CVPR 2026 Papers — Page 30
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
Rethinking 2D-3D Registration: A Novel Network for High-Value Zone Selection and Representation Consistency Alignment
Zhixin Cheng (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
Pose EstimationConvolutional Neural NetworkTransformerReinforcement LearningContrastive LearningImagePoint Cloud
🎯 What it does: Propose a network that combines high-value region selection and cross-modal representation consistent alignment to improve the registration accuracy between images and point clouds.
Rethinking Asymmetric Quantization: Hidden Symmetry in Vision Model Weights
Masafumi Mori (DENSO CORPORATION), Mitsuru Ambai (DENSO IT Laboratory)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose Dense and Additive Sparse Quantization (DASQ), decomposing weight quantization into dense symmetric components and sparse outlier components, eliminating zero-point offset to achieve zero-point-free symmetric quantization.
Rethinking BCE Loss for Multi-Label Image Recognition with Fine-Tuning
Ao Zhou (Nanjing University), Qing Gu (Nanjing University)
ClassificationRecognitionSupervised Fine-TuningPrompt EngineeringVision Language ModelImageText
🎯 What it does: This paper proposes Class-wise Covariance Regularization (CCR), introducing structured calibration regularization in the fine-tuning of multi-label image recognition with CLIP to correct confidence distortion caused by BCE loss.
Rethinking Box Supervision: Bias-Free Weakly Supervised Medical Segmentation
Jun Wei (Shenzhen University), Hui Huang (Shenzhen University)
SegmentationConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: Developed the WeakMed framework, which trains medical image segmentation models using only box annotations, eliminating the reliance on pixel-level annotations and providing a low-cost, high-quality segmentation solution.
Rethinking Camera Choice: An Empirical Study on Fisheye Camera Properties in Robotic Manipulation
Han Xue (Shanghai Jiao Tong University), Chuan Wen (Shanghai Jiao Tong University)
Robotic IntelligenceConvolutional Neural NetworkTransformerVision-Language-Action ModelDiffusion modelImage
🎯 What it does: This paper systematically investigates the impact of wrist-mounted fisheye cameras in robot imitation learning through simulation and real-world experiments, focusing on three aspects: spatial localization, scene generalization, and hardware generalization.
Rethinking Concept Bottleneck Models: From Pitfalls to Solutions
Merve Tapli (Middle East Technical University), Emre Akbas (Middle East Technical University)
Explainability and InterpretabilityKnowledge DistillationRepresentation LearningTransformerVision Language ModelMultimodality
🎯 What it does: Propose the CBM-Suite method, systematically addressing four major defects of concept bottleneck models: concept relevance assessment, linear problems, accuracy gap, and selection of backbones and VLMs.
Rethinking Cross-Modal Anchor Alignment for Mitigating Error Accumulation
Bin Liu (Northwest A & F University), Haixi Zhang (Northwest A & F University)
RetrievalRepresentation LearningContrastive LearningMultimodality
🎯 What it does: Proposed a cross-modal matching framework (GSL) that combines frequency domain alignment, geometry-aware label correction, and semantic constraints to address error accumulation caused by noisy correspondences.
Rethinking Dataset Distillation: Hard Truths about Soft Labels
Priyam Dey (Indian Institute of Science), Venkatesh Babu Radhakrishnan (Indian Institute of Science)
ClassificationKnowledge DistillationData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: Systematically analyze the scalability and data quality impact of dataset distillation under different label supervision (hard labels HL, fixed soft labels SL, teacher soft labels SL+KD), propose DCS (Distillation Correlation Score) as a fast metric to evaluate distillation objectives, and design CAD-Prune sampling metric based on computational budget and CA2D dataset distillation method, significantly improving performance on ImageNet-1K HL.
Rethinking Diffusion Model-Based Video Super-Resolution: Leveraging Dense Guidance from Aligned Features
Jingyi Xu (Beihang University), Mai Xu (Beihang University)
Super ResolutionConvolutional Neural NetworkDiffusion modelOptical FlowVideo
🎯 What it does: Propose DGAF-VSR, which combines diffusion models with a densely feature-guided video super-resolution network;
Rethinking Glyph Spatial Information in Font Generation
Peng Su (Jilin University), Xi Yang (Jilin University)
GenerationDiffusion modelImageStochastic Differential Equation
🎯 What it does: Proposed the Spatial Preservation Rendering (SPR) scheme, constructed a large-scale OFL-licensed Chinese font dataset, and designed a two-stage GlyphSpatialNet model to achieve end-to-end generation of high-quality vector fonts from reference fonts.
Rethinking Intermediate Representation for VLM-based Robot Manipulation
Weiliang Tang (Chinese University of Hong Kong), Chi-Wing Fu (Chinese University of Hong Kong)
OptimizationRobotic IntelligenceTransformerVision Language ModelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Designed a new intermediate representation SEAM, combining context-free grammar to balance the interpretability of VLM and the generalization of actions, and proposed a RAG-based few-shot open-vocabulary segmentation scheme for precise object part localization.
Rethinking Knowledge Transfer in Image Quality Assessment: A Perceptual Preference Structure Alignment Perspective
Aobo Li (Xidian University), Weisheng Dong (Xidian University)
Domain AdaptationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a knowledge transfer framework called PreSTA based on perceptual preference structure alignment, aiming to enhance the generalization performance of image quality assessment models across different scenarios.
