ECCV 2024 Papers — Page 18
European Conference on Computer Vision · 2387 papers
Radiative Gaussian Splatting for Efficient X-ray Novel View Synthesis
Yuanhao Cai (Johns Hopkins University), Alan Yuille (Johns Hopkins University)
Data SynthesisGaussian SplattingBiomedical DataComputed Tomography
🎯 What it does: Propose a novel X-ray novel view synthesis method called X-Gaussian based on 3D Gaussian point clouds;
RaFE: Generative Radiance Fields Restoration
Zhongkai Wu, Dong Xu (City University of Hong Kong)
RestorationDiffusion modelNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: Proposed a general NeRF recovery framework called RaFE, which can recover high-quality radiance fields from low-quality images with various distortions (low resolution, blur, noise, compression, etc.), while preserving both geometric and appearance details;
Raindrop Clarity: A Dual-Focused Dataset for Day and Night Raindrop Removal
Yeying Jin (National University of Singapore), Malu Zhang (University of Electronic Science and Technology of China)
RestorationData SynthesisDiffusion modelImageBenchmark
🎯 What it does: This paper constructs a large-scale real-world raindrop removal dataset called Raindrop Clarity, which includes paired and triplet images of daytime and nighttime scenes, as well as focused raindrops and focused backgrounds.
Raising the Ceiling: Conflict-Free Local Feature Matching with Dynamic View Switching
Xiaoyong Lu (Southeast University), Songlin Du (Southeast University)
Pose EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose a semi-sparse and coarse-to-fine matching combined feature matching framework RCM, including dynamic view switching, conflict-free coarse matching, and multi-scale fine matching, to improve matching accuracy and efficiency.
Random Walk on Pixel Manifolds for Anomaly Segmentation of Complex Driving Scenes
Zelong Zeng (SenseTime Japan), Kaname Tomite (SenseTime Japan)
Anomaly DetectionAutonomous DrivingGraph Neural NetworkImage
🎯 What it does: Propose an anomaly segmentation method called RWPM based on pixel manifold random walks, which reshapes pixel embeddings through random walks to enhance segmentation accuracy.
RangeLDM: Fast Realistic LiDAR Point Cloud Generation
Qianjiang Hu (Peking University), Wei Hu (Peking University)
GenerationData SynthesisDiffusion modelAuto EncoderGenerative Adversarial NetworkPoint Cloud
🎯 What it does: This paper proposes a LiDAR point cloud generation method called RangeLDM based on latent diffusion models, which can quickly generate high-quality 3D point clouds and support point cloud upsampling and filling.
RANRAC: Robust Neural Scene Representations via Random Ray Consensus
Benno Buschmann (Delft University of Technology), Bernhard Egger (Delft University of Technology)
GenerationNeural Radiance FieldImage
🎯 What it does: Proposes a robust neural scene representation method based on Random Ray Consensus (RANRAC), capable of achieving high-quality reconstruction even in the presence of occlusions and camera parameter errors.
RAP: Retrieval-Augmented Planner for Adaptive Procedure Planning in Instructional Videos
Ali Zare (Columbia University), Shih-Fu Chang (Columbia University)
RetrievalTransformerVision Language ModelVideoRetrieval-Augmented Generation
🎯 What it does: Proposed an instruction video program planning framework named RAP with adaptive step length.
RAPiD-Seg: Range-Aware Pointwise Distance Distribution Networks for 3D LiDAR Segmentation
Li Li (Durham University), Toby P Breckon
SegmentationAutonomous DrivingConvolutional Neural NetworkAuto EncoderContrastive LearningPoint Cloud
🎯 What it does: This paper proposes the RAPiD (Range-Aware Pointwise Distance Distribution) feature and the RAPiD-Seg network for 3D LiDAR point cloud semantic segmentation.
Rasterized Edge Gradients: Handling Discontinuities Differentially
Stanislav Pidhorskyi (Meta), Jason Saragih (Meta)
Computational EfficiencyImageVideo
🎯 What it does: Propose a method based on micro-edges that can maintain the traditional discrete pixel rasterization results while efficiently approximating and solving the gradients of differentiable renderers at visibility discontinuities.
Rate-Distortion-Cognition Controllable Versatile Neural Image Compression
Jinming Liu (Shanghai Jiao Tong University), Xin Jin (Ningbo Institute of Digital Twin)
CompressionAuto EncoderContrastive LearningImage
🎯 What it does: Designed an image compression model that achieves controllable bitrate, distortion, and machine task performance within a single neural network, termed Rate-Distortion-Cognition Controllable Multifunctional Neural Image Compression.
RAVE: Residual Vector Embedding for CLIP-Guided Backlit Image Enhancement
Tatiana Gaintseva (Queen Mary University of London), Gregory Slabaugh (Queen Mary University of London)
RestorationConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes two improved CLIP-guided methods for single-image backlit enhancement tasks—CLIP-LIT-Latent and RAVE, achieving higher quality image enhancement.
RAW-Adapter: Adapting Pretrained Visual Model to Camera RAW Images
Ziteng Cui (University of Tokyo), Tatsuya Harada (University of Tokyo)
Object DetectionSegmentationDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Proposes the RAW-Adapter framework to adapt pre-trained sRGB visual models to camera RAW images, enhancing compatibility between RAW inputs and sRGB models through input-layer and model-layer adapters.
