CVPR 2024 Papers — Page 20
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
PREGO: Online Mistake Detection in PRocedural EGOcentric Videos
Alessandro Flaborea (Sapienza University of Rome), Fabio Galasso (Sapienza University of Rome)
ClassificationRecognitionAnomaly DetectionTransformerLarge Language ModelVideo
🎯 What it does: This paper proposes an online one-shot classification model called PREGO, which can instantly detect unseen program errors in first-person videos.
Preserving Fairness Generalization in Deepfake Detection
Li Lin (Purdue University), Shu Hu (Purdue University)
ClassificationDomain AdaptationGenerative Adversarial NetworkContrastive LearningImageVideo
🎯 What it does: A framework is proposed that combines feature disentanglement, dual-layer fairness loss, and loss surface smoothing to achieve fairness generalization in deep fake detection models.
Previously on ... From Recaps to Story Summarization
Aditya Kumar Singh (International Institute of Information Technology Hyderabad), Makarand Tapaswi (International Institute of Information Technology Hyderabad)
GenerationTransformerVideoTextMultimodality
🎯 What it does: This paper proposes a method for generating multimodal story summaries using TV program recaps. It first constructs the PlotSnap dataset, which includes episodes from 24 and Prison Break, and then designs a hierarchical Transformer structure called TaleSumm to encode video shots and dialogues, focusing on story segments through sparse attention, ultimately achieving multimodal summarization.
Privacy-Preserving Face Recognition Using Trainable Feature Subtraction
Yuxi Mi (Fudan University), Shuigeng Zhou (Fudan University)
RecognitionSafty and PrivacyConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A privacy-preserving facial recognition framework named MinusFace is proposed, which obtains the residual by performing feature subtraction between the original facial image and its generated reconstructed image. This residual is then processed through high-dimensional mapping and random channel permutation to generate a final recognizable and privacy-preserving representation.
Privacy-Preserving Optics for Enhancing Protection in Face De-Identification
Jhon Lopez (Universidad Industrial de Santander), Bernard Ghanem (King Abdullah University of Science and Technology)
GenerationSafty and PrivacyGenerative Adversarial NetworkImageVideo
🎯 What it does: A hardware-level facial de-identification scheme is proposed, which jointly trains an optical phase mask and a heatmap regression network, and then uses a reference image-driven GAN to generate new facial images, ensuring the elimination of personal identity information during the acquisition phase.
Probabilistic Sampling of Balanced K-Means using Adiabatic Quantum Computing
Jan-Nico Zaech (ETH Zurich), Luc Van Gool (ETH Zurich)
OptimizationImageTabular
🎯 What it does: The study utilizes adiabatic quantum computing (AQC) for probabilistic sampling of balanced k-means clustering, generating calibrated posterior probabilities from all measurement results to achieve confidence quantification of the clusters.
Probabilistic Speech-Driven 3D Facial Motion Synthesis: New Benchmarks Methods and Applications
Karren D. Yang (Apple), Oncel Tuzel (Apple)
GenerationData SynthesisKnowledge DistillationTransformerAuto EncoderVideoMeshBenchmark
🎯 What it does: A 3D facial action dataset based on the large-scale VoxCeleb2 videos has been constructed, metrics for measuring the quality of probabilistic models have been designed, and a two-stage residual vector quantization (RVQ) autoregressive model has been proposed to achieve speech-driven diverse 3D facial animation generation.
Probing Synergistic High-Order Interaction in Infrared and Visible Image Fusion
Naishan Zheng (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
Image TranslationRestorationTransformerImage
🎯 What it does: A fusion paradigm SHIP based on high-order spatial fine-grained interaction and channel statistical interaction is proposed for the fusion of infrared and visible images, achieving efficient computation of high-order interactions within this framework.
Probing the 3D Awareness of Visual Foundation Models
Mohamed El Banani (University of Michigan), Varun Jampani (Google Research)
Depth EstimationImage
🎯 What it does: This paper evaluates the three-dimensional perception capabilities of frozen features from visual foundation models by probing them for single-view depth, normal vector prediction, and multi-view consistency.