Rethinking MLLM Itself as a Segmenter with a Single Segmentation Token
Anqi Zhang (Beijing Institute Of Technology), Yunchao Wei (University Of Birmingham)
SegmentationTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: Propose a novel MLLM-based segmentation method, SELF1E, which can achieve pixel-level segmentation without using a specialized mask decoder, relying solely on a single [SEG] token.
Rethinking Model Selection in VLM Through the Lens of Gromov-Wasserstein Distance
Muyang Li (Sydney AI Centre, University of Sydney), Tongliang Liu (Sydney AI Centre, University of Sydney)
Representation LearningVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper systematically evaluates the applicability of 18 mainstream visual encoders in vision-language models (VLMs) and proposes a training-free model selection method based on the Gromov-Wasserstein (GW) distance.
Rethinking Occlusion Modeling for UAV Tracking
Jian Zhang (Sichuan University), Yi Lin (Sichuan University)
Object TrackingTransformerVideoBenchmark
🎯 What it does: Proposes OCTrack, a single-stream Transformer UAV tracker that integrates Clustered Occlusion Modeling and Cost-Aware Depth Bias.
Rethinking Pose Refinement in 3D Gaussian Splatting under Pose Prior and Geometric Uncertainty
Mangyu Kong (Yonsei University), Euntai Kim (Yonsei University)
Pose EstimationGaussian SplattingImage
🎯 What it does: Propose an untrained camera pose refinement method within the 3D Gaussian Splatting (3DGS) framework, taking into account both the uncertainty of pose priors and geometric uncertainty;
Rethinking Position Embedding as a Context Controller for Multi-Reference and Multi-Shot Video Generation
Binyuan Huang (Wuhan University), Daiqing Yang (Wuhan University)
GenerationTransformerVision-Language-Action ModelDiffusion modelVideo
🎯 What it does: Proposes PoCo, which utilizes position embeddings as a context controller to address the reference confusion problem in multi-reference, multi-shot video generation, achieving consistent generation of characters and backgrounds.
Rethinking Prompt Design for Inference-time Scaling in Text-to-Visual Generation
Subin Kim (KAIST), Tobias Hinz (Adobe)
GenerationLarge Language ModelPrompt EngineeringVision Language ModelImageVideoBenchmark
🎯 What it does: Propose the PRIS framework, which dynamically rewrites prompts during inference based on failure modes in visual generation to achieve joint scaling of prompts and visuals;
Rethinking SNN Online Training and Deployment: Gradient-Coherent Learning via Hybrid-Driven LIF Model
Zecheng Hao (Peking University), Tiejun Huang (Peking University)
Computational EfficiencySpiking Neural NetworkImageTime Series
🎯 What it does: Proposed the Hybrid-Driven Leaky-Integrate-and-Fire (HD-LIF) model family, achieving gradient consistency in online learning and improving training and inference efficiency through various variants.
Rethinking Token Reduction for Large Vision-Language Models
Yi Wang (Zhejiang University), Xinchao Wang (National University Of Singapore)
Computational EfficiencyTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a learning-based, prompt-agnostic visual token compression method called MetaCompress, designed for efficient inference in large-scale vision-language models during multi-round question-answering tasks.
Rethinking Two-Stage Referring-by-Tracking in Referring Multi-Object Tracking: Make it Strong Again
Weize Li (Beijing University of Posts and Telecommunications), Fei Su (Beijing University of Posts and Telecommunications)
Object TrackingTransformerOptical FlowVideoText
🎯 What it does: Propose a novel two-stage referential tracking framework called FlexHook for tracking multiple objects in videos based on natural language instructions.
Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training
Peng Sun (Westlake University), Tao Lin (Westlake University)
GenerationRepresentation LearningTransformerSupervised Fine-TuningPrompt EngineeringFlow-based ModelAuto EncoderImageMultimodality
🎯 What it does: Train the visual generation module of a unified multimodal model, proposing a two-stage image-only pre-training and hybrid fine-tuning framework called IOMM.
Rethinking Visual Rearrangement from A Diffusion Perspective
Tianliang Qi (Chinese Academy of Sciences), Shuqiang Jiang (University of Chinese Academy of Sciences)
Robotic IntelligenceTransformerDiffusion modelPoint Cloud
🎯 What it does: Proposes a visual rearrangement method based on the diffusion bridge model, utilizing a Gaussian Mixture Model (GMM) for distribution representation of point clouds and employing a Transformer for progressive denoising to predict the rearrangement target.
RetimeGS: Continuous-Time Reconstruction of 4D Gaussian Splatting
Xuezhen Wang (Hong Kong University of Science and Technology), Pedro V. Sander (Hong Kong University of Science and Technology)
GenerationGaussian SplattingOptical FlowVideo
🎯 What it does: Propose the RetimeGS 4D Gaussian Splatting approach to achieve continuous-time dynamic scene reconstruction and frame interpolation.
RetouchIQ: MLLM Agents for Instruction-Based Image Retouching with Generalist Reward
Qiucheng Wu (Adobe Research), Handong Zhao (Adobe Research)
Image TranslationTransformerLarge Language ModelReinforcement LearningAgentic AIVision Language ModelMultimodalityBenchmark
🎯 What it does: Developed RETOUCHIQ, a multi-modal large language model (MLLM)-based agent that can perform interpretable parameterized image retouching in Adobe Lightroom according to user instructions.
Retrieve and Segment: Are a Few Examples Enough to Bridge the Supervision Gap in Open-Vocabulary Segmentation?
Tilemachos Aravanis (Czech Technical University in Prague), Giorgos Tolias (Czech Technical University in Prague)
SegmentationRetrievalSupervised Fine-TuningVision Language ModelContrastive LearningImageTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose a retrieval-enhanced test-time adapter RNS, which utilizes a few visual support images with pixel annotations and text descriptions to train a lightweight linear classifier on each test image, achieving open-vocabulary segmentation.