Rawformer: Unpaired Raw-to-Raw Translation for Learnable Camera ISPs
Georgy Perevozchikov (University of Würzburg), Radu Timofte (University of Würzburg)
Image TranslationDomain AdaptationTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: Propose Rawformer, a fully unsupervised Transformer-based raw-to-raw translation framework that can map raw images captured by one camera to the raw domain of another camera, enabling pre-trained neural ISPs to directly process raw images from new cameras without the need to re-collect paired data.
Ray Denoising: Depth-aware Hard Negative Sampling for Multi-view 3D Object Detection
Feng Liu (University of Chinese Academy of Sciences), Yanzhao Zhou (University of Chinese Academy of Sciences)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: Propose the Ray Denoising method, which samples depth-aware hard negative samples along camera rays in multi-view 3D detection, helping the model distinguish real objects from pseudo targets generated by depth ambiguity.
Ray-Distance Volume Rendering for Neural Scene Reconstruction
Ruihong Yin (University of Amsterdam), Theo Gevers (University of Amsterdam)
GenerationNeural Radiance FieldImage
🎯 What it does: Propose a volumetric rendering method based on ray-specific SRDF for indoor scene reconstruction.
RCS-Prompt: Learning Prompt to Rearrange Class Space for Prompt-based Continual Learning
Longrong Yang (Zhejiang University), Xi Li (Ant Services Group)
ClassificationRepresentation LearningTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: Designed a novel Prompt training method called RCS-Prompt for Prompt-Based lifelong learning without example archives, aiming to reduce overlap between old and new class spaces by learning to rearrange the category space, thereby enhancing classification boundaries.
Real Appearance Modeling for More General Deepfake Detection
Jiahe Tian (University Of Chinese Academy Of Sciences), Jizhong Han
RestorationAnomaly DetectionTransformerAuto EncoderImage
🎯 What it does: This paper proposes a deepfake detection framework based on Realistic Appearance Modeling (RAM), which generates forged samples using custom facial perturbations (Texture Disturbance and Structure Disturbance), and learns the fine-grained appearance of real human faces on forged images through a Recovery Autoencoder (RAE), thereby achieving generalized detection of unknown deepfake techniques.
Real-data-driven 2000 FPS Color Video from Mosaicked Chromatic Spikes
Siqi Yang (Peking University), Boxin Shi (Peking University)
RestorationGenerationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: Propose an end-to-end self-supervised reconstruction framework that generates high dynamic range color videos at 2000 FPS using a single camera equipped with a Bayer CFA;
Real-time 3D-aware Portrait Editing from a Single Image
Qingyan Bai (Hong Kong University of Science and Technology), Qifeng Chen (Hong Kong University of Science and Technology)
GenerationKnowledge DistillationPrompt EngineeringVision Language ModelDiffusion modelGenerative Adversarial NetworkImageTextMultimodality
🎯 What it does: This paper proposes 3DPE, a real-time 3D-aware portrait editing method based on a single portrait image, supporting both image and text prompts;
Real-time Holistic Robot Pose Estimation with Unknown States
Shikun Ban, Yizhou Wang (Peking University)
Pose EstimationDepth EstimationRobotic IntelligenceConvolutional Neural NetworkGraph Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: This paper proposes an end-to-end, real-time panoramic robot pose estimation framework that can directly predict the 6D pose from the camera to the robot, joint angles, root depth, and keypoint positions under the condition of using only a single RGB image and without knowing the robot's joint states.
ReALFRED: An Embodied Instruction Following Benchmark in Photo-Realistic Environments
Taewoong Kim (Seoul National University), Jonghyun Choi (Seoul National University)
Robotic IntelligenceVision-Language-Action ModelTextMultimodalityPoint CloudMeshBenchmark
🎯 What it does: Proposed and made public the ReALFRED benchmark, which uses multi-room environments with real 3D scans combined with interactive objects and free-form natural language instructions, aiming to train robots to complete long-term daily household tasks.
RealGen: Retrieval Augmented Generation for Controllable Traffic Scenarios
Wenhao Ding (Carnegie Mellon University), Marco Pavone (NVIDIA Research)
GenerationData SynthesisAutonomous DrivingTransformerContrastive LearningPoint CloudRetrieval-Augmented Generation
🎯 What it does: This paper proposes RealGen, a traffic scene generation framework based on retrieval-augmented generation, achieving controllable and realistic traffic scenario synthesis.
Realistic Human Motion Generation with Cross-Diffusion Models
Zeping Ren (Tsinghua Shenzhen International Graduate School), Xiu Li (Tsinghua Shenzhen International Graduate School)
GenerationTransformerVision Language ModelDiffusion modelTextMultimodality
🎯 What it does: Achieving high-quality 3D and 2D human motion generation based on text through cross-diffusion models.
RealViformer: Investigating Attention for Real-World Video Super-Resolution
Yuehan Zhang (National University of Singapore), Angela Yao (National University of Singapore)
RestorationSuper ResolutionTransformerOptical FlowVideo
🎯 What it does: Studies the attention mechanisms in real-time video super-resolution (RWVSR), exploring the sensitivity of spatial attention and channel attention on real-world degraded videos, and proposes an improved solution to address the feature redundancy caused by channel attention. Based on this research, designs RealViformer: a unidirectional recursive Transformer structure that employs channel attention fusion and improved channel attention (ICA), enhancing RWVSR performance without introducing a large number of parameters.
