningPoint Cloud
🎯 What it does: Proposes the Relation3D method, achieving point cloud instance segmentation based on Transformer.
RelationField: Relate Anything in Radiance Fields
Sebastian Koch, Timo Ropinski
SegmentationKnowledge DistillationLarge Language ModelNeural Radiance FieldPoint Cloud
🎯 What it does: Proposes RelationField, which can directly extract open vocabulary object relationships in neural radiance fields and support relationship queries.
Relative Pose Estimation through Affine Corrections of Monocular Depth Priors
Yifan Yu (ETH Zurich), Viktor Larsson (Lund University)
Pose EstimationDepth EstimationSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A depth prior using a monocular depth estimation model is proposed, and the relative pose of the camera is solved through explicit scale and offset correction.
Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization
Siyan Dong (University of Hong Kong), Yanchao Yang (University of Hong Kong)
Pose EstimationTransformerSimultaneous Localization and MappingImage
🎯 What it does: Proposes the Reloc3r framework, which achieves efficient and generalizable visual localization through a large-scale relative pose regression network and a minimal motion averaging module;
RELOCATE: A Simple Training-Free Baseline for Visual Query Localization Using Region-Based Representations
Savya Khosla (University of Illinois Urbana-Champaign), Derek Hoiem (University of Illinois Urbana-Champaign)
Object DetectionObject TrackingRetrievalTransformerContrastive LearningVideo
🎯 What it does: A training-free visual query localization method called RELOCATE is proposed, which can locate the last appearance of a specified object in long videos.
Remote Photoplethysmography in Real-World and Extreme Lighting Scenarios
Hang Shao (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
TransformerContrastive LearningVideo
🎯 What it does: An end-to-end remote photoplethysmography (rPPG) model based on visual Transformer is proposed, capable of non-contact measurement of physiological indicators such as heart rate under extreme lighting and complex interference scenarios in the real world.
Removing Reflections from RAW Photos
Eric Kee (Adobe Inc), Marc Levoy (Adobe Inc)
Image TranslationRestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Train an end-to-end RAW camera-level reflection removal system that utilizes optional reference photos as context to output clear and reflection-free images.
ReNeg: Learning Negative Embedding with Reward Guidance
Xiaomin Li (Dalian University of Technology), Emad Barsoum (Advanced Micro Devices)
GenerationData SynthesisOptimizationReinforcement LearningDiffusion modelImageVideoText
🎯 What it does: A negative embedding learning framework called ReNeg is proposed, which automatically optimizes negative embeddings through gradient descent to enhance the visual quality and human preference of text-to-image/video generation.
RENO: Real-Time Neural Compression for 3D LiDAR Point Clouds
Kang You (Nanjing University), Zhan Ma (University of California Riverside)
CompressionAutonomous DrivingPoint Cloud
🎯 What it does: This paper proposes a lightweight neural encoder named RENO, which achieves real-time compression of 3D LiDAR point clouds.
RePerformer: Immersive Human-centric Volumetric Videos from Playback to Photoreal Reperformance
Yuheng Jiang (ShanghaiTech University), Lan Xu (ShanghaiTech University)
GenerationData SynthesisConvolutional Neural NetworkGaussian SplattingVideo
🎯 What it does: For multi-view human dynamic scenes, a hierarchical representation based on Gaussian distribution (motion Gaussian + appearance Gaussian) is proposed, achieving a full-process video rendering from playback to reenactment (playback + new action reenactment).
Reproducible Vision-Language Models Meet Concepts Out of Pre-Training
Ziliang Chen (Peng Cheng Laboratory), Liang Lin (Sun Yat-sen University)
ClassificationRecognitionRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes the LAION-Beyond benchmark to study the generalization of CLIP to out-of-training concepts and introduces methods such as name fine-tuning, FSNL, and ZSNL to enhance the recognition of OOP concepts.
Repurposing Pre-trained Video Diffusion Models for Event-based Video Interpolation
Jingxi Chen (University of Maryland), Yiannis Aloimonos (University of Maryland)
RestorationGenerationData SynthesisDiffusion modelAuto EncoderVideo
🎯 What it does: Adapting the pre-trained video diffusion model (Stable Video Diffusion) to an event-driven video frame interpolation (EVFI) model, achieving high frame rate video generation by using events as motion guidance.
Repurposing Stable Diffusion Attention for Training-Free Unsupervised Interactive Segmentation
Markus Karmann (Vivo Tech Research), Onay Urfalioglu (Vivo Tech Research)
SegmentationTransformerDiffusion modelImage
🎯 What it does: This paper proposes a training-free, point-prompt interactive image segmentation framework M2N2, which constructs a Markov mapping using the self-attention of Stable Diffusion 2, and achieves segmentation through flood filling and truncated nearest neighbors.
ReRAW: RGB-to-RAW Image Reconstruction via Stratified Sampling for Efficient Object Detection on the Edge
Radu Berdan (Sony AI), Daisuke Iso (Sony AI)
Image TranslationRestorationObject DetectionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes the ReRAW model, which can reverse-generate high-quality RAW images from RGB images and pre-train edge object detectors using synthetic RAW data.
ResCLIP: Residual Attention for Training-free Dense Vision-language Inference
Yuhang Yang (University of Electronic Science and Technology of China), Lixin Duan (University of Electronic Science and Technology of China)
SegmentationTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes an untrained framework called ResCLIP, which utilizes the intermediate layer self-attention of the CLIP model to improve dense visual-language inference, particularly for open vocabulary semantic segmentation.