Retrieve-to-Restore: Efficient All-in-One Image Restoration with a Retrieval-Based Degradation Bank
Chenxu Wang (Nanjing University), Jian Yang (Nanjing University)
RestorationRetrievalConvolutional Neural NetworkImage
🎯 What it does: Propose an All-in-One image restoration framework R2R, which externalizes degradation information by leveraging a retrieval-based degradation knowledge base, enabling a single model to simultaneously handle multiple degradations.
Retrieving Counterfactuals Improves Visual In-Context Learning
Guangzhi Xiong (University of Virginia), Aidong Zhang (University of Virginia)
RetrievalExplainability and InterpretabilityVision Language ModelContrastive LearningMultimodalityRetrieval-Augmented Generation
🎯 What it does: Propose the CIRCLES framework, which leverages attribute-guided synthetic image retrieval (CIR) to proactively construct counterfactual examples, combined with traditional similarity retrieval to enhance visual context learning.
RevINN: An End-to-End Invertible Neural Network for Reversible Adversarial Examples Generation
Jielun Huang (University of Macau), Guoheng Huang (Guangdong University of Technology)
Adversarial AttackConvolutional Neural NetworkFlow-based ModelGenerative Adversarial NetworkImage
🎯 What it does: Proposed an end-to-end reversible neural network called RevINN, which can directly generate reversible adversarial examples (RAE) from the frequency information of images and restore the original image during the reverse process.
Revisiting 2D Foundation Models for Scalable 3D Medical Image Classification
Han Liu (Siemens Healthineers), Sasa Grbic (Siemens Healthineers)
ClassificationTransformerSupervised Fine-TuningMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Proposed a scalable 3D medical image classification framework called AnyMC3D, which adapts 2D foundation models into 3D tasks through lightweight plugins.
Revisiting 3D Reconstruction Kernels as Low-Pass Filters
Shengjun Zhang (Tsinghua University), Yueqi Duan (Wuhan University)
Neural Radiance FieldGaussian SplattingImageBenchmark
🎯 What it does: The study treats 3D reconstruction as a signal reconstruction problem, re-analyzes the low-pass characteristics of the reconstruction kernel in the frequency domain, proposes an ideal low-pass Jinc kernel and a modulation kernel that balances spatial attenuation, and implements corresponding rendering methods.
Revisiting F-measure Optimization in Multi-Label Classification: A Sampling-based Approach
Zixun Wang (Chinese University of Hong Kong)
ClassificationImageText
🎯 What it does: Propose a sampling-based multi-label classification F-measure optimization framework, and derive traditional matrix multiplication as a convolution implementation, reducing the prediction complexity from O(q³) to O(q² log q).
Revisiting Geometric Obfuscation with Dual Convergent Lines for Privacy-Preserving Image Queries in Visual Localization
Jeonggon Kim (Hanyang University), Je Hyeong Hong (Hanyang University)
Pose EstimationSafty and PrivacySimultaneous Localization and MappingImageBenchmark
🎯 What it does: This study proposes the Dual Convergent Lines (DCL) keypoint hiding method to achieve privacy protection in cloud-based visual localization systems while maintaining localization performance.
Revisiting Learning with Noisy Labels: Active Forgetting and Noise Suppression
Mengmeng Sheng (Nanjing University of Science and Technology), Fumin Shen (University of Electronic Science and Technology of China)
ClassificationData-Centric LearningImage
🎯 What it does: This paper proposes a framework called FINE for learning tasks with noisy labels.
Revisiting Model Stitching In the Foundation Model Era
Zheda Mai (Ohio State University), Cheng-Hao Kuo (Amazon)
ClassificationSegmentationComputational EfficiencyRepresentation LearningTransformerImage
🎯 What it does: This paper investigates how to achieve compatibility and fusion among different objectives, data, and modalities of visual foundation models (VFM) in the era of large-scale VFM through model stitching, proposing a systematic stitching strategy and verifying its effectiveness;
Revisiting Monocular SLAM with Spatio-Temporal Scene Modeling
Valter Piedade (Mitsubishi Electric Research Laboratories), Pedro Miraldo (Mitsubishi Electric Research Laboratories)
Pose EstimationDepth EstimationRetrievalSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: Developed a real-time monocular visual SLAM pipeline called SLAM-MER based on spatiotemporal scene modeling, enabling efficient localization and map construction without requiring calibrated camera intrinsics.
Revisiting Multimodal KV Cache Compression: A Frequency-Domain-Guided Outlier-KV-Aware Approach
Yaoxin Yang (Fudan University), Tao Chen (Fudan University)
Computational EfficiencyTransformerMultimodality
🎯 What it does: To address the inference overhead caused by KV cache expansion with increasing visual input length in multimodal large language models (MLLMs), this paper proposes FlashCache, a frequency-domain guided and outlier KV-aware KV cache compression framework. The framework performs KV compression during the prefilling stage, is compatible with efficient attention kernels such as FlashAttention, and does not require recalculating attention scores or additional training.
Revisiting Optimal Coding for I-ToF under Practical Sensor Constraints
Wenbin Luo (Kyushu University), Hiroshi Kawasaki (Kyushu University)
Depth EstimationOptimizationImage
🎯 What it does: This paper addresses indirect time-of-flight (I-ToF) cameras, deriving a new depth variance metric under practical hardware constraints and noise models, and searching for optimal encoding schemes under this metric. The proposed high and low SNR encodings outperform traditional dual-slope and Hamiltonian encodings in both simulations and real measurements.