Reason2Drive: Towards Interpretable and Chain-based Reasoning for Autonomous Driving
Ming Nie (Fudan University), Li Zhang (Fudan University)
Autonomous DrivingExplainability and InterpretabilityTransformerVision Language ModelVideoTextChain-of-Thought
🎯 What it does: This paper proposes an autonomous driving framework based on interpretable chain reasoning, constructs a large-scale dataset named Reason2Drive, and designs a specialized evaluation metric called ADRScore for chain reasoning, further improving visual language models (VLMs) to better utilize visual prior information.
Rebalancing Using Estimated Class Distribution for Imbalanced Semi-Supervised Learning under Class Distribution Mismatch
Taemin Park (KAIST), Heeyoung Kim (KAIST)
ClassificationImage
🎯 What it does: Propose a semi-supervised learning framework RECD that addresses label imbalance and distribution mismatch between unlabeled and labeled data by estimating the class distribution of unlabeled data to rebalance the model.
Receler: Reliable Concept Erasing of Text-to-Image Diffusion Models via Lightweight Erasers
Chi-Pin Huang (National Taiwan University), Yu-Chiang Frank Wang (National Taiwan University)
GenerationComputational EfficiencyPrompt EngineeringDiffusion modelImageText
🎯 What it does: The paper proposes a lightweight concept elimination method called Receler, which reliably removes specified concepts from text-to-image diffusion models without accessing image data.
ReCON: Training-Free Acceleration for Text-to-Image Synthesis with Retrieval of Concept Prompt Trajectories
Chen-Yi Lu (Purdue University), Somali Chaterji (Purdue University)
GenerationComputational EfficiencyLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Propose a training-agnostic retrieval-based framework called ReCon for accelerating text-to-image diffusion models. It generates faster and higher fidelity images by extracting visual concepts from prompts and retrieving noise trajectories of similar concepts.
Reconstruction and Simulation of Elastic Objects with Spring-Mass 3D Gaussians
Licheng Zhong (Stanford University), Yunzhu Li (Stanford University)
OptimizationGaussian SplattingVideoMeshPhysics Related
🎯 What it does: Propose Spring-Gaus, a reversible and simulatable 3D object representation method that combines a 3D elastic spring-mass model with a 3D Gaussian kernel, enabling simultaneous reconstruction of appearance, geometry, and physical dynamics parameters from multi-view videos;
Rectify the Regression Bias in Long-Tailed Object Detection
Ke Zhu (Nanjing University), Jianxin Wu (Nanjing University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: What was done: The study addresses the regression bias problem in long-tailed object detection, proposing methods to correct regression bias by introducing class-agnostic regression heads, clustering heads, and merging heads, thereby improving detection accuracy.
RecurrentBEV: A Long-term Temporal Fusion Framework for Multi-view 3D Detection
Ming Chang (Cambricon Technologies), Shaoli Liu (Cambricon Technologies)
Object DetectionAutonomous DrivingConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: Propose a novel recursive temporal fusion framework called RecurrentBEV for bird's-eye-view 3D object detection using multi-view cameras.
Recursive Visual Programming
Jiaxin Ge (UC Berkeley), Trevor Darrell (UC Berkeley)
AI Code AssistantTransformerLarge Language ModelPrompt EngineeringImageVideoMultimodality
🎯 What it does: Propose a Recursive Visual Programming (RVP) approach, enabling Large Language Models (LLMs) to generate modular, dynamically typed code through recursive self-calls in Visual Question Answering (VQA) tasks, progressively decomposing complex problems layer by layer;
REDIR: Refocus-free Event-based De-occlusion Image Reconstruction
Qi Guo (Institute of Microelectronics of Chinese Academy of Sciences), Xingyu Gao (Institute of Microelectronics of Chinese Academy of Sciences)
RestorationSuper ResolutionConvolutional Neural NetworkSpiking Neural NetworkImage
🎯 What it does: Designed and implemented an end-to-end event camera reconstruction model named REDIR, which can accomplish event alignment, filtering, and reconstruction in occlusion-free event-based synthetic aperture imaging (E-SAI), ultimately generating high-resolution images without occlusion.
Ref-AVS: Refer and Segment Objects in Audio-Visual Scenes
Yaoting Wang (Renmin University of China), Di Hu (Renmin University of China)
SegmentationTransformerPrompt EngineeringVision Language ModelVideoTextMultimodalityBenchmarkAudio
🎯 What it does: This paper proposes a new multimodal reference segmentation task called Ref-AVS, requiring the model to precisely segment target objects in videos based on natural language expressions containing audio, visual, and temporal cues;
Referring Atomic Video Action Recognition
Kunyu Peng (Karlsruhe Institute of Technology), Alina Roitberg (University of Stuttgart)
RecognitionTransformerVision Language ModelVision-Language-Action ModelMultimodalityBenchmark
🎯 What it does: Propose a new task—Reference-based Atomic Video Action Recognition (RAVAR), and design the corresponding RefAVA dataset; implement and evaluate various baseline methods, ultimately proposing RefAtomNet which significantly improves performance through cross-stream attention fusion.
Reflective Instruction Tuning: Mitigating Hallucinations in Large Vision-Language Models
Jinrui Zhang (Southern University of Science and Technology), Feng Zheng (Cloud Computing and IT Institute of ZTE Corporation)
TransformerSupervised Fine-TuningVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposed the 'Reflective Instruction Tuning' method, combining positive and negative reasoning steps (rationale) to train large-scale vision-language models, thereby reducing hallucinations.