Producing and Leveraging Online Map Uncertainty in Trajectory Prediction
Xunjiang Gu (University of Toronto), Boris Ivanovic (NVIDIA Research)
Autonomous DrivingGraph Neural NetworkTransformerSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper proposes modeling the uncertainty of the position and category of map elements in online HD map estimation, and directly injecting this uncertainty information into the trajectory prediction model to enhance prediction performance.
Programmable Motion Generation for Open-Set Motion Control Tasks
Hanchao Liu (Tsinghua University), Ying Shan (Tencent AI Lab)
OptimizationRobotic IntelligenceLarge Language ModelDiffusion modelMultimodality
🎯 What it does: A programmable motion generation framework is proposed, utilizing a pre-trained motion diffusion model to optimize latent noise so that the generated motion meets any constraints defined by an error function, achieving open-set motion control.
Progress-Aware Online Action Segmentation for Egocentric Procedural Task Videos
Yuhan Shen (Northeastern University), Ehsan Elhamifar (Northeastern University)
SegmentationRecurrent Neural NetworkTransformerVideo
🎯 What it does: A real-time action segmentation framework for first-person process videos, ProTAS, is proposed, which combines causal action segmentation, action progress prediction, and task graph reasoning, significantly improving online segmentation performance.
Progressive Divide-and-Conquer via Subsampling Decomposition for Accelerated MRI
Chong Wang (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
RestorationOptimizationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes an evolutionary partitioning and divide-and-conquer (PDAC) framework, which first decomposes the severe undersampling problem under high compression rates into a series of progressively decreasing moderate undersampling subproblems, and at each iteration stage, only recovers the missing information corresponding to the subspace;
Progressive Semantic-Guided Vision Transformer for Zero-Shot Learning
Shiming Chen (Mohamed bin Zayed University of AI), Fahad Shahbaz Khan (Mohamed bin Zayed University of AI)
ClassificationRecognitionTransformerImage
🎯 What it does: This paper proposes an evolutionary semantic-guided visual Transformer (ZSLViT) that explicitly discovers semantically relevant visual features and eliminates irrelevant information for zero-shot learning by progressively introducing semantic embedding token learning and visual enhancement in the Transformer.
Projecting Trackable Thermal Patterns for Dynamic Computer Vision
Mark Sheinin (Carnegie Mellon University), Srinivasa G. Narasimhan (Carnegie Mellon University)
Object TrackingPose EstimationConvolutional Neural NetworkSimultaneous Localization and MappingOptical FlowImagePoint Cloud
🎯 What it does: By projecting thermal patterns with a laser and capturing them in real-time using a thermal camera, combined with a learning model for inverse thermal diffusion, dynamic visual tasks for texture-less objects are achieved (structured light, optical flow, AR, localization).
ProMark: Proactive Diffusion Watermarking for Causal Attribution
Vishal Asnani (Adobe Research), Shruti Agarwal (Adobe Research)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: ProMark is proposed, an active watermarking technique that embeds invisible watermarks in training images, allowing diffusion models to retain the corresponding watermark when generating images, thus achieving causal attribution between the generated images and the training data concepts.
ProMotion: Prototypes As Motion Learners
Yawen Lu (Purdue University), Heng Fan (University of North Texas)
Object DetectionObject TrackingSegmentationDepth EstimationTransformerOptical FlowImageVideo
🎯 What it does: A unified ProMotion framework is proposed, utilizing prototype learning and feature denoising to simultaneously accomplish optical flow, scene depth estimation, and their transfer to various downstream tasks.
Prompt Augmentation for Self-supervised Text-guided Image Manipulation
Rumeysa Bodur (Imperial College London), Tae-Kyun Kim (KAIST)
Image TranslationGenerationPrompt EngineeringDiffusion modelContrastive LearningImageText
🎯 What it does: This paper proposes a text-guided image editing framework based on prompt enhancement and soft contrast loss, which automatically obtains attention masks through the generation of multiple target prompts, achieving local self-supervised editing.