Revisiting Pose Sensitivity in Splat-based Computed Tomography under Sparse-view Reconstruction
Kiseok Choi (KAIST), Min H. Kim (KAIST)
OptimizationGaussian SplattingBiomedical DataComputed Tomography
🎯 What it does: This paper analyzes the pose sensitivity of splat-based CT and proposes a self-calibrating Gaussian splatting reconstruction framework that simultaneously optimizes volumetric and geometric parameters under sparse-view conditions, significantly eliminating stripe and needle artifacts.
Revisiting Sparsity Constraint Under High-Rank Property in Partial Multi-Label Learning
Chongjie Si (Shanghai Jiao Tong University), Wei Shen (Southeast University)
ClassificationOptimization
🎯 What it does: Propose a new partial multi-label learning method called Schirn, which imposes sparse constraints on the noisy label matrix while maintaining a high-rank structure on the predicted label matrix;
Revisiting the Necessity of Full Accuracy: Weakly Supervised Object-Level Offset Correction for Misaligned Building Labels
Junda Xu (Beijing Normal University), Libao Zhang (Beijing Normal University)
SegmentationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: Proposed an object-level multi-stage alignment framework (OMAF) that corrects spatial offsets between Google Earth images and building labels through weakly supervised methods, generating high-quality aligned labels for semantic segmentation training.
Revisiting the Necessity of Lengthy Chain-of-Thought in Vision-centric Reasoning Generalization
Yifan Du (Gaoling School of Artificial Intelligence, Renmin University of China), Ji-Rong Wen (Gaoling School of Artificial Intelligence, Renmin University of China)
TransformerSupervised Fine-TuningReinforcement LearningVision Language ModelImageMultimodalityChain-of-Thought
🎯 What it does: Investigated the impact of different chain-of-thought (CoT) formats on the generalization ability of vision-language models in visual reasoning tasks.
Revisiting Token Compression for Accelerating ViT-based Sparse Multi-View 3D Object Detectors
Mingqian Ji (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
Object DetectionAutonomous DrivingComputational EfficiencyTransformerImagePoint Cloud
🎯 What it does: This paper proposes the SEPatch3D framework, which accelerates sparse multi-view 3D object detection models based on Vision Transformer through dynamic patch size selection, information patch screening, and cross-granularity feature enhancement.
Revisiting Unknowns: Towards Effective and Efficient Open-Set Active Learning
Chen-Chen Zong (Nanjing University of Aeronautics and Astronautics), Sheng-Jun Huang (Nanjing University of Aeronautics and Astronautics)
ClassificationAnomaly DetectionComputational EfficiencyContrastive LearningImage
🎯 What it does: Propose a unified, detector-free open-set active learning framework named E2OAL, which leverages unlabeled unknown samples through adaptive clustering to discover potential category structures, and uses them as an auxiliary classification head to enhance known class learning and confidence calibration. Subsequently, a two-stage querying strategy is designed using logit-margin purity scores and JS information measures to automatically control query accuracy.
Revisiting Visual Corruptions in LVLMs: A Shape-Texture Perspective on Model Failures
Xinkuan Qiu (Chinese Academy of Sciences), Shiguang Shan (Chinese Academy of Sciences)
Anomaly DetectionComputational EfficiencyVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Propose a training-free, shape-and-texture dual-channel contrastive decoding framework named ST-CD, which constructs two contrastive views through edge extraction and jigsaw permutation, and significantly enhances the robustness of large vision-language models against various visual distortions (noise, blur, geometric distortion, color shift, etc.) via entropy-weighted adaptive fusion.
REVISOR: Beyond Textual Reflection, Towards Multimodal Introspective Reasoning in Long-Form Video Understanding
Jiaze Li (MiLM Plus, Xiaomi Inc.), Jian Luan (Renmin University of China)
Large Language ModelReinforcement LearningVision Language ModelVideoMultimodality
🎯 What it does: Propose the REVISOR framework, combining tool-enhanced multimodal self-reflection mechanisms to improve long-term video understanding.
REVIVE 3D: Refinement via Encoded Voluminous Inflated prior for Volume Enhancement
Hankyeol Lee (Yonsei University), Jongyoo Kim (Yonsei University)
GenerationTransformerDiffusion modelImageMesh
🎯 What it does: Proposed a two-stage 3D generation framework called REVIVE 3D, which first constructs a 3D prior with specific volume and local details through a 'dilated prior', and then refines this prior in the 3D latent space using a diffusion model to generate volume-rich and detail-abundant 3D meshes from 2D images.
Reviving ConvNeXt for Efficient Convolutional Diffusion Models
Taesung Kwon (KAIST), Vinicius Azevedo (ETH Zürich)
GenerationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Propose a fully convolutional diffusion model (FCDM), adapting the ConvNeXt architecture into a generator applicable for conditional diffusion generation.
Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillation
Yunhong Lu (Zhejiang University), Min Zhang (Zhejiang University)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelVideoText
🎯 What it does: Propose the Reward Forcing framework, which maintains global context via EMA-Sink and enhances motion dynamics through Rewarded Distribution Matching Distillation (Re-DMD), achieving high-quality, real-time streaming video generation.
Reward Sharpness-Aware Fine-Tuning for Diffusion Models
Kwanyoung Kim (GIST), Byeongsu Sim (Samsung Research)
GenerationReinforcement LearningDiffusion modelImage
🎯 What it does: This paper proposes a fine-tuning method called RSA-FT based on reward surface smoothing, aiming to address the reward hacking problem in reward-driven diffusion model training.