REFRAME: Reflective Surface Real-Time Rendering for Mobile Devices
Chaojie Ji, Yiyi Liao (Zhejiang University)
GenerationNeural Radiance FieldGaussian SplattingImageMesh
🎯 What it does: This paper proposes a grid-based real-time rendering framework called REFRAME, specifically designed for novel view synthesis of reflective surfaces on mobile devices.
Region-Adaptive Transform with Segmentation Prior for Image Compression
Yuxi Liu (Beijing Jiaotong University), Yao Zhao (Beijing Jiaotong University)
SegmentationCompressionConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes an end-to-end learning-based image compression framework named Segmentation-Prior-Guided Image Compression (SegPIC), which utilizes class-agnostic semantic masks during training to guide the network to generate region-adaptive transformations, thereby improving pixel-level compression quality.
Region-aware Distribution Contrast: A Novel Approach to Multi-Task Partially Supervised Learning
Meixuan Li (University of Electronic Science and Technology of China), Jie Zou (University of Electronic Science and Technology of China)
SegmentationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: By converting the multi-task sparse supervision learning problem into regional-level distribution comparison, this paper achieves cross-task consistency among different tasks.
Region-Aware Sequence-to-Sequence Learning for Hyperspectral Denoising
JiaHua Xiao, Xing Wei (Xi'an Jiaotong University)
RestorationRecurrent Neural NetworkImageBenchmark
🎯 What it does: This paper proposes a region-aware sequence-to-sequence learning framework named RAS2S for hyperspectral image denoising.
Region-centric Image-Language Pretraining for Open-Vocabulary Detection
Dahun Kim (Google DeepMind), Weicheng Kuo (Google DeepMind)
Object DetectionKnowledge DistillationTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Integrate a detection head into CLIP pre-training, using random box sampling and RoI-Align to achieve region-level image-text alignment, while introducing Shifted-Window Learning during the detection fine-tuning phase to enhance ViT's detection performance.
Region-Native Visual Tokenization
Mengyu Wang (Beijing Jiaotong University), Shuicheng Yan (Skywork AI)
RestorationGenerationRepresentation LearningTransformerAuto EncoderImage
🎯 What it does: Propose a region-based visual tokenization method called Reader, constructing an autoencoder to achieve efficient encoding and local editing.
RegionDrag: Fast Region-Based Image Editing with Diffusion Models
Jingyi Lu (University of Hong Kong), Kai Han (University of Hong Kong)
Image TranslationDiffusion modelImage
🎯 What it does: This paper proposes RegionDrag, a fast region drag image editing method based on diffusion models, where users move or deform image content by drawing drag regions (handle region) and target regions (target region).
ReGround: Improving Textual and Spatial Grounding at No Cost
Phillip Y. Lee, Minhyuk Sung
GenerationTransformerDiffusion modelImageMultimodality
🎯 What it does: For the text-and-spatial-prompt-driven image generation model GLIGEN, it was found that there exists a 'description missing' issue in balancing text and spatial constraints, and by re-wiring the network structure (changing from sequential to parallel), the simultaneous accuracy of text and spatial aspects was improved.
Regularizing Dynamic Radiance Fields with Kinematic Fields
Woobin Im (KAIST), Sungeui Yoon (NAVER Cloud)
GenerationNeural Radiance FieldVideoPhysics Related
🎯 What it does: Propose a method that introduces joint learning of kinematic fields (velocity, acceleration, jerk) and dynamic radiance fields, achieving dynamic radiance field reconstruction from monocular videos using physics-driven regularization.
Regulating Model Reliance on Non-Robust Features by Smoothing Input Marginal Density
Peiyu Yang (University of Western Australia), Ajmal Mian (University of Western Australia)
ClassificationExplainability and InterpretabilityAdversarial AttackImage
🎯 What it does: This paper proposes a regularization method that utilizes the gradient of the marginal density of input samples to smooth out, reducing the model's reliance on non-robust features, and distinguishes robust from non-robust features through gradient consistency.
Reinforcement Learning Friendly Vision-Language Model for Minecraft
Haobin Jiang (Peking University), Zongqing Lu (Beijing Academy of Artificial Intelligence)
TransformerReinforcement LearningContrastive LearningVideoTextMultimodality
🎯 What it does: Proposes a cross-modal contrastive learning framework called CLIP4MC, which trains an RL-friendly vision-language model as a task reward function and constructs a high-quality Minecraft YouTube dataset.
Reinforcement Learning Meets Visual Odometry
Nico Messikommer (University of Zurich), Davide Scaramuzza (University of Zurich)
TransformerReinforcement LearningSimultaneous Localization and MappingImageVideo
🎯 What it does: Model visual odometry (VO) as a sequence decision problem, training an agent network using reinforcement learning to dynamically adjust parameters such as keyframe selection and grid size, thereby improving the accuracy and robustness of VO.
Reinforcement Learning via Auxillary Task Distillation
Abhinav N Harish (University of Wisconsin-Madison), Andrew Szot (Georgia Institute of Technology)
Knowledge DistillationRobotic IntelligenceConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningBenchmark
🎯 What it does: Propose a reinforcement learning method for long-term robot control tasks that parallelly learns the main task and more easily learnable auxiliary tasks, and transfers behaviors from auxiliary tasks to the main task using a correlation-weighted distillation loss.