Prompt Highlighter: Interactive Control for Multi-Modal LLMs
Yuechen Zhang, Jiaya Jia
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringVision Language ModelTextMultimodality
🎯 What it does: A token-level highlighting interactive reasoning method called Prompt Highlighter is proposed, which enables controllable generation of multimodal LLMs without training any models.
Prompt Learning via Meta-Regularization
Jinyoung Park (Korea University), Hyunwoo J. Kim (Korea University)
Meta LearningTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: ProMetaR framework is proposed to automatically adjust regularization and soft prompts through meta-learning to enhance the generalization ability of VLM tasks.
Prompt-Driven Dynamic Object-Centric Learning for Single Domain Generalization
Deng Li (Tianjin University), Yahong Han (Tianjin University)
Object DetectionDomain AdaptationTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: A prompt-based dynamic object-centered learning framework is proposed, which adaptively adjusts the network using object-centered features and scene prompts to enhance single-domain generalization performance.
Prompt-Driven Referring Image Segmentation with Instance Contrasting
Chao Shang (University of Electronic Science and Technology of China), Hongliang Li (University of Electronic Science and Technology of China)
Object DetectionSegmentationTransformerPrompt EngineeringContrastive LearningImageMultimodality
🎯 What it does: Proposes the Prompt-RIS framework, which utilizes cross-modal prompts to bridge CLIP and SAM end-to-end, achieving Referring Image Segmentation (RIS).
Prompt-Enhanced Multiple Instance Learning for Weakly Supervised Video Anomaly Detection
Junxi Chen (University of Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
Anomaly DetectionTransformerPrompt EngineeringVideoMultimodalityAudio
🎯 What it does: This paper proposes a weakly supervised video anomaly detection framework based on prompt-enhanced multi-instance learning (MIL), which utilizes abnormal-aware prompts and normal context prompts to enhance video features, achieving precise anomaly event detection through Transformer-based temporal fusion and multi-scale prediction heads.
Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models
Xingqian Xu (SHI Labs at University of Illinois Urbana-Champaign), Humphrey Shi (Georgia Institute of Technology)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: Proposes the Prompt-Free Diffusion framework, which directly drives the text-to-image diffusion model using reference images and optional structural conditions without the need for text prompts.
Prompt3D: Random Prompt Assisted Weakly-Supervised 3D Object Detection
Xiaohong Zhang (Nanjing University), Jie Guo (Nanjing University)
Object DetectionDomain AdaptationPrompt EngineeringPoint Cloud
🎯 What it does: This paper proposes a weakly supervised 3D object detection method called Prompt3D, which utilizes random text prompts to generate diverse 3D shapes and construct synthetic scenes, achieving domain adaptation through prototype proposal feature alignment.
Promptable Behaviors: Personalizing Multi-Objective Rewards from Human Preferences
Minyoung Hwang (Allen Institute for AI), Kiana Ehsani (Allen Institute for AI)
Robotic IntelligenceReinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningPrompt EngineeringImageTextChain-of-Thought
🎯 What it does: Proposes the Promptable Behaviors framework, which uses a single multi-objective reinforcement learning strategy combined with a reward weight vector to achieve instant personalization of robot behaviors, and predicts weights through human demonstrations, trajectory preference comparisons, or language instructions without the need to retrain the strategy.
PromptAD: Learning Prompts with only Normal Samples for Few-Shot Anomaly Detection
Xiaofan Li (East China Normal University), Lizhuang Ma (East China Normal University)
Anomaly DetectionPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes PromptAD, which utilizes single-class normal samples to automatically learn prompts to enhance few-shot anomaly detection.
PromptCoT: Align Prompt Distribution via Adapted Chain-of-Thought
Junyi Yao (Peking University), Shanghang Zhang (Peking University)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageVideoTextChain-of-Thought
🎯 What it does: An automated prompt enhancer named PromptCoT has been developed, utilizing large language models and the Chain-of-Thought (CoT) mechanism to refine text prompts in a fine-grained manner, significantly improving the generation quality of diffusion models.
Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models
Shweta Mahajan (University of British Columbia), Leonid Sigal (University of British Columbia)
GenerationOptimizationPrompt EngineeringDiffusion modelImage
🎯 What it does: For the target image, a discrete optimization method PH2P is proposed, which directly recovers readable hard prompts through delayed projection and gradients at later noisy time steps.
