CVPR 2025 Papers — Page 22
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Resilient Sensor Fusion Under Adverse Sensor Failures via Multi-Modal Expert Fusion
Konyul Park (Seoul National University), Jun Won Choi (Seoul National University)
Object DetectionAutonomous DrivingTransformerMixture of ExpertsMultimodalityPoint Cloud
🎯 What it does: This paper proposes a LiDAR-Camera fusion 3D detection framework called MoME, which maintains robustness under sensor failures or adverse weather conditions.
ReSpec: Relevance and Specificity Grounded Online Filtering for Learning on Video-Text Data Streams
Chris Dongjoo Kim (Seoul National University), Christopher Clark (Allen Institute for AI)
RetrievalOptimizationComputational EfficiencyTransformerVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes an online filtering framework named ReSpec, which is designed to real-time select high-quality samples that meet task requirements from video-text data streams to improve online learning efficiency and performance.
RestorGS: Depth-aware Gaussian Splatting for Efficient 3D Scene Restoration
Yuanjian Qiao, Kai Xu
RestorationConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: This paper proposes RestorGS, a unified deep perception Gaussian projection framework for restoring high-quality rendering of various distorted 3D scenes such as underwater, nighttime, and foggy environments.
Retaining Knowledge and Enhancing Long-Text Representations in CLIP through Dual-Teacher Distillation
Yuheng Feng (Fudan University), Siyu Zhu (Fudan University)
RetrievalKnowledge DistillationTransformerContrastive LearningImageTextMultimodality
🎯 What it does: By retraining CLIP through a dual-teacher distillation framework (Long-CLIP as the long text teacher and the original CLIP as the base teacher), significant improvements are achieved in long text retrieval, short text retrieval, and zero-shot image classification tasks.
Rethinking Correspondence-based Category-Level Object Pose Estimation
Huan Ren (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
Object DetectionPose EstimationImagePoint Cloud
🎯 What it does: This paper presents SpotPose, a two-stage correspondence-based method that achieves more accurate category-level object pose estimation by extracting shape-sensitive and pose-invariant features during the correspondence prediction stage and predicting and discarding outlier correspondences during the pose fitting stage.
Rethinking Decoder Design: Improving Biomarker Segmentation Using Depth-to-Space Restoration and Residual Linear Attention
Saad Wazir (Korea Advanced Institute of Science and Technology), Daeyoung Kim (Korea Advanced Institute of Science and Technology)
SegmentationConvolutional Neural NetworkBiomedical Data
🎯 What it does: A new multi-scale convolutional attention deep-to-space (MCADS) decoder is proposed for biomarker medical image segmentation, achieving high-precision segmentation in conjunction with an improved U2-Net encoder.
Rethinking Diffusion for Text-Driven Human Motion Generation: Redundant Representations, Evaluation, and Masked Autoregression
Zichong Meng (Northeastern University), Huaizu Jiang (Northeastern University)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderVideoText
🎯 What it does: Redesign a diffusion-based text-driven 3D human motion generation model, eliminating redundant dimensions and employing masked autoregressive training;
Rethinking End-to-End 2D to 3D Scene Segmentation in Gaussian Splatting
Runsong Zhu (Chinese University of Hong Kong), Chi-Wing Fu (Chinese University of Hong Kong)
SegmentationContrastive LearningGaussian SplattingImage
🎯 What it does: A unified lifting framework called Unified-Lift is designed, which does not require pre-processing or post-processing, to elevate multi-view 2D instance segmentation to 3D Gaussian Splatting scenes, achieving high-quality and view-consistent 3D segmentation.
Rethinking Epistemic and Aleatoric Uncertainty for Active Open-Set Annotation: An Energy-Based Approach
Chen-Chen Zong (Nanjing University of Aeronautics and Astronautics), Sheng-Jun Huang (Nanjing University of Aeronautics and Astronautics)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: An energy-based active open set annotation framework (EAOA) is proposed, which simultaneously considers knowledge uncertainty and noise uncertainty in open set active learning to guide sample queries.
Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages
Matteo Farina (University of Trento), Elisa Ricci (Fondazione Bruno Kessler)
ClassificationDomain AdaptationTransformerSupervised Fine-TuningVision Language ModelImageMultimodality
🎯 What it does: This paper proposes a two-stage few-shot adaptation method (2SFS), which first learns general features using PEFT and then trains a linear classifier.
Rethinking Lanes and Points in Complex Scenarios for Monocular 3D Lane Detection
Yifan Chang (Institute of Automation, Chinese Academy of Sciences), Xingang Wang (Institute of Automation, Chinese Academy of Sciences)
Autonomous DrivingConvolutional Neural NetworkSupervised Fine-TuningPoint Cloud
🎯 What it does: A new endpoint completion strategy and EP-head for sparse point-based 3D lane detection are proposed, along with a PL-attention module based on geometric priors.
Rethinking Noisy Video-Text Retrieval via Relation-aware Alignment
Huakai Lai (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
RetrievalContrastive LearningVideoText
🎯 What it does: The study improves video-text retrieval in the presence of noise.