Rejection Sampling IMLE: Designing Priors for Better Few-Shot Image Synthesis
Chirag Vashist (Simon Fraser University), Ke Li (Simon Fraser University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Propose a rejection sampling-based IMLE (RS-IMLE) method for few-shot image synthesis, addressing the inconsistency between potential code distributions in training and inference stages by redesigning the prior distribution.
Relation DETR: Exploring Explicit Position Relation Prior for Object Detection
Xiuquan Hou (Xi'an Jiaotong University), Xuguang Lan (Xi'an Jiaotong University)
Object DetectionTransformerVision Language ModelMultimodality
🎯 What it does: Propose a Transformer-based visual grounding model that uses box features for self-attention modeling to predict the target bounding box corresponding to the text description.
Reliability in Semantic Segmentation: Can We Use Synthetic Data?
Thibaut Loiseau (valeo.ai), Matthieu Cord (Sorbonne Université)
SegmentationData SynthesisAnomaly DetectionDiffusion modelImage
🎯 What it does: Utilizing Stable Diffusion + ControlNet fine-tuned with only in-domain data to generate zero-shot synthetic images and OOD object interpolations for a comprehensive evaluation of pre-trained semantic segmentation models in terms of robustness, OOB detection, confidence calibration, and training.
Reliable and Efficient Concept Erasure of Text-to-Image Diffusion Models
Chao Gong (Fudan University), Yu-Gang Jiang (Fudan University)
GenerationComputational EfficiencyTransformerDiffusion modelImageText
🎯 What it does: Investigated concept erasure in text-to-image diffusion models, proposing a fast and reliable method called RECE;
Reliable Spatial-Temporal Voxels For Multi-Modal Test-Time Adaptation
Haozhi Cao (Nanyang Technological University), Lihua Xie (Nanyang Technological University)
SegmentationDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningContrastive LearningSimultaneous Localization and MappingMultimodality
🎯 What it does: This paper proposes Latte, a 3D semantic segmentation method that achieves multi-modal test-time adaptation (MM-TTA) by leveraging reliable spatiotemporal voxels.
Relightable 3D Gaussians: Realistic Point Cloud Relighting with BRDF Decomposition and Ray Tracing
Jian Gao (Nanjing University), Yao Yao (Nanjing University)
GenerationNeural Radiance FieldGaussian SplattingPoint Cloud
🎯 What it does: Construct a re-lightable 3D Gaussian point cloud by inverse rendering from multi-view images through differentiable point-based rendering, obtaining geometric, material, and lighting information, and achieving photorealistic relighting, shadows, and scene editing.
Relightable Neural Actor with Intrinsic Decomposition and Pose Control
Diogo Carbonera Luvizon (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)
GenerationNeural Radiance FieldVideo
🎯 What it does: This work proposes a neural actor model that can learn from multi-view static illumination videos to generate high-quality images under arbitrary poses and lighting conditions, supporting appearance editing.
ReLoo: Reconstructing Humans Dressed in Loose Garments from Monocular Video in the Wild
Chen Guo (ETH Zürich), Otmar Hilliges (ETH Zürich)
GenerationPose EstimationNeural Radiance FieldVideo
🎯 What it does: Proposed a method to reconstruct high-quality 3D models of humans wearing loose clothing from monocular videos
ReMamber: Referring Image Segmentation with Mamba Twister
Yuhuan Yang (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
SegmentationVision Language ModelMultimodality
🎯 What it does: Proposed the ReMamber model, which achieves efficient image-text fusion in Referring Image Segmentation (RIS) tasks by utilizing the Mamba Twister block.
ReMatching: Low-Resolution Representations for Scalable Shape Correspondence
Filippo Maggioli (University of Milano-Bicocca), Simone Melzi (University of Milano-Bicocca)
Computational EfficiencyRepresentation LearningMesh
🎯 What it does: Propose ReMatching: first perform geometry-preserving low-resolution reconstruction, then conduct functional mapping on the low-resolution model, and finally project the results back to the original high resolution, achieving scalable correspondence for shapes with millions of vertices.
ReMoS: 3D Motion-Conditioned Reaction Synthesis for Two-Person Interactions
Anindita Ghosh (German Research Center for Artificial Intelligence), Philipp Slusallek (German Research Center for Artificial Intelligence)
GenerationData SynthesisTransformerDiffusion modelTime Series
🎯 What it does: Propose ReMoS, a 3D two-person interactive reaction motion synthesis method based on denoising diffusion models, capable of generating the corresponding full-body and hand motions of one person given only the motion of the other person;
Remove Projective LiDAR Depthmap Artifacts via Exploiting Epipolar Geometry
Shengjie Zhu (Michigan State University), Xiaoming Liu (Michigan State University)
Depth EstimationAutonomous DrivingImagePoint Cloud
🎯 What it does: Proposed the RePLAy method, which constructs a stereo system using a virtual LiDAR camera and RGB camera, and detects and removes projection artifacts in LiDAR projected depth maps through epipolar geometry analysis;
Removing Distributional Discrepancies in Captions Improves Image-Text Alignment
Mu Cai (University of Wisconsin-Madison), Krishna Kumar Singh (Adobe Research)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Generate mixed-type negative descriptions using a large language model and eliminate distributional biases through a text classifier to construct high-quality image-text alignment training data; subsequently, fine-tune LLaVA-1.5 and BLIP-2 to obtain LLaVA-score, which measures the semantic alignment between images and text.