Prompting Vision Foundation Models for Pathology Image Analysis
Chong Yin (Hong Kong Baptist University), Pong C. Yuen (Hong Kong Baptist University)
ClassificationRecognitionAnomaly DetectionTransformerPrompt EngineeringImageBiomedical Data
🎯 What it does: A quantitative attribute-based prompting method (QAP) is proposed for liver pathological image analysis, transforming spatial (K-function) and morphological (histogram) attributes into visual prompts to enhance model performance.
PromptKD: Unsupervised Prompt Distillation for Vision-Language Models
Zheng Li (Nankai University), Jian Yang (Nankai University)
ClassificationKnowledge DistillationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes an unsupervised Prompt Distillation framework called PromptKD, which utilizes a large teacher CLIP model to train a lightweight student model on unlabeled domain images.
ProS: Prompting-to-simulate Generalized knowledge for Universal Cross-Domain Retrieval
Kaipeng Fang (University of Electronic Science and Technology of China), Heng Tao Shen (Tongji University)
RetrievalDomain AdaptationPrompt EngineeringContrastive LearningImage
🎯 What it does: This paper proposes a two-stage Prompt-to-Simulate method (ProS) based on CLIP, which learns domain and semantic Prompt units and utilizes a context-aware Prompt simulator to generate dynamic Prompts for universal cross-domain retrieval (UCDR).
ProTeCt: Prompt Tuning for Taxonomic Open Set Classification
Tz-Ying Wu (University of California), Nuno Vasconcelos (University of California)
ClassificationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: A taxonomic open set (TOS) classification method called ProTeCt is proposed for visual-language foundational models, which maintains prediction consistency across different levels of label sets.
ProxyCap: Real-time Monocular Full-body Capture in World Space via Human-Centric Proxy-to-Motion Learning
Yuxiang Zhang (Tsinghua University), Yebin Liu (Harbin Institute of Technology)
Data SynthesisPose EstimationSimultaneous Localization and MappingVideo
🎯 What it does: Developed ProxyCap, a real-time monocular full-body motion capture method that utilizes 2D skeletal proxies and human-centered learning to achieve world-space pose recovery, incorporating a contact-based neural descent module.
PrPSeg: Universal Proposition Learning for Panoramic Renal Pathology Segmentation
Ruining Deng (Vanderbilt University), Yuankai Huo (Vanderbilt University)
SegmentationConvolutional Neural NetworkBiomedical Data
🎯 What it does: A scalable panoramic nephropathy tissue segmentation model, PrPSeg, is proposed and implemented, capable of achieving multi-scale and multi-class segmentation from the tissue level to the cellular level.
PSDPM: Prototype-based Secondary Discriminative Pixels Mining for Weakly Supervised Semantic Segmentation
Xinqiao Zhao (University of Liverpool), Jimin Xiao (University of Liverpool)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: The PSDPM (Prototype-based Secondary Discriminative Pixels Mining) framework is proposed, utilizing the Foreground Pixel Estimation Module (FPEM) and consistency loss to encourage the weakly supervised semantic segmentation model to learn primary discriminative pixels and secondary discriminative pixels, thereby generating a more complete Class Activation Map (CAM).
Pseudo Label Refinery for Unsupervised Domain Adaptation on Cross-dataset 3D Object Detection
Zhanwei Zhang (Fabu Inc.), Deng Cai (Zhejiang University)
Object DetectionDomain AdaptationAutonomous DrivingPoint Cloud
🎯 What it does: A pseudo-label refinement framework PERE is proposed for unsupervised 3D object detection across datasets.
Psychometry: An Omnifit Model for Image Reconstruction from Human Brain Activity
Ruijie Quan (Zhejiang University), Yi Yang (Zhejiang University)
GenerationRetrievalTransformerMixture of ExpertsDiffusion modelContrastive LearningImageMagnetic Resonance Imaging
🎯 What it does: A unified omnifit model called Psychometry was constructed to reconstruct natural images from fMRI data of multiple subjects using a single network.