Rethinking Personalized Aesthetics Assessment: Employing Physique Aesthetics Assessment as An Exemplification
Haobin Zhong (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
Recommendation SystemGraph Neural NetworkLarge Language ModelSupervised Fine-TuningGenerative Adversarial NetworkImageMultimodality
🎯 What it does: This paper proposes a new personalized aesthetic assessment framework PAA+, which achieves precise predictions of user personalized aesthetics through three stages (pre-training, fine-tuning, and continuous learning), using physical aesthetics as an experimental case.
Rethinking Query-based Transformer for Continual Image Segmentation
Yuchen Zhu (ShanghaiTech University), Sibei Yang (Sun Yat-sen University)
SegmentationTransformerImage
🎯 What it does: The SimCIS framework is proposed to address the issues of catastrophic forgetting and background semantic drift in class-incremental image segmentation.
Rethinking Reconstruction and Denoising in the Dark: New Perspective, General Architecture and Beyond
Tengyu Ma (Dalian University of Technology), Risheng Liu (Dalian University of Technology)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A low-light RAW image reconstruction and denoising method called CANS/CANS++ based on a backbone-head structure is proposed.
Rethinking Spiking Self-Attention Mechanism: Implementing a-XNOR Similarity Calculation in Spiking Transformers
Yichen Xiao (University of Electronic Science and Technology of China), Malu Zhang (University of Electronic Science and Technology of China)
Spiking Neural NetworkTransformerImage
🎯 What it does: A pulse self-attention module based on α-XNOR similarity, α-SSA, is designed to improve the performance of SNN Transformers.
Rethinking Temporal Fusion with a Unified Gradient Descent View for 3D Semantic Occupancy Prediction
Dubing Chen (University of Macau), Jianbing Shen (University of Macau)
SegmentationAutonomous DrivingOptimizationComputational EfficiencyRecurrent Neural NetworkPoint Cloud
🎯 What it does: A unified gradient descent-based RNN perspective temporal fusion framework GDFusion is proposed to enhance vision-based 3D semantic occupancy prediction.
Rethinking the Adversarial Robustness of Multi-Exit Neural Networks in an Attack-Defense Game
Keyizhi Xu (Wuhan University), Chao Liang (Wuhan University)
Adversarial AttackConvolutional Neural NetworkTransformerReinforcement LearningImage
🎯 What it does: This study investigates the robustness of multi-output neural networks under adversarial attacks and proposes a game theory-based evaluation method AIMER and a defense method NEED.
Rethinking Token Reduction with Parameter-Efficient Fine-Tuning in ViT for Pixel-Level Tasks
Cheng Lei (University of Electronic Science and Technology of China), Le Zhang (University of Electronic Science and Technology of China)
RestorationSegmentationKnowledge DistillationTransformerSupervised Fine-TuningImage
🎯 What it does: This paper studies the feasibility of combining Parameter-Efficient Fine-Tuning (PEFT) with Token Reduction (TR) in pixel-level tasks and proposes the Diverse Attention Restorer (DAR) to reduce the number of tokens while maintaining attention diversity and high-frequency information.
Rethinking Training for De-biasing Text-to-Image Generation: Unlocking the Potential of Stable Diffusion
Eunji Kim (Seoul National University), Sungroh Yoon (Seoul National University)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: Without the need for additional training, by analyzing the noise space of Stable Diffusion (SD), it was found that the noise corresponding to a few attributes clusters into a 'minority zone', and a weak guidance method is proposed to direct random noise towards these zones, thereby reducing gender and racial bias.
Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector
Xiao Guo (Michigan State University), Xiaoming Liu (Michigan State University)
ClassificationRecognitionExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: A multi-modal facial forgery detector M2F2-Det is proposed, which can simultaneously provide authenticity probabilities and natural language explanations.
Retrieving Semantics from the Deep: an RAG Solution for Gesture Synthesis
M. Hamza Mughal (Max Planck Institute for Informatics), Christian Theobalt (Saarland University)
GenerationData SynthesisRetrievalDiffusion modelTextRetrieval-Augmented GenerationAudio
🎯 What it does: This paper proposes a retrieval-augmented generation (RAG) based diffusion model for generating natural and semantically rich co-speech gestures under speech and text conditions.
Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition
Hongda Liu (Institute of Automation, Chinese Academy of Sciences), Zhenan Sun (Institute of Automation, Chinese Academy of Sciences)
RecognitionGraph Neural NetworkContrastive LearningVideo
🎯 What it does: This paper proposes ProtoGCN, which achieves fine-grained skeleton action recognition through a prototype reconstruction network, allowing for better differentiation of similar actions.
Reversible Decoupling Network for Single Image Reflection Removal
Hao Zhao (Tianjin University), Xiaojie Guo (Tianjin University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A reversible decoupling network RDNet for single image reflection removal is proposed, which achieves precise separation of the transmission layer and reflection layer using a reversible encoder, columnar multi-scale structure, and adaptive transmission rate hint generator.
Reversing Flow for Image Restoration
Haina Qin (University of Chinese Academy of Sciences), Weiming Hu (University of Chinese Academy of Sciences)
RestorationFlow-based ModelImageOrdinary Differential Equation
🎯 What it does: Proposes ResFlow, an image restoration framework based on deterministic continuous normalizing flows, which can directly reverse the degradation path from high quality to low quality.
ReVisionLLM: Recursive Vision-Language Model for Temporal Grounding in Hour-Long Videos
Tanveer Hannan (Ludwig Maximilian University of Munich), Gedas Bertasius (University of Oxford)
RetrievalOptimizationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodality
🎯 What it does: A recursive visual language model, ReVisionLLM, is proposed for the temporal localization task of hour-long videos.