Removing Rows and Columns of Tokens in Vision Transformer enables Faster Dense Prediction without Retraining
Diwei Su (East China University of Science and Technology), Jianxu Luo (East China University of Science and Technology)
ClassificationSegmentationComputational EfficiencyTransformerImage
🎯 What it does: Propose Token Adapter, which compresses Vision Transformer tokens by deleting entire rows and columns of the image feature map, achieving acceleration for dense prediction without training.
ReNoise: Real Image Inversion Through Iterative Noising
Daniel Garibi (Tel Aviv University Google Research), Danny Cohen-Or
RestorationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Proposes the ReNoise method, which utilizes iterative denoising techniques to achieve high-quality inversion of real images within diffusion models, thereby enabling text-driven image editing.
Repaint123: Fast and High-quality One Image to 3D Generation with Progressive Controllable Repainting
Junwu Zhang (Peking University), Li Yuan (Peking University)
GenerationTransformerVision Language ModelDiffusion modelScore-based ModelAuto EncoderGaussian SplattingImageMesh
🎯 What it does: Propose Repaint123, a two-stage image-to-3D generation method that first initializes with 3D Gaussian Splatting and then enhances texture quality through progressive controllable repainting, generating high-quality, view-consistent 3D content.
RePOSE: 3D Human Pose Estimation via Spatio-Temporal Depth Relational Consistency
Ziming Sun (South China University of Technology), Shengfeng He (Singapore Management University)
Pose EstimationTransformerVideo
🎯 What it does: Proposes a method called RePOSE, which addresses occlusion problems in video-based 3D human pose estimation by introducing spatial-temporal relative depth consistency supervision.
Representation Enhancement-Stabilization: Reducing Bias-Variance of Domain Generalization
Wei Huang (Technical University of Munich), Xiao Xiang Zhu (Technical University of Munich)
Domain AdaptationRepresentation LearningConvolutional Neural NetworkImageBenchmark
🎯 What it does: Proposes a domain generalization framework RES based on the bias-variance decomposition perspective, which includes a stabilization module with feature frequency domain enhancement and parameter mutual fusion;
Representing Topological Self-Similarity Using Fractal Feature Maps for Accurate Segmentation of Tubular Structures
Jiaxing Huang (Institute of Automation, Chinese Academy of Sciences), Ge Yang (Institute of Automation, Chinese Academy of Sciences)
SegmentationConvolutional Neural NetworkBiomedical Data
🎯 What it does: Accurate segmentation of long, thin tubular structures was achieved by combining pixel-level fractal feature maps (FFM) with a multi-decoder network (MD-Net).
Reprojection Errors as Prompts for Efficient Scene Coordinate Regression
Ting-Ru Liu (National Tsing Hua University), Chun-Yi Lee (National Tsing Hua University)
Pose EstimationPrompt EngineeringSimultaneous Localization and MappingImage
🎯 What it does: This study proposes an error-guided feature selection (EGFS) framework based on reprojection error, aiming to perform scene coordinate regression (SCR) more efficiently for visual localization.
RepVF: A Unified Vector Fields Representation for Multi-task 3D Perception
Jianbing Shen (University of Macau), Sanyuan Zhao (Beijing Institute of Technology)
Autonomous DrivingComputational EfficiencyRepresentation LearningConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: This paper proposes a unified vector field representation, RepVF, to integrate multiple tasks such as 3D object detection and 3D lane detection into a single framework, and constructs the RFTR network to achieve single-head multi-task learning, significantly reducing computational redundancy and task competition;
Reshaping the Online Data Buffering and Organizing Mechanism for Continual Test-Time Adaptation
Zhilin Zhu (Harbin Institute of Technology), Yaowei Wang (Pengcheng Laboratory)
SegmentationDomain AdaptationGraph Neural NetworkImage
🎯 What it does: Designed a source-free continuous test-time adaptation framework based on uncertainty-adaptive buffering and graph structure preservation, which can efficiently collect reliable samples and prevent catastrophic forgetting and error accumulation under online unsupervised environments.
Resilience of Entropy Model in Distributed Neural Networks
Milin Zhang (Northeastern University), Francesco Restuccia (Northeastern University)
CompressionAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper systematically evaluates the robustness of entropy coding models in distributed deep networks and proposes a defense method based on decoupling in frequency and spatial domains.
Resolving Scale Ambiguity in Multi-view 3D Reconstruction using Dual-Pixel Sensors
Kohei Ashida (Osaka University), Yasuyuki Matsushita (Osaka University)
Pose EstimationDepth EstimationSimultaneous Localization and MappingImage
🎯 What it does: Automatically resolve scale ambiguity in structure from motion and multi-view 3D reconstruction by analyzing the size of focal blur from multi-view images captured using dual-pixel (Dual-Pixel, DP) sensors.
Responsible Visual Editing
Minheng Ni (Hong Kong Polytechnic University), Wangmeng Zuo (Harbin Institute of Technology)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringVision-Language-Action ModelDiffusion modelImageChain-of-Thought
🎯 What it does: Proposes the Responsible Visual Editing task, which leverages a multimodal model to automatically identify and edit risky concepts in images, reducing the need for human manual intervention.
Restore Anything with Masks: Leveraging Mask Image Modeling for Blind All-in-One Image Restoration
Chujie Qin, Chongyi Li (Samsung Electronics)
RestorationTransformerSupervised Fine-TuningPrompt EngineeringImage
🎯 What it does: Proposed RAM, a universal blind image restoration framework based on masked image modeling, which includes a two-phase strategy of mask pre-training and fine-tuning only key layers.