RewardFlow: Generate Images by Optimizing What You Reward
Onkar Susladkar (University of Illinois Urbana-Champaign), Ismini Lourentzou (University of Illinois Urbana-Champaign)
GenerationPrompt EngineeringDiffusion modelFlow-based ModelImageTextMultimodalityBenchmarkStochastic Differential Equation
🎯 What it does: Designed and implemented RewardFlow, a training-free, inversion-free multi-reward Langevin dynamics framework for text-guided image editing and generation.
ReWeaver: Towards Simulation-Ready and Topology-Accurate Garment Reconstruction
Ming Li (Zhejiang University), Xiangru Huang (Westlake University)
GenerationData SynthesisConvolutional Neural NetworkTransformerImageMesh
🎯 What it does: Simultaneously reconstruct 3D garment geometry and 2D seam patterns from four sparse perspective images, maintaining a strict correspondence between the two.
Rewis3d: Reconstruction Improves Weakly-Supervised Semantic Segmentation
Jonas Ernst (Saarland University), Bernt Schiele (Max Planck Institute for Informatics)
SegmentationAutonomous DrivingTransformerVideoPoint Cloud
🎯 What it does: Leveraging the 3D geometric structure recovered from 2D videos as an auxiliary supervisory signal significantly enhances semantic segmentation performance under sparse annotations (points, line drawings, rough annotations).
RF4D:Neural Radar Fields for Novel View Synthesis in Outdoor Dynamic Scenes
Jiarui Zhang (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
GenerationData SynthesisNeural Radiance FieldOptical FlowPoint Cloud
🎯 What it does: Constructed a time-space neural field based on millimeter-wave radar (RF4D), achieving view synthesis and occupancy estimation for outdoor dynamic scenes.
RFDM: Residual Flow Diffusion Models for Video Editing
Mohammadreza Salehi (Samsung AI Center), Abhinav Mehrotra (Samsung AI Center)
GenerationDiffusion modelFlow-based ModelVideoText
🎯 What it does: This paper proposes an autoregressive video editing framework based on residual flow diffusion models, which can recursively edit videos frame by frame with text guidance while maintaining temporal consistency.
RGB-Event based Pedestrian Attribute Recognition: A Benchmark Dataset and An Asymmetric RWKV Fusion Framework
Xiao Wang (Anhui University), Chenglong Li (Anhui University)
RecognitionTransformerMultimodalityBenchmark
🎯 What it does: Constructed the first multi-modal RGB-Event pedestrian attribute recognition dataset, EventPAR, and proposed an RWKV-based heterogeneous RGB-Event fusion framework for joint recognition of pedestrian appearance attributes and emotional attributes.
RHCNet: Residual-Guided Hierarchical Calibration Network for Robust Underwater Object Detection
Yueying Wang (Shanghai University), Xin Xu (East China University of Science and Technology)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Proposed a residual-guided hierarchical calibration network (RHCNet), achieving structural restoration, semantic focusing, and cross-scale alignment for blurred, low-contrast targets in seawater environments through the residual-guided feature enhancement module (RGFE) and hierarchical feature calibration pyramid (HFCP).
RHINO: Reconstructing Human Interactions with Novel Objects from Monocular Videos
Lixin Xue (ETH Zürich), Dimitrios Tzionas (University of Amsterdam)
GenerationPose EstimationDepth EstimationOptimizationNeural Radiance FieldVideoBenchmark
🎯 What it does: Propose the RHINO framework, which can recover 4D scenes (including humans, unseen objects, and static environments) of dynamic human-object interactions from monocular videos.
RHO: Robust Holistic OSM-Based Metric Cross-View Geo-Localization
Junwei Zheng (Karlsruhe Institute of Technology), Rainer Stiefelhagen (Karlsruhe Institute of Technology)
Pose EstimationRetrievalAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: Proposed a robust metric cross-view geolocation framework called RHO for OSM maps, and created a large-scale robust dataset named CV-RHO.
RI-Mamba: Rotation-Invariant Mamba for Robust Text-to-Shape Retrieval
Khanh Nguyen (University of Western Australia), Ajmal Mian (University of Western Australia)
RetrievalVision Language ModelContrastive LearningTextMultimodalityPoint Cloud
🎯 What it does: Designed a rotation-invariant Mamba model for text-to-shape retrieval, performing unsupervised cross-modal contrastive learning on large-scale data.
RigMo: Unifying Rig and Motion Learning for Generative Animation
Hao Zhang (Snap Inc.), Bing Zhou (University Of Illinois Urbana Champaign)
GenerationTransformerDiffusion modelAuto EncoderMesh
🎯 What it does: Propose a unified generation framework named RigMo that can simultaneously learn skeletal structures and motion parameters from raw mesh sequences, achieving end-to-end self-supervised learning without manual annotations.
RINO: Rotation-Invariant Non-Rigid Correspondences
Maolin Gao, Daniel Cremers
Diffusion modelMesh
🎯 What it does: This paper proposes an unsupervised rotation-invariant RINO framework that can directly learn dense correspondences from raw geometry.
RISE: Single Static Radar-based Indoor Scene Understanding
Kaichen Zhou (Massachusetts Institute of Technology), Fadel Adib (Massachusetts Institute of Technology)
Object DetectionGenerationData SynthesisDiffusion modelPoint CloudBenchmark
🎯 What it does: Developed RISE, a single static millimeter-wave radar indoor scene understanding system that jointly achieves layout reconstruction and object detection.