Revisiting Audio-Visual Segmentation with Vision-Centric Transformer
Shaofei Huang (Hefei University of Technology), Meng Wang (Hefei University of Technology)
SegmentationTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: Proposes the Vision-Centric Transformer (VCT) framework, which utilizes visually derived queries to iteratively obtain corresponding audio information and fine-grained visual features in a multi-layer audio-visual Transformer decoder, achieving more accurate sound source segmentation and boundary delineation.
Revisiting Backdoor Attacks against Large Vision-Language Models from Domain Shift
Siyuan Liang (Nanyang Technological University), Dacheng Tao (Nanyang Technological University)
Domain AdaptationAdversarial AttackTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: The study evaluates the generalization effect of backdoor attacks under domain transfer during the instruction tuning process of visual-language models.
Revisiting Fairness in Multitask Learning: A Performance-Driven Approach for Variance Reduction
Xiaohan Qin (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
ImageBenchmark
🎯 What it does: A dynamic weighted gradient aggregation method based on performance differences (PIVRG) is proposed for multi-task learning, aiming to reduce performance variance between tasks and enhance overall performance.
Revisiting Generative Replay for Class Incremental Object Detection
Shizhou Zhang (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
Object DetectionGenerationDiffusion modelImage
🎯 What it does: This paper proposes an image-level generative replay method based on Stable Diffusion and a similarity-based cross-sampling mechanism to prevent catastrophic forgetting in class-incremental object detection.
Revisiting MAE Pre-training for 3D Medical Image Segmentation
Tassilo Wald (German Cancer Research Center), Klaus Maier-Hein (German Cancer Research Center)
SegmentationConvolutional Neural NetworkAuto EncoderImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper uses Masked AutoEncoder for self-supervised pre-training on a large-scale brain MRI dataset and fine-tunes it on various brain segmentation tasks, significantly improving segmentation performance.
Revisiting Source-Free Domain Adaptation: Insights into Representativeness, Generalization, and Variety
Ronghang Zhu (University of Georgia), Sheng Li (University of Virginia)
ClassificationDomain AdaptationImage
🎯 What it does: Under the premise that source domain data is inaccessible, this paper theoretically analyzes and proposes an upper bound on the target domain empirical risk for Source-Free Domain Adaptation (SFDA), and based on this, designs a practical algorithm that includes cross-sampling, confidence-weighted semantic alignment, and progressive optimization.
Reward Fine-Tuning Two-Step Diffusion Models via Learning Differentiable Latent-Space Surrogate Reward
Zhiwei Jia (Zoom Communications), Gengdai Liu (Zoom Communications)
GenerationData SynthesisReinforcement Learning from Human FeedbackReinforcement LearningDiffusion modelImageText
🎯 What it does: This paper studies a reward fine-tuning method for two-step fast diffusion models, proposing the use of differentiable proxy reward learning in latent space to achieve gradient guidance for non-differentiable rewards.
REWIND: Real-Time Egocentric Whole-Body Motion Diffusion with Exemplar-Based Identity Conditioning
Jihyun Lee (KAIST), Jason Saragih (Meta)
GenerationPose EstimationKnowledge DistillationGraph Neural NetworkTransformerDiffusion modelVideo
🎯 What it does: A real-time, causal, diffusion model-based egocentric full-body motion estimation framework named REWIND is proposed.
ReWind: Understanding Long Videos with Instructed Learnable Memory
Anxhelo Diko (La Sapienza University of Roma), Ioannis Patras (Huawei)
RecognitionRetrievalComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideoTextMultimodality
🎯 What it does: ReWind is designed, a long video visual language model based on learnable memory, which can efficiently understand long video content while maintaining temporal consistency.
RGBAvatar: Reduced Gaussian Blendshapes for Online Modeling of Head Avatars
Linzhou Li (Zhejiang University), Kun Zhou (Zhejiang University)
GenerationCompressionComputational EfficiencyGaussian SplattingVideo
🎯 What it does: By compressing 3D Gaussian mixture shapes into a learnable small basis, high-fidelity, animatable head avatars are constructed in real-time using FLAME parameters.
RICCARDO: Radar Hit Prediction and Convolution for Camera-Radar 3D Object Detection
Yunfei Long (Michigan State University), Daniel Morris (Michigan State University)
Object DetectionAutonomous DrivingImageMultimodalityPoint Cloud
🎯 What it does: A camera-radar 3D object detection framework called RICCARDO is proposed, which predicts radar strike distribution and aligns it through convolution to obtain precise locations using monocular detection priors.
RigGS: Rigging of 3D Gaussians for Modeling Articulated Objects in Videos
Yuxin Yao (City University of Hong Kong), Junhui Hou (City University of Hong Kong)
GenerationData SynthesisPose EstimationGaussian SplattingVideoMesh
🎯 What it does: Automatically reconstructs editable and driveable 3D joint structures from monocular video using a 3D Gaussian representation and a skeleton-driven dynamic model, enabling the editing and transfer of new actions.