Restoring Images in Adverse Weather Conditions via Histogram Transformer
Shangquan Sun (Chinese Academy of Sciences), Xiaochun Cao (Sun Yat-sen University)
RestorationTransformerImage
🎯 What it does: Proposed a Transformer network named Histoformer for unified removal of image degradation caused by adverse weather conditions such as rain, fog, and snow.
ReSyncer: Rewiring Style-based Generator for Unified Audio-Visually Synced Facial Performer
Jiazhi Guan (Tsinghua University), Ziwei Liu (Nanyang Technological University)
Image TranslationGenerationData SynthesisTransformerGenerative Adversarial NetworkVideoMultimodalityMeshAudio
🎯 What it does: Proposed a unified ReSyncer framework that leverages audio-driven 3D facial mesh and StyleGAN generator to achieve high-quality lip-sync video generation, speaking style transfer, personalized fine-tuning, and face replacement.
Retargeting Visual Data with Deformation Fields
Tim Elsner (RWTH Aachen University), Leif Kobbelt (RWTH Aachen University)
Image TranslationRestorationImageMesh
🎯 What it does: Proposed a deformation field-based visual data redirection method, aiming to learn deformations through neural networks to deform in low-information content regions, thereby achieving better content-aware redirection.
Rethinking and Improving Visual Prompt Selection for In-Context Learning Segmentation Framework
Wei Suo (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
SegmentationReinforcement LearningPrompt EngineeringVision Language ModelImage
🎯 What it does: Investigated the impact of visual prompt selection in In-Context Learning (ICL) segmentation, and proposed a Stepwise Context Search (SCS) method based on clustering and reinforcement learning to achieve automated high-quality example selection;
Rethinking Data Augmentation for Robust LiDAR Semantic Segmentation in Adverse Weather
Junsung Park (Korea Advanced Institute of Science & Technology), Hyunjung Shim (Korea Advanced Institute of Science & Technology)
SegmentationAutonomous DrivingData-Centric LearningReinforcement LearningPoint Cloud
🎯 What it does: This paper addresses the problem of performance degradation of LiDAR semantic segmentation under adverse weather conditions, proposing two data augmentation-based methods to enhance robustness;
Rethinking Data Bias: Dataset Copyright Protection via Embedding Class-wise Hidden Bias
Jinhyeok Jang (Electronics And Telecommunications Research Institute), Chan-Hyun Youn (Electronics And Telecommunications Research Institute)
Safty and PrivacyAuto EncoderImage
🎯 What it does: Proposed a dataset watermarking technique called 'undercover bias,' which embeds invisible hidden biases (i.e., watermarks) into each category of the target dataset, enabling trained models to recognize and classify these watermarks, thereby verifying unauthorized data usage in black-box scenarios.
Rethinking Deep Unrolled Model for Accelerated MRI Reconstruction
Bingyu Xin (Rutgers University), Dimitris N. Metaxas (Rutgers University)
RestorationConvolutional Neural NetworkBiomedical DataMagnetic Resonance Imaging
🎯 What it does: To accelerate MRI reconstruction, this paper proposes an adaptive gradient and momentum accelerated deep unfolding model, achieving more efficient and accurate multi-coil reconstruction through multi-slice related low-memory sensitivity map estimation.
Rethinking Directional Parameterization in Neural Implicit Surface Reconstruction
Zijie Jiang (Tokyo Institute of Technology), Hiroharu Kato (Preferred Networks, Inc.)
GenerationNeural Radiance FieldImage
🎯 What it does: Proposed a hybrid direction parameterization method for neural implicit surface reconstruction, capable of effectively handling both specular highlights and complex geometries simultaneously.
Rethinking Fast Adversarial Training: A Splitting Technique To Overcome Catastrophic Overfitting
Masoumeh Zareapoor (East China Normal University), Pourya Shamsolmoali (Queen's University Belfast)
ClassificationAdversarial AttackImage
🎯 What it does: The paper proposes a fast adversarial training framework (RAT) based on the Douglas-Rachford splitting technique, aiming to effectively avoid catastrophic overfitting by stabilizing training dynamics.
Rethinking Features-Fused-Pyramid-Neck for Object Detection
Hulin Li (Chongqing Jiaotong University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Designed and verified an Independent Hierarchical Pyramid (IHP) without feature fusion, proposed Soft Nearest Neighbor Interpolation (SNI) and Extended Spatial Window Adaptive Downsampling (ESD), improved the lightweight GSConv (GSConvE), and integrated these techniques into a secondary feature alignment (SA) scheme for real-time detection.
Rethinking Few-shot Class-incremental Learning: Learning from Yourself
Yu-Ming Tang (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
ClassificationRepresentation LearningMeta LearningTransformerImage
🎯 What it does: This paper proposes a new evaluation metric gAcc and a lightweight feature refinement (FR) module based on Vision Transformer to enhance the performance of few-shot class-incremental learning (FSCIL).
Rethinking Image Super Resolution from Training Data Perspectives
Go Ohtani (Keio University), Yoshimitsu Aoki (Keio University)
Super ResolutionImage
🎯 What it does: Proposes an automated image evaluation pipeline, constructing the DiverSeg dataset, which is low-resolution but high-quality and object-diverse, for training image super-resolution models.