RiskProp: Collision-Anchored Self-Supervised Risk Propagation For Early Accident Anticipation
Yiyang Zou (Wuhan University), Yu Wu (Wuhan University)
Anomaly DetectionAutonomous DrivingConvolutional Neural NetworkVideo
🎯 What it does: Propose a self-supervised risk propagation framework called RiskProp that uses only collision frames as supervision for early prediction of collisions.
RL-ScanIQA: Reinforcement-Learned Scanpaths for Blind 360deg Image Quality Assessment
Yujia Wang (Victoria University of Wellington), Neil.A Dodgson (Victoria University of Wellington)
Recurrent Neural NetworkTransformerReinforcement LearningImage
🎯 What it does: Propose the RL-ScanIQA framework to achieve end-to-end joint optimization of 360° no-reference image quality assessment and viewpoint scanning path generation.
RLFTSim: Realistic and Controllable Multi-Agent Traffic Simulation via Reinforcement Learning Fine-Tuning
Ehsan Ahmadi (University of Alberta), Kasra Rezaee (Huawei Technologies Canada)
Autonomous DrivingReinforcement Learning
🎯 What it does: This study proposes the RLFTSim framework, which utilizes reinforcement learning to fine-tune a pre-trained multi-agent traffic simulator, achieving controllability while maintaining high realism;
RMAE-ProGRess: Advancing Semantic Segmentation in Unstructured Environments
Manish Bhurtel (Howard University), Danda B. Rawat (Howard University)
SegmentationTransformerAuto EncoderImage
🎯 What it does: A lightweight RMAE-ProGRess model is proposed for semantic segmentation in unstructured environments.
RMIR: A Benchmark Dataset for Reasoning-Intensive Multimodal Image Retrieval
Yijiang Li (University of California San Diego), Sunny Dasgupta (Amazon)
RetrievalTransformerLarge Language ModelVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposed the RMIR (Reasoning-Intensive Multimodal Image Retrieval) benchmark dataset to evaluate the performance of multimodal retrieval models on three categories of reasoning tasks: functional, temporal, and causal.
RNED: Rotary Number Encoding and Decoding for Medical VLMs
Fengbei Liu (Cornell University), Mert R. Sabuncu (Cornell University)
Representation LearningSupervised Fine-TuningVision Language ModelScore-based ModelMultimodalityBiomedical DataComputed Tomography
🎯 What it does: This paper proposes a new numerical representation method called Rotational Numerical Encoding and Decoding (RNED) for achieving precise and continuous numerical prediction in medical vision-language models.
RnG: A Unified Transformer for Complete 3D Modeling from Partial Observations
Mochu Xiang (Northwestern Polytechnical University), Yuchao Dai (Northwestern Polytechnical University)
GenerationDepth EstimationTransformerNeural Radiance FieldImagePoint Cloud
🎯 What it does: Designed a unified Transformer structure RnG, which can reconstruct complete 3D structures and generate high-quality RGBD outputs from arbitrary perspectives under the condition of only a few uncalibrated images.
RNN as Linear Transformer: A Closer Investigation into Representational Potentials of Visual Mamba Models
Timing Yang (Johns Hopkins University), Guoyizhe Wei (Johns Hopkins University)
ClassificationObject DetectionSegmentationRepresentation LearningImage
🎯 What it does: Systematically study the representational capability of Mamba in visual tasks and propose the Binary-AUC evaluation metric.
RoadGIE: Towards A Global-Scale Aerial Benchmark for Generalizable Interactive Road Extraction
Chenxu Peng (Nankai University), Xiang Li (NKIARI)
SegmentationConvolutional Neural NetworkTransformerPrompt EngineeringImageBenchmark
🎯 What it does: Proposed a global-scale road segmentation dataset named WorldRoadSeg-360K (366,947 satellite images) and developed an interactive road extraction framework called RoadGIE based on connectivity-aware prompts.
RoadSceneBench: A Lightweight Benchmark for Mid-Level Road Scene Understanding
Xiyan Liu (Baidu Inc), Zhen Lu (Baidu Inc)
Autonomous DrivingTransformerReinforcement LearningVision Language ModelImageVideoBenchmark
🎯 What it does: Proposed RoadSceneBench lightweight benchmark and HRRP-T reinforcement learning framework for evaluating and enhancing visual-language models in mid-level road semantic reasoning.
Robo-SGG: Exploiting Layout-Oriented Normalization and Restitution Can Improve Robust Scene Graph Generation
Changsheng Lv (Beijing University of Posts and Telecommunications), Mengshi Qi (Beijing University of Posts and Telecommunications)
RecognitionConvolutional Neural NetworkImage
🎯 What it does: Proposed a pluggable module Robo-SGG that enhances the robustness of scene graph generation using layout information;
RoboAgent: Chaining Basic Capabilities for Embodied Task Planning
Peiran Xu (Peking University), Yadong Mu (Peking University)
Robotic IntelligenceTransformerSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Proposes RoboAgent, a capability-driven planning framework based on visual language models (VLM), which decomposes complex embodied task planning into multiple basic capabilities directly processable by VLM, achieving closed-loop control from perception to action instructions through a single model that jointly executes the scheduler and five specialized capabilities.