RipVIS: Rip Currents Video Instance Segmentation Benchmark for Beach Monitoring and Safety
Andrei Dumitriu (University of Wurzburg), Radu Timofte (University of Bucharest)
Object DetectionSegmentationVideoBenchmark
🎯 What it does: This paper presents the RipVIS dataset and its benchmark for surge flow detection in video instance segmentation, along with multiple model baselines and post-processing methods.
RivuletMLP: An MLP-based Architecture for Efficient Compressed Video Quality Enhancement
Gang He (Xidian University), Yunsong Li (Ant Group)
RestorationCompressionVideo
🎯 What it does: A compression video quality enhancement framework based on MLP, named RivuletMLP, is proposed, which utilizes three main modules: dynamic guided deformable alignment, spatiotemporal feature flow, and benign selection compensation to achieve effective alignment of multi-frame features and global information modeling.
RL-RC-DoT: A Block-level RL agent for Task-Aware Video Compression
Uri Gadot (Technion NVIDIA Research), Assaf Hallak (NVIDIA)
CompressionAutonomous DrivingReinforcement LearningVideo
🎯 What it does: A macroblock-level quantization parameter controller based on reinforcement learning, RL-RC-DoT, is proposed for real-time, task-aware video compression on the standard x264 encoder.
RLAIF-V: Open-Source AI Feedback Leads to Super GPT-4V Trustworthiness
Tianyu Yu (Tsinghua University), Maosong Sun (Tsinghua University)
GenerationData SynthesisOptimizationTransformerLarge Language ModelReinforcement LearningTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: The RLAIF-V framework is proposed, which enhances model credibility through AI feedback learning using open-source large multimodal language models (MLLM).
RNG: Relightable Neural Gaussians
Jiahui Fan (Nanjing University of Science and Technology), Beibei Wang (Nanjing University)
GenerationOptimizationNeural Radiance FieldGaussian SplattingImage
🎯 What it does: This paper proposes a relightable 3D asset framework based on 3D Gaussian splatting, modeling the radiance of each Gaussian point using implicit feature vectors, and achieving high-quality relighting of hard surface and soft boundary objects when the viewpoint and lighting direction change.
RoadSocial: A Diverse VideoQA Dataset and Benchmark for Road Event Understanding from Social Video Narratives
Chirag Parikh (International Institute of Information Technology Hyderabad), Ravi Kiran Sarvadevabhatla (International Institute of Information Technology Hyderabad)
Data SynthesisAutonomous DrivingTransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: RoadSocial has been constructed, which is a large-scale, multi-perspective, multi-national Road VideoQA dataset based on social media videos and their comments, containing 13.2k videos, 260k question-answer pairs, and 674 categories of labels;
RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete
Yuheng Ji (Peking University), Shanghang Zhang (Beijing Academy of Artificial Intelligence)
Robotic IntelligenceTransformerLarge Language ModelVision Language ModelImageVideoTextMultimodality
🎯 What it does: This paper proposes a unified robotic brain model called RoboBrain, which can convert abstract instructions into concrete execution plans and achieve operability perception and trajectory prediction for objects.
RoboGround: Robotic Manipulation with Grounded Vision-Language Priors
Haifeng Huang (Zhejiang University), Zhou Zhao (Shanghai AI Laboratory)
Robotic IntelligenceTransformerVision Language ModelMultimodality
🎯 What it does: This paper proposes the ROBOGROUND framework, which utilizes pixel-level localization masks as an intermediate representation to guide low-level grasping and placing strategies, and constructs a large-scale simulation dataset to enhance the robot's generalization ability to new objects and new scenes.
RoboPEPP: Vision-Based Robot Pose and Joint Angle Estimation through Embedding Predictive Pre-Training
Raktim Gautam Goswami (New York University), Farshad Khorrami (New York University)
Pose EstimationRobotic IntelligenceTransformerSupervised Fine-TuningImage
🎯 What it does: A vision-based robot pose and joint angle estimation framework called RoboPEPP is proposed, which utilizes self-supervised embedding prediction with joint occlusion to enhance the encoder's understanding of the robot's physical model.
RoboSense: Large-scale Dataset and Benchmark for Egocentric Robot Perception and Navigation in Crowded and Unstructured Environments
Haisheng Su (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
Object DetectionObject TrackingAutonomous DrivingRobotic IntelligenceTransformerSimultaneous Localization and MappingMultimodalityPoint CloudBenchmark
🎯 What it does: This paper presents RoboSense, a panoramic multimodal dataset and benchmark for social mobile robots in crowded unstructured environments, containing 133K frames of synchronized data, 1.4M 3D boxes, 216K trajectories, and occupancy labels;
RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Robotics
Chan Hee Song (Ohio State University), Stan Birchfield (NVIDIA)
Robotic IntelligenceTransformerSupervised Fine-TuningVision Language ModelImageTextPoint CloudBenchmark
🎯 What it does: The ROBOSPATIAL dataset is proposed for training and evaluating vision-language models aimed at robotic spatial reasoning.
Robotic Visual Instruction
Yanbang Li (Independent Researcher), Xianzheng Ma (South China University of Technology)
Object DetectionRobotic IntelligenceTransformerLarge Language ModelVision Language ModelImageTextChain-of-Thought
🎯 What it does: A hand-drawn visual instruction (RoVI) is proposed to control robots, and a VIEW pipeline is designed to convert RoVI into executable 3D actions.
RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins
Yao Mu (University of Hong Kong), Ping Luo (University of Hong Kong)
GenerationData SynthesisRobotic IntelligenceTransformerLarge Language ModelDiffusion modelImagePoint CloudBenchmark
🎯 What it does: Proposed and implemented the RoboTwin framework, which utilizes a single RGB image to automatically generate diverse digital twin objects and precise dual-arm demonstration data through a 3D generative model and LLM, and establishes a unified benchmark for dual-arm robot evaluation.
RobSense: A Robust Multi-modal Foundation Model for Remote Sensing with Static, Temporal, and Incomplete Data Adaptability
Minh Kha Do (La Trobe University), Wei Xiang (La Trobe University)
ClassificationSegmentationTransformerAuto EncoderContrastive LearningMultimodalityTime Series
🎯 What it does: Developed the RobSense multimodal foundation model, supporting static, time series, and missing input for multispectral (MS) and synthetic aperture radar (SAR) data;
Robust 3D Shape Reconstruction in Zero-Shot from a Single Image in the Wild
Junhyeong Cho (POSTECH), Tae-Hyun Oh (KAIST)
GenerationData SynthesisDepth EstimationTransformerDiffusion modelImagePoint CloudMesh
🎯 What it does: A unified regression model is proposed to achieve zero-shot and occlusion-aware 3D shape reconstruction from a single image in real scenes, and large-scale diverse synthetic data is constructed for training through a generative model.
Robust Audio-Visual Segmentation via Audio-Guided Visual Convergent Alignment
Chen Liu (University of Queensland), Xin Yu (University of Queensland)
SegmentationTransformerContrastive LearningMultimodalityAudio
🎯 What it does: An end-to-end framework is proposed that combines Audio-Guided Modality Alignment and Uncertainty Estimation to enhance the robustness and accuracy of Audio-Visual Segmentation (AVS).
Robust Message Embedding via Attention Flow-Based Steganography
Huayuan Ye (East China Normal University), Chenhui Li (East China Normal University)
Data SynthesisCompressionSafty and PrivacyTransformerFlow-based ModelImage
🎯 What it does: A robust information embedding framework called RMSteg based on attention flow is proposed, utilizing reversible QR code transfer, reversible token fusion, and attention coupling networks to achieve high capacity, robustness, and high-quality image steganography.
Robust Multi-Object 4D Generation for In-the-wild Videos
Wen-Hsuan Chu (Carnegie Mellon University), Katerina Fragkiadaki (Carnegie Mellon University)
GenerationData SynthesisDepth EstimationDiffusion modelScore-based ModelGaussian SplattingVideo
🎯 What it does: This paper presents GenMOJO, a framework capable of generating complete 4D scenes (spatial-temporal 3D models) from monocular multi-object videos that include severe occlusions and high-speed motion.
Robust Multimodal Survival Prediction with Conditional Latent Differentiation Variational AutoEncoder
Junjie Zhou (Nanjing University of Aeronautics and Astronautics), Wei Shao (Nanjing University of Aeronautics and Astronautics)
GenerationData SynthesisRepresentation LearningTransformerAuto EncoderMultimodalityBiomedical Data
🎯 What it does: This study investigates the generation of multifunctional genomic embeddings from pathological images for robust multimodal survival prediction in the context of missing genomic data.
Robust-MVTON: Learning Cross-Pose Feature Alignment and Fusion for Robust Multi-View Virtual Try-On
Nannan Zhang (CUHKSZ), Xiaoguang Han (CUHKSZ)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes Robust-MVTON, an end-to-end multi-view virtual try-on method that utilizes front and back clothing images and achieves high-quality try-on effects from arbitrary angles through cross-pose feature alignment and fusion.
ROCKET-1: Mastering Open-World Interaction with Visual-Temporal Context Prompting
Shaofei Cai (Peking University), Yitao Liang (Peking University)
Robotic IntelligenceTransformerReinforcement LearningPrompt EngineeringVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: This paper proposes a hierarchical architecture based on visual-temporal context prompts, utilizing VLM to generate segmentation and interaction instructions, training the low-level policy ROCKET-1, and tracking targets in real-time through SAM-2 to accomplish complex interactions and long-term tasks in Minecraft.
ROD-MLLM: Towards More Reliable Object Detection in Multimodal Large Language Models
Heng Yin (Tongji University), Yongtao Hao (Tongji University)
Object DetectionTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: ROD-MLLM is proposed, a multimodal large language model capable of reliable object detection based on free text descriptions.
RoGSplat: Learning Robust Generalizable Human Gaussian Splatting from Sparse Multi-View Images
Junjin Xiao (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
GenerationData SynthesisGaussian SplattingImage
🎯 What it does: RoGSplat proposes a general human novel view rendering method based on 3D Gaussian Splatting, capable of generating high-fidelity novel views from sparse multi-view images without the need for optimization for each subject.
ROICtrl: Boosting Instance Control for Visual Generation
Yuchao Gu (National University of Singapore), Mike Zheng Shou (National University of Singapore)
Object DetectionGenerationDiffusion modelImageBenchmark
🎯 What it does: An adapter named ROICtrl is proposed, which enables multi-instance control in pre-trained diffusion models, allowing each instance to be precisely generated through free text descriptions and corresponding bounding boxes.