Rethinking Image-to-Video Adaptation: An Object-centric Perspective
Rui Qian (Chinese University of Hong Kong), Dahua Lin (Chinese University of Hong Kong)
RecognitionSegmentationDomain AdaptationComputational EfficiencyRepresentation LearningTransformerContrastive LearningImageVideo
🎯 What it does: Proposes a method for image-to-video adaptation from an object-centric perspective: freezing the pre-trained Vision Transformer, using slot attention with learnable queries to perform unsupervised object decomposition on each frame, and modeling temporal state changes at the object level to achieve efficient video action recognition and zero-shot video object segmentation.
Rethinking LiDAR Domain Generalization: Single Source as Multiple Density Domains
Jaeyeul Kim (DGIST), Sunghoon Im (DGIST)
SegmentationDomain AdaptationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: Developed a Density Discriminative Feature Embedding (DDFE) module, achieving cross-domain semantic segmentation model generalization by mining multi-density distributions in single-source LiDAR point clouds, addressing density discrepancy issues caused by different LiDAR sensors.
Rethinking Normalization Layers for Domain Generalizable Person Re-identification
Ren Nie (University Of Electronic Science And Technology Of China), Xi Li (Zhejiang University)
RecognitionDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: Propose a novel DG-ReID framework named ReNorm, which employs two forward passes respectively adopting Remix Normalization (RN) and Emulation Normalization (EN), while introducing a Domain Frozen (DF) mechanism in both to suppress overfitting of normalization layers on the source domain.
Rethinking Tree-Ring Watermarking for Enhanced Multi-Key Identification
Hai Ci (National University of Singapore), Mike Zheng Shou (National University of Singapore)
RecognitionImage
🎯 What it does: Propose the RingID method, improving the Tree-Ring watermark to achieve multi-key identification and stronger robustness
Rethinking Unsupervised Outlier Detection via Multiple Thresholding
Zhonghang Liu (Singapore Management University), Wen-Yan Lin (Singapore Management University)
Anomaly DetectionImage
🎯 What it does: Propose a multi-threshold (Multi-T) module that automatically generates two thresholds using unlabeled data, classifying samples into clean inliers and outliers, thereby improving the scores and annotation effectiveness of unsupervised anomaly detection.
Rethinking Video Deblurring with Wavelet-Aware Dynamic Transformer and Diffusion Model
Chen Rao (Zhejiang University), Dalong Zhang (Zhejiang University)
RestorationTransformerDiffusion modelVideo
🎯 What it does: Proposed a video deblurring framework named VD-Diff that combines Wavelet-Aware Dynamic Transformer with diffusion models;
Rethinking Video-Text Understanding: Retrieval from Counterfactually Augmented Data
Wufei Ma (Meta Reality Labs), Alan Yuille (Meta AI)
RetrievalRepresentation LearningData-Centric LearningLarge Language ModelVision Language ModelContrastive LearningVideoTextMultimodalityBenchmark
🎯 What it does: This paper proposes a new video-text understanding evaluation task called Adversarial Enhanced Data Retrieval (RCAD) and constructs the Feint6K dataset based on this task, aiming to test whether video-text models truly understand cross-frame action semantics; subsequently, a contrastive learning method named LLM-teacher is designed, leveraging large language models (LLMs) as teachers to enhance the discriminativeness of action embeddings; experiments on multiple mainstream video-text models verify the difficulty of RCAD and the effectiveness of LLM-teacher.
Rethinking Weakly-supervised Video Temporal Grounding From a Game Perspective
Xiang Fang (Huazhong University of Science and Technology), Daizong Liu (Chinese University of Hong Kong)
RetrievalConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningVideoTextMultimodality
🎯 What it does: Propose a weakly supervised video temporal localization framework based on cooperative game theory, treating video frames and query words as players to learn fine-grained frame-word correspondence relationships, achieving precise temporal localization without candidate segments.
Retrieval Robust to Object Motion Blur
Rong Zou (ETH Zürich), Denys Rozumnyi (ETH Zürich)
RetrievalConvolutional Neural NetworkContrastive LearningImageVideo
🎯 What it does: Proposed an image retrieval method for motion-blurred objects that can match blurred and clear images.
Revising Densification in Gaussian Splatting
Samuel Rota Bulò (Meta Reality Labs), Peter Kontschieder (Meta Reality Labs)
OptimizationGaussian SplattingImage
🎯 What it does: Improve the adaptive density control (ADC) in 3D Gaussian Splatting (3DGS) by introducing pixel error-driven density decisions, transparency correction during cloning operations, and control over the total number of primitives and incremental growth rates;
REVISION: Rendering Tools Enable Spatial Fidelity in Vision-Language Models
Agneet Chatterjee (Arizona State University), Chitta R Baral
Image TranslationGenerationData SynthesisVision Language ModelDiffusion modelScore-based ModelImageTextMultimodalityMeshBenchmark
🎯 What it does: Built REVISION, a 3D rendering pipeline based on Blender, capable of precisely synthesizing high-quality images containing over 100 3D assets, 11 types of spatial relationships, diverse backgrounds, and perspectives. These synthetic images are utilized to enhance the spatial consistency of text-image models without training, while introducing the RevQA evaluation framework to assess the spatial reasoning capabilities of multimodal large language models.
Revisit Anything: Visual Place Recognition via Image Segment Retrieval
Kartik Garg (Indian Institute of Science), Sourav Garg (University of Adelaide)
SegmentationRetrievalImage
🎯 What it does: By leveraging open-set segmentation for visual place recognition to obtain SuperSegment, and performing feature aggregation and retrieval at the segment level, the robustness to viewpoint changes is enhanced;