PTM-VQA: Efficient Video Quality Assessment Leveraging Diverse PreTrained Models from the Wild
Kun Yuan (Kuaishou Technology), Yansong Tang (Tsinghua University)
OptimizationComputational EfficiencyVideo
🎯 What it does: PTM-VQA is proposed, which extracts and fuses features by freezing various pre-trained models for no-reference video quality assessment;
PTQ4SAM: Post-Training Quantization for Segment Anything
Chengtao Lv (Beihang University), Xianglong Liu (Beihang University)
Object DetectionSegmentationTransformerSupervised Fine-TuningImage
🎯 What it does: A post-training quantization framework PTQ4SAM is proposed to specifically address the bimodal activation distribution and diversified softmax distribution issues in the Segment Anything Model (SAM).
PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection
Kuan-Chih Huang (University of California), Yi-Hsuan Tsai (Google)
Object DetectionAutonomous DrivingComputational EfficiencyTransformerPoint Cloud
🎯 What it does: A Point-Trajectory Transformer (PTT) is proposed, which utilizes only the current frame point cloud and multi-frame predicted box trajectories for temporal 3D object detection, significantly reducing memory usage.
Puff-Net: Efficient Style Transfer with Pure Content and Style Feature Fusion Network
Sizhe Zheng (Nanjing University of Aeronautics and Astronautics), Jie Qin (Nanjing University of Aeronautics and Astronautics)
Image TranslationComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposes Puff-Net, an efficient style transfer framework that combines pure content and pure style feature extractors with a transformer that only contains an encoder;
Purified and Unified Steganographic Network
Guobiao Li (Fudan University), Xinpeng Zhang (Fudan University)
RestorationData SynthesisConvolutional Neural NetworkImage
🎯 What it does: A Purified and Unified Steganographic Network (PUSNet) is proposed, which can embed and recover secrets within a sparse ordinary network and is activated by a key.
Putting the Object Back into Video Object Segmentation
Ho Kei Cheng (University of Illinois Urbana-Champaign), Alexander Schwing (University of Illinois Urbana-Champaign)
Object DetectionSegmentationTransformerVideo
🎯 What it does: Introducing Cutie, a VOS network that combines pixel-level memory reading and object querying, utilizing object-level memory re-reading to enhance segmentation robustness.
Q-Instruct: Improving Low-level Visual Abilities for Multi-modality Foundation Models
Haoning Wu, Weisi Lin
RecognitionData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Collected 58K human low-level visual feedback and generated 200K instruction-response pairs, constructing the Q-Pathway and Q-Instruct datasets for low-level visual instruction tuning of multimodal large models.
QDFormer: Towards Robust Audiovisual Segmentation in Complex Environments with Quantization-based Semantic Decomposition
Xiang Li (Carnegie Mellon University), Bhiksha Raj (Microsoft Research Asia)
SegmentationTransformerVideoMultimodalityAudio
🎯 What it does: This paper proposes a quantization-based semantic decomposition method for achieving more robust audio-video segmentation (AVS) and semantic segmentation (AVSS) in multi-source noisy environments.
QN-Mixer: A Quasi-Newton MLP-Mixer Model for Sparse-View CT Reconstruction
Ishak Ayad (Cergy Paris University), Mai K. Nguyen (Cergy Paris University)
RestorationOptimizationImageBiomedical DataComputed Tomography
🎯 What it does: A second-order iterative network QN-Mixer based on the quasi-Newton method is proposed for sparse-angle CT reconstruction.
QUADify: Extracting Meshes with Pixel-level Details and Materials from Images
Maximilian Frühauf (ETH Zurich), Christopher Schroers (Disney Research Studios)
GenerationData SynthesisNeural Radiance FieldImageMesh
🎯 What it does: This paper proposes an end-to-end 3D reconstruction framework that can directly extract quadrilateral dominant meshes with pixel-level displacement, material, and lighting information from multi-view images.