RoboTAG: End-to-end Robot Pose Estimation via Topological Alignment Graph
Yifan Liu (Tsinghua University), Hanspeter Pfister (Harvard University)
Pose EstimationRobotic IntelligenceConvolutional Neural NetworkGraph Neural NetworkImage
🎯 What it does: This paper proposes an end-to-end robotic configuration estimation framework called RoboTAG, which constructs a topology graph containing 2D and 3D branches to achieve mutual alignment and co-learning between camera and robot states, significantly reducing dependence on labeled data;
RobotSeg: A Model and Dataset for Segmenting Robots in Image and Video
Haiyang Mei (National University of Singapore), Mike Zheng Shou (National University of Singapore)
SegmentationRobotic IntelligenceTransformerPrompt EngineeringContrastive LearningImageVideo
🎯 What it does: Propose RobotSeg, a robot segmentation model based on SAM 2 and the VRS video robot segmentation dataset, supporting image and video segmentation, and enabling zero-shot, single-point, box selection, and interactive refinement;
RoboWheel: A Data Engine from Real-World Human Demonstrations for Cross-Embodiment Robotic Learning
Yuhong Zhang (Tsinghua University), Haoqian Wang (Tsinghua University)
Data SynthesisDepth EstimationOptimizationData-Centric LearningRobotic IntelligenceReinforcement LearningVideoMultimodalityPoint Cloud
🎯 What it does: Studies how to convert real-world hand-object interaction videos into data usable for cross-modal machine learning, forming a complete data engine through high-precision reconstruction, physical feasibility optimization, cross-modal mapping, and simulation augmentation.
Robust Promptable Video Object Segmentation
Sohyun Lee (POSTECH), Suha Kwak (POSTECH)
SegmentationTransformerPrompt EngineeringVideoBenchmark
🎯 What it does: The study addresses the robustness of Promptable Video Object Segmentation (PVOS) under real-world harsh conditions such as noise, rain, snow, fog, and nighttime, proposing a new benchmark and an improved model;
Robust Remote Sensing Image-Text Retrieval with Noisy Correspondence
Qiya Song (Hunan Normal University), Xudong Kang (Hunan University)
RetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Investigate the noise correspondence problem in remote sensing image-text retrieval and propose a robust remote sensing image-text retrieval (RRSITR) framework.
Robust Spiking Neural Networks by Temporal Mutual Information
Mengting Xu (Zhejiang University), Gang Pan (Zhejiang University)
ClassificationAdversarial AttackSpiking Neural NetworkImageTime Series
🎯 What it does: This paper proposes a regularization method based on Temporal Mutual Information (TMI) to enhance the robustness of Spiking Neural Networks (SNN).
Robust3DGSW: Toward Robust Watermarking for Quantization-Aware 3D Gaussian Splatting
Boyu Wang (East China Normal University), Mingsong Chen (East China Normal University)
Gaussian SplattingImageBenchmark
🎯 What it does: Proposed a quantization-aware 3D Gaussian Splatting watermark embedding method aimed at simultaneously enhancing watermark robustness and rendering quality under low-bit quantization environments.
Robustness Under Data Scarcity: Few-Shot Continual Adversarial Training for Evolving Threats
Wenxuan Wang (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
ClassificationAdversarial AttackData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: Proposes a continuous adversarial training framework FS-CAT under few adversarial samples, and designs three key mechanisms: adversarial boundary loss, GMM prototype replay, and multi-domain balanced loss.
RobustVisRAG: Causality-Aware Vision-Based Retrieval-Augmented Generation under Visual Degradations
I-Hsiang Chen (National Taiwan University), Wei-Ting Chen (Microsoft)
GenerationRetrievalVision Language ModelContrastive LearningImageTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposes the RobustVisRAG framework to improve the retrieval and generation performance of Vision-based Retrieval-Augmented Generation (VisRAG) under visual degradation conditions, and constructs the Distortion-VisRAG benchmark dataset.
Role-SynthCLIP: A Role-Play Driven Diverse Synthetic Data Approach
Yuanxiang Huangfu (PatSnap Company Limited), Weilei Wang (PatSnap Company Limited)
ClassificationData SynthesisRetrievalKnowledge DistillationLarge Language ModelPrompt EngineeringContrastive LearningImageTextMultimodality
🎯 What it does: Propose the Role-SynthCLIP framework, which enables multimodal large language models (MLLM) to generate multiple perspectives and fine-grained captions for the same image through multi-role prompting, thereby constructing high-quality, semantically rich synthetic image-text pairs.
RoMo: A Large-Scale, Richly Organized Dataset and Semantic Taxonomy for Human Motion Generation
Jiahao Zhang (Australian National University), Yizhak Ben-Shabat (Roblox)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelScore-based ModelVideoTextSequentialBenchmark
🎯 What it does: Propose the RoMo dataset, which collects 820K 3D human motion clips with five text descriptions, structured using a hierarchical semantic taxonomy, and releases the Motion Toolbox as a unified evaluation and visualization tool.
Roots Beneath the Cut: Uncovering the Risk of Concept Revival in Pruning-Based Unlearning for Diffusion Models
Ci Zhang (University of Georgia), Geng Yuan (Stevens Institute of Technology)
Safty and PrivacyAdversarial AttackDiffusion modelImage
🎯 What it does: Studied the safety of pruning in diffusion models and proposed a data- and training-agnostic attack framework. The framework uses low-rank matrix completion, Top-K sign preservation, and neuron maximum scaling to recover pruned weights and restore forgotten concepts; meanwhile, Gaussian blur defense is proposed.