ROLL: Robust Noisy Pseudo-label Learning for Multi-View Clustering with Noisy Correspondence
Yuan Sun (Sichuan University), Peng Hu (Sichuan University)
Auto EncoderContrastive LearningMultimodality
🎯 What it does: A robust pseudo-label learning framework ROLL is proposed to simultaneously address pseudo-label noise (NPP) and correspondence noise (NCP) issues in multi-view clustering.
RoomPainter: View-Integrated Diffusion for Consistent Indoor Scene Texturing
Zhipeng Huang (Peking University), Yonghong Tian (Peking University)
GenerationData SynthesisDiffusion modelImagePoint Cloud
🎯 What it does: This paper presents RoomPainter, which utilizes a 2D diffusion model to achieve high-quality, cross-view consistent texture generation for indoor scenes.
RoomTour3D: Geometry-Aware Video-Instruction Tuning for Embodied Navigation
Mingfei Han (MBZUAI), Ivan Laptev (MBZUAI)
Robotic IntelligenceTransformerLarge Language ModelVision Language ModelVideoTextRetrieval-Augmented Generation
🎯 What it does: Proposes the RoomTour3D dataset, which automatically generates geometrically aware navigation trajectories and natural language instructions from indoor tour videos, supporting multi-task VLN training.
RORem: Training a Robust Object Remover with Human-in-the-Loop
Ruibin Li (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
RestorationObject DetectionKnowledge DistillationDiffusion modelImage
🎯 What it does: A robust object removal model RORem based on semi-supervised learning and human-computer interaction is proposed, and a dataset of approximately 200K object removal pairs is constructed.
ROS-SAM: High-Quality Interactive Segmentation for Remote Sensing Moving Object
Zhe Shan (Hainan University), Xia Xie (Tianjin University)
Object DetectionSegmentationTransformerSupervised Fine-TuningImageVideo
🎯 What it does: To achieve high-quality interactive segmentation of moving objects in remote sensing videos, the ROS-SAM method is proposed, along with a specially designed data and inference pipeline.
Rotation-Equivariant Self-Supervised Method in Image Denoising
Hanze Liu (Xi'an Jiaotong University), Deyu Meng (Macau University of Science and Technology)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a self-supervised image denoising network called AdaReNet, which integrates rotational equivariance priors into the self-supervised denoising framework for the first time, and theoretically analyzes the impact of upsampling and downsampling on equivariance within the U-Net structure.
RSAR: Restricted State Angle Resolver and Rotated SAR Benchmark
Xin Zhang (Nankai University), Xiang Li (Nankai University)
Object DetectionImageBenchmark
🎯 What it does: To address the issue of discontinuity in angle prediction boundaries in rotating object detection, a Unit Circle Constrained Angle Resolver (UCR) is proposed, which enhances the angle prediction accuracy of weakly supervised models. Based on this, the first large-scale multi-class rotating SAR object detection dataset, RSAR, is constructed.
RUBIK: A Structured Benchmark for Image Matching across Geometric Challenges
Thibaut Loiseau (Ecole des Ponts Université Gustave Eiffel), Guillaume Bourmaud (Université de Bordeaux)
Pose EstimationAutonomous DrivingSimultaneous Localization and MappingImageBenchmark
🎯 What it does: The RUBIK benchmark is proposed, which constructs 16.5K pairs of images based on nuScenes images, categorized into 33 difficulty levels according to overlap, scale ratio, and viewpoint difference, for the systematic evaluation of camera pose estimation methods.
S^3-Face: SSS-Compliant Facial Reflectance Estimation via Diffusion Priors
Xingyu Ren (Shanghai Jiao Tong University), Chao Ma (Shanghai Jiao Tong University)
RestorationGenerationDiffusion modelImage
🎯 What it does: Proposes the S3-Face framework, using the Stable Diffusion model as a prior, with a two-stage training process to estimate SSS-compatible facial reflection features from a single image, including five channels: diffuse reflection, specular reflection, normals, hemoglobin, and melanin.
S2D-LFE: Sparse-to-Dense Light Field Event Generation
Yutong Liu (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationGenerationData SynthesisSuper ResolutionDiffusion modelAuto EncoderImageVideo
🎯 What it does: Proposed the S2D-LFE method, which utilizes sparse perspective light field events (LFE) to generate dense, spatiotemporally consistent new perspective LFEs;
S2Gaussian: Sparse-View Super-Resolution 3D Gaussian Splatting
Yecong Wan (China University of Petroleum), Wangmeng Zuo (Harbin Institute of Technology)
RestorationDepth EstimationSuper ResolutionGaussian SplattingPoint Cloud
🎯 What it does: The S2Gaussian framework is proposed, which can reconstruct high-quality, detail-rich 3D scenes using only sparse and low-resolution views.
S4-Driver: Scalable Self-Supervised Driving Multimodal Large Language Model with Spatio-Temporal Visual Representation
Yichen Xie (University of California Berkeley), Mingxing Tan (Waymo LLC)
Autonomous DrivingTransformerLarge Language ModelMultimodality
🎯 What it does: A scalable self-supervised end-to-end driving planning framework S4-Driver is proposed, utilizing the multimodal large language model PaLI to generate driving trajectories in three-dimensional space.