Quantifying Task Priority for Multi-Task Optimization
Wooseong Jeong (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
SegmentationOptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a multi-task learning optimization method based on connection strength, which can learn and maintain task priorities within shared parameters, thereby reducing negative transfer between multiple tasks and improving overall performance.
Quantifying Uncertainty in Motion Prediction with Variational Bayesian Mixture
Juanwu Lu (Purdue University), Ziran Wang (Purdue University)
GenerationAutonomous DrivingExplainability and InterpretabilityRecurrent Neural NetworkAuto EncoderTime SeriesSequential
🎯 What it does: A variational Bayesian mixture model called SeNeVA is proposed for generative prediction of future trajectories of individual traffic participants and quantification of prediction uncertainty.
Querying as Prompt: Parameter-Efficient Learning for Multimodal Language Model
Tian Liang (Zhejiang University), Qiang Zhu (Beijing Information Science and Technology University)
OptimizationTransformerLarge Language ModelPrompt EngineeringVideoTextMultimodality
🎯 What it does: A parameter-efficient multimodal language model learning strategy called QaP is proposed, which injects multimodal information into a frozen pre-trained language model through queries as prompts.
Question Aware Vision Transformer for Multimodal Reasoning
Roy Ganz (Technion), Ron Litman (AWS AI Labs)
RecognitionData-Centric LearningTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Introducing a Question-Answering Perceptive Visual Encoder (QA-ViT) in visual-language (VL) models, which achieves question-conditioned encoding of images by embedding questions into the self-attention layers of ViT; this method can be inserted into any VL architecture while keeping the visual encoder frozen.
Quilt-LLaVA: Visual Instruction Tuning by Extracting Localized Narratives from Open-Source Histopathology Videos
Mehmet Saygin Seyfioglu (University of Washington), Linda Shapiro (University of Washington)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageVideoTextMultimodality
🎯 What it does: This study constructed a multimodal instruction tuning dataset QLUT-INSTRUCT based on educational pathology videos and trained a multimodal model QLUT-LLAVA with cross-image reasoning capabilities using it;
R-Cyclic Diffuser: Reductive and Cyclic Latent Diffusion for 3D Clothed Human Digitalization
Kennard Yanting Chan (Nanyang Technological University), Weisi Lin (Nanyang Technological University)
GenerationPose EstimationDiffusion modelImage
🎯 What it does: A R-Cyclic Diffuser has been developed, which achieves 3D clothed human reconstruction from a single view by combining the Zero-1-to-3 latent diffusion method with a pixel-aligned implicit model.
RadarDistill: Boosting Radar-based Object Detection Performance via Knowledge Distillation from LiDAR Features
Geonho Bang (Hanyang University), Jun Won Choi (Seoul National University)
Object DetectionAutonomous DrivingKnowledge DistillationConvolutional Neural NetworkPoint Cloud
🎯 What it does: Utilizing knowledge distillation to transfer features from LiDAR models to radar models, enhancing the 3D object detection performance of radar point clouds.
RadSimReal: Bridging the Gap Between Synthetic and Real Data in Radar Object Detection With Simulation
Oded Bialer (General Motors), Yuval Haitman (General Motors)
Object DetectionData SynthesisImage
🎯 What it does: A physical radar simulation framework named RadSimReal is proposed, which can generate annotated radar images for training object detection networks without requiring detailed radar hardware information.
RAM-Avatar: Real-time Photo-Realistic Avatar from Monocular Videos with Full-body Control
Xiang Deng (Tsinghua University), Yebin Liu (Tsinghua University)
GenerationPose EstimationGenerative Adversarial NetworkVideo
🎯 What it does: We propose RAM-Avatar, a real-time and realistic full-body control human avatar learning method based on monocular video;
Random Entangled Tokens for Adversarially Robust Vision Transformer
Huihui Gong (University of Sydney), Chang Xu (University of Sydney)
RecognitionAdversarial AttackTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes a Vision Transformer scheme called ReiT that combines randomization and adversarial training, utilizing input-independent random entangled self-attention (II-ReSA) to enhance adversarial robustness.