RoSAMDepth: Robust Self-supervised Depth Estimation Leveraging Segment Anything Model
Xuanang Gao (Shanghai Jiao Tong University), Wei Liu (Shanghai Jiao Tong University)
Depth EstimationAutonomous DrivingTransformerContrastive LearningImage
🎯 What it does: This paper proposes RoSAMDepth, a framework that leverages object-level priors from the Segment Anything Model (SAM) to enhance the robustness of self-supervised monocular depth estimation. It injects SAM's segmentation information at the feature level and improves depth smoothing constraints and pseudo-label weighting methods, resulting in sharper and more accurate depth maps under various adverse conditions.
ROSE: Rotate Your Large Language Model to See
Tongtian Yue (Chinese Academy of Sciences), Jing Liu (Chinese Academy of Sciences)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes ROSE, an efficient multimodal language model, by applying an orthogonal rotation matrix in the parameter space of a pre-trained large language model to directly inject visual information into the model.
Rosetta Stone For Unified MLLMs: A Unified Tokenizer to Decipher Understanding and Generation
Wenyu Sun (Taobao & Tmall Group of Alibaba), Yuning Jiang (Taobao & Tmall Group of Alibaba)
Representation LearningConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelAuto EncoderContrastive LearningImageTextMultimodality
🎯 What it does: Proposed a unified visual tokenizer using a hierarchical decoupling and attention-priority mapping training framework to achieve a unified representation for image reconstruction and semantic alignment, and built a 7B large-scale multimodal LLM based on this tokenizer.
Rotation Invariant and Symmetry Aware Pixel Difference Network for Remote Sensing Object Detection
Jialei Zhan (National University of Defense Technology), Ming-Ming Cheng (Nankai University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Proposed a remote sensing object detection network named RIS-PiDiNet that integrates rotation invariance and symmetry modeling.
Rounded or Streamlined Head? Bridging Concept Bottleneck Models and Attribute-Described Object Parts
Yang Liu (DAMO Academy, Alibaba Group), Ling Zhang (DAMO Academy, Alibaba Group)
ClassificationObject DetectionSegmentationExplainability and InterpretabilityTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose an object-aware concept bottleneck model (OA-CBM) that simultaneously incorporates semantic consistency and object consistency into a vision-language interpretability framework;
Routing on Demand: DSNet for Efficient Progressive Point Cloud Denoising
Xiaoqian Cheng (University of Science and Technology of China), Renjie Chen (University of Science and Technology of China)
RestorationComputational EfficiencyGraph Neural NetworkPoint Cloud
🎯 What it does: This paper proposes DSNet, a progressive denoising network for point clouds that achieves adaptive processing of local noise through a dynamic routing mechanism.
RPGFusion: 4D Radar Prior-Guided Multi-Modal Fusion for 3D Detection
Xin Qiu (Zhejiang University), Wenjie Liu (Zhejiang University)
Object DetectionAutonomous DrivingMultimodalityPoint Cloud
🎯 What it does: Proposes an RPGFusion framework that leverages 4D radar priors for multi-modal fusion to enhance 3D object detection in autonomous driving.
rPPG-VQA: A Video Quality Assessment Framework for Unsupervised rPPG Training
Tianyang Dai (University of Science and Technology of China), Yang Hu (University of Science and Technology of China)
OptimizationData-Centric LearningTransformerLarge Language ModelVideo
🎯 What it does: Propose a video quality assessment framework specifically designed for unsupervised remote photoplethysmography (rPPG) training, named rPPG-VQA, which combines signal-level consensus SNR with scene disturbance evaluation based on a multi-modal large language model, and implements two-stage adaptive sampling based on quality scores.
RS-SSM: Refining Forgotten Specifics in State Space Model for Video Semantic Segmentation
Kai Zhu (Peking University), Jiahuan Zhou (University of Chinese Academy of Sciences)
SegmentationTransformerVideo
🎯 What it does: Designed and implemented a video semantic segmentation framework RS-SSM based on the state space model (SSM), achieving more accurate pixel-level segmentation by compensating for the spatiotemporal details lost during SSM compression.
RT-Splatting: Joint Reflection-Transmission Modeling with Gaussian Splatting
Ji Shi (Peking University), Wenzhen Yue (Peking University)
GenerationGaussian SplattingImage
🎯 What it does: Propose the RT-Splatting framework, jointly modeling reflection and transmission of semi-transparent surfaces to achieve high-fidelity rendering.
RunawayEvil: Jailbreaking the Image-to-Video Generative Models
Songping Wang (Nanjing University), Caifeng Shan (Nanjing University)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelReinforcement LearningImageVideoMultimodality
🎯 What it does: Propose a self-evolving multimodal jailbreak framework RunawayEvil based on the Strategy-Tactic-Action (S-T-A) paradigm, which can collaborate with text and image inputs to bypass the security mechanisms of Image-to-Video (I2V) models.
RxnCaption: Reformulating Reaction Diagram Parsing as Visual Prompt Guided Captioning
Jiahe Song (Shanghai Jiao Tong University), Conghui He (Shanghai Artificial Intelligence Laboratory)
Object DetectionDrug DiscoveryConvolutional Neural NetworkLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Redefined the chemical reaction diagram parsing task as an image description task, proposed and implemented the RxnCaption framework;
S$^2$-MLLM: Boosting Spatial Reasoning Capability of MLLMs for 3D Visual Grounding with Structural Guidance
Beining Xu, Hesheng Wang (Shanghai Jiao Tong University)
Computational EfficiencyRepresentation LearningLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningTextMultimodalityPoint Cloud
🎯 What it does: Propose the S-MLLM framework, which equips multi-modal large language models with implicit 3D spatial reasoning capabilities by incorporating feed-forward 3D reconstruction as spatial guidance during training, enabling 3D visual localization tasks.