SACB-Net: Spatial-awareness Convolutions for Medical Image Registration
Xinxing Cheng (University of Birmingham), Jinming Duan (University of Manchester)
Image TranslationSegmentationConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a 3D medical image registration network named SACB-Net, which utilizes Spatially Aware Convolution Blocks (SACB) to adaptively generate convolution kernels, thereby enhancing feature representation in different spatial regions.
SAIST: Segment Any Infrared Small Target Model Guided by Contrastive Language-Image Pretraining
Mingjin Zhang (Xidian University), Jing Zhang (Wuhan University)
Object DetectionContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a multimodal framework SAIST, which utilizes the joint information of text and infrared images to achieve small target detection.
SALAD: Skeleton-aware Latent Diffusion for Text-driven Motion Generation and Editing
Seokhyeon Hong (Korea Advanced Institute of Science and Technology), Junyong Noh (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderVideoText
🎯 What it does: A text-driven action generation and editing framework based on Skeleton-aware Latent Diffusion (SALAD) is proposed;
Saliuitl: Ensemble Salience Guided Recovery of Adversarial Patches against CNNs
Mauricio Byrd Victorica (KTH Royal Institute of Technology), Henrik Sandberg (KTH Royal Institute of Technology)
RestorationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A two-stage adversarial patch recovery method called Saliuitl is proposed, which first detects the presence of patches through a multi-threshold feature map set, and then performs recovery;
SALOVA: Segment-Augmented Long Video Assistant for Targeted Retrieval and Routing in Long-Form Video Analysis
Junho Kim (Korea Advanced Institute of Science and Technology), Yong Man Ro (Korea Advanced Institute of Science and Technology)
RetrievalTransformerLarge Language ModelVideoMultimodalityRetrieval-Augmented Generation
🎯 What it does: The SALOVA framework is proposed, utilizing retrieval-driven long video understanding, and the SceneWalk dataset is constructed to support model learning.
SAM-I2V: Upgrading SAM to Support Promptable Video Segmentation with Less than 0.2% Training Cost
Haiyang Mei (National University of Singapore), Mike Zheng Shou (National University of Singapore)
SegmentationConvolutional Neural NetworkPrompt EngineeringImageVideo
🎯 What it does: By upgrading the pre-trained Segment Anything Model (SAM) from image to video, SAM-I2V is proposed to support Promptable video segmentation.
SAM-REF: Introducing Image-Prompt Synergy during Interaction for Detail Enhancement in the Segment Anything Model
Chongkai Yu (Meitu Inc), Xiaolin Hu (Tsinghua University)
SegmentationTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes SAM-REF, an interactive segmentation framework that combines early and late fusion, utilizing a lightweight refiner to achieve global and local refinement based on SAM, significantly improving detail segmentation quality.
SAM2-LOVE: Segment Anything Model 2 in Language-aided Audio-Visual Scenes
Yuji Wang (Tsinghua University), Yansong Tang (Zhejiang University)
Object DetectionSegmentationTransformerPrompt EngineeringVideoTextMultimodalityBenchmarkAudio
🎯 What it does: Proposes the SAM2-LOVE framework, which utilizes text, audio, and visual tri-modal information to generate learnable tokens through a fusion module and prompts SAM2, achieving pixel-level reference segmentation (Ref-AVS) tasks for language-assisted audio-video scenes.
SAM2Object: Consolidating View Consistency via SAM2 for Zero-Shot 3D Instance Segmentation
Jihuai Zhao (University of Science and Technology Beijing), Huimin Ma (University of Science and Technology Beijing)
Object DetectionSegmentationPoint Cloud
🎯 What it does: Proposes SAM2Object, which utilizes the multi-view consistency of SAM2 to achieve zero-shot 3D instance segmentation.
SaMam: Style-aware State Space Model for Arbitrary Image Style Transfer
Hongda Liu (Sun Yat-Sen University), Yulan Guo (Sun Yat-Sen University)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: A style-aware network called SaMam based on Mamba is proposed for arbitrary image style transfer.
Samba: A Unified Mamba-based Framework for General Salient Object Detection
Jiahao He (Sichuan University), Qijun Zhao (Sichuan University)
Object DetectionConvolutional Neural NetworkTransformerImageVideoMultimodality
🎯 What it does: A unified framework called Samba based on Mamba is proposed to handle various salient object detection tasks, including RGB/RGB-D/RGB-T, video SOD, and RGB-D video SOD;
SAMBLE: Shape-Specific Point Cloud Sampling for an Optimal Trade-Off Between Local Detail and Global Uniformity
Chengzhi Wu (Karlsruhe Institute of Technology), Jürgen Beyerer (Fraunhofer IOSB)
ClassificationSegmentationOptimizationTransformerPoint Cloud
🎯 What it does: A SAMBLE point cloud sampling method is proposed, utilizing sparse attention maps to learn shape-specific sampling strategies that balance local detail and global uniformity.
Sample- and Parameter-Efficient Auto-Regressive Image Models
Elad Amrani (Apple), Alex Bronstein (Technion)
RecognitionGenerationTransformerImage
🎯 What it does: We propose XTRA, a self-regressive image model based on Vision Transformer, which predicts pixels block by block using a Block Causal Mask;
Sampling Innovation-Based Adaptive Compressive Sensing
Zhifu Tian (Information Engineering University), Shu Wang (Information Engineering University)
RestorationCompressionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Sampling Innovation-based Adaptive Compressed Sensing (SIB-ACS) framework, which combines multi-stage negative feedback adaptive sampling with a novel reconstruction network PCCD-Net to achieve high-fidelity image reconstruction.
SAMWISE: Infusing Wisdom in SAM2 for Text-Driven Video Segmentation
Claudia Cuttano (Politecnico di Torino), Giuseppe Averta (Focoos AI)
Object TrackingSegmentationTransformerVision Language ModelVideoText
🎯 What it does: By combining SAM2 with textual information, the SAMWISE model is proposed to achieve text-driven video segmentation (RVOS) and address tracking deviation issues in a streaming processing environment.
SapiensID: Foundation for Human Recognition
Minchul Kim (Michigan State University), Xiaoming Liu (Michigan State University)
RecognitionRetrievalTransformerImage
🎯 What it does: SapiensID is proposed, a unified human identification model that can simultaneously process facial and body features and achieve recognition under various poses and scales.
SAR3D: Autoregressive 3D Object Generation and Understanding via Multi-scale 3D VQVAE
Yongwei Chen, Xingang Pan
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningImageTextPoint Cloud
🎯 What it does: The SAR3D framework is proposed, which discretizes and encodes 3D objects using a multi-scale 3D VQVAE, and employs an autoregressive Transformer based on next-scale prediction to quickly generate 3D objects from images or text. It also fine-tunes a pre-trained LLM using truncated multi-scale 3D codes to achieve natural language descriptions and explanations of 3D objects.
SASep: Saliency-Aware Structured Separation of Geometry and Feature for Open Set Learning on Point Clouds
Jinfeng Xu (Huazhong University of Science and Technology), Min Chen (South China University of Technology)
ClassificationRecognitionGraph Neural NetworkPoint Cloud
🎯 What it does: This paper proposes an open set recognition method for point clouds called SASep, which decomposes objects using semantic saliency and distinguishes between known and unknown categories through geometric synthesis and feature distance enhancement.
SAT-HMR: Real-Time Multi-Person 3D Mesh Estimation via Scale-Adaptive Tokens
Chi Su (Peking University), Yizhou Wang (Peking University)
Pose EstimationComputational EfficiencyTransformerImageMesh
🎯 What it does: A real-time multi-person 3D human mesh estimation framework based on scale-adaptive tokens, SAT-HMR, is proposed, achieving efficient and accurate multi-person mesh regression from a single RGB image.
SATA: Spatial Autocorrelation Token Analysis for Enhancing the Robustness of Vision Transformers
Nick Nikzad (Griffith University), Jun Zhou (Griffith University)
ClassificationTransformerImage
🎯 What it does: Without the need for additional training, Spatial Autocorrelation Token Analysis (SATA) is introduced, enhancing the model's representation ability and robustness by segmenting and merging the spatial autocorrelation of ViT feature tokens.
Satellite Observations Guided Diffusion Model for Accurate Meteorological States at Arbitrary Resolution
Siwei Tu (Fudan University), Lei Bai (Fudan University)
GenerationData SynthesisOptimizationDiffusion modelTime Series
🎯 What it does: A diffusion model conditioned on satellite observations (SGD) has been constructed, which generates high-resolution meteorological fields at a scale of 6.25 km by using low-resolution ERA5 images, GridSat satellite data, and meteorological station observations, achieving precise downscaling of meteorological states.
Satellite to GroundScape - Large-scale Consistent Ground View Generation from Satellite Views
Ningli Xu (Ohio State University), Rongjun Qin (Ohio State University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a cross-view synthesis framework for generating multi-view consistent ground views from satellite images.
Scalable Autoregressive Monocular Depth Estimation
Jinhong Wang (Zhejiang University), Jian Wu (Zhejiang University)
Depth EstimationTransformerImage
🎯 What it does: A self-autoregressive monocular depth estimation framework (DAR) is proposed, which first predicts a low-resolution depth map and then gradually refines it to high resolution, while recursively refining the discrete intervals of depth values, achieving dual autoregression at both resolution and fine-grained levels.
Scalable Video-to-Dataset Generation for Cross-Platform Mobile Agents
Yunseok Jang (University of Michigan), Honglak Lee (University of Michigan)
Object DetectionData SynthesisRobotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningVideoMultimodality
🎯 What it does: An automated process was developed to extract and annotate mobile operation workflows from YouTube tutorial videos, resulting in the creation of the cross-platform (iOS/Android) mobile OS navigation dataset MONDAY, which includes 20K videos and 313K frames. Pre-training and fine-tuning of models were conducted on this dataset.
Scale Efficient Training for Large Datasets
Qing Zhou (Northwestern Polytechnical University), Qi Wang (Northwestern Polytechnical University)
SegmentationRetrievalComputational EfficiencyImage
🎯 What it does: A dynamic sample pruning framework called SeTa is proposed, which achieves training acceleration on large-scale datasets through random downsampling, loss-based clustering, and a sliding window strategy, while reducing training costs without compromising or even enhancing model performance.
ScaleLSD: Scalable Deep Line Segment Detection Streamlined
Zeran Ke (Wuhan University), Nan Xue (Wuhan University)
Object DetectionTransformerImage
🎯 What it does: This study proposes a self-supervised learning line segment detection model, ScaleLSD, that can operate on a large number of unlabeled images and achieves zero-shot performance improvements across various tasks.