RankED: Addressing Imbalance and Uncertainty in Edge Detection Using Ranking-based Losses
Bedrettin Cetinkaya (Middle East Technical University), Emre Akbas (Middle East Technical University)
Object DetectionSegmentationTransformerImage
🎯 What it does: A new edge detection method called RankED is proposed, which addresses the issues of positive-negative sample imbalance and label uncertainty using a ranking-based loss.
Ranking Distillation for Open-Ended Video Question Answering with Insufficient Labels
Tianming Liang (Sun Yat-sen University), Jian-Fang Hu (Sun Yat-sen University)
RetrievalKnowledge DistillationVideo
🎯 What it does: This paper addresses the generalization challenge caused by label scarcity in Open-ended Video Question Answering (OE-VQA) by proposing a ranking distillation framework called RADI, which enriches training information by generating answer rankings through a teacher model trained on incomplete labels.
RankMatch: Exploring the Better Consistency Regularization for Semi-supervised Semantic Segmentation
Huayu Mai (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
SegmentationKnowledge DistillationImage
🎯 What it does: Proposes the RankMatch method, which constructs pixel-wise correlations by introducing proxy points in semi-supervised semantic segmentation, and enhances model generalization using ranking-aware consistency regularization;
Ranni: Taming Text-to-Image Diffusion for Accurate Instruction Following
Yutong Feng (Alibaba Group), Jingren Zhou (Alibaba Group)
Image TranslationGenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: By constructing a semantic panel, text instructions are transformed into structured object information, which is then mapped to images using a diffusion model, achieving precise instruction following and fine-grained editing.
Rapid 3D Model Generation with Intuitive 3D Input
Tianrun Chen (KOKONI3D), Lingyun Sun (Zhejiang University)
GenerationData SynthesisDiffusion modelPoint CloudMesh
🎯 What it does: Deep3DVRSketch is proposed, allowing users to hand-draw 3D sketches in VR and generate high-quality 3D models within seconds.
Rapid Motor Adaptation for Robotic Manipulator Arms
Yichao Liang (University of Cambridge), João Henriques (University of Oxford)
Robotic IntelligenceConvolutional Neural NetworkReinforcement LearningMultimodalityBenchmark
🎯 What it does: A rapid adaptation method for robotic arms based on RMA is proposed, enabling good generalization across various manipulation tasks.
RAVE: Randomized Noise Shuffling for Fast and Consistent Video Editing with Diffusion Models
Ozgur Kara, Pinar Yanardag
GenerationData SynthesisComputational EfficiencyDiffusion modelVideoMultimodality
🎯 What it does: A zero-shot video editing framework called RAVE has been developed, utilizing a pre-trained text-to-image diffusion model and a random noise scrambling strategy to achieve fast video editing while maintaining spatiotemporal consistency.
RCBEVDet: Radar-camera Fusion in Bird's Eye View for 3D Object Detection
Zhiwei Lin (Peking University), Ce Zhu (University of Electronic Science and Technology of China)
Object DetectionAutonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: A radar-camera multimodal 3D object detection framework RCBEVDet is proposed, which integrates radar and multi-view camera information in the bird's-eye view (BEV) space to enhance detection accuracy and robustness.
RCL: Reliable Continual Learning for Unified Failure Detection
Fei Zhu (Centre for Artificial Intelligence and Robotics), Zhaoxiang Zhang (Centre for Artificial Intelligence and Robotics)
Anomaly DetectionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A reliable continuous learning framework (RCL) is proposed, which first allows the model to learn misclassification detection, and then fine-tunes it to achieve unified failure detection (misclassification + OOD) while retaining the original knowledge.
RCooper: A Real-world Large-scale Dataset for Roadside Cooperative Perception
Ruiyang Hao (Tsinghua University), Zaiqing Nie (Tsinghua University)
Object DetectionObject TrackingAutonomous DrivingSimultaneous Localization and MappingImageMultimodalityPoint Cloud
🎯 What it does: RCooper has been released, which is the first large-scale, real-world roadside collaborative perception dataset that supports detection and tracking tasks.
Re-thinking Data Availability Attacks Against Deep Neural Networks
Bin Fang (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
OptimizationAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a two-stage min-max-min optimization framework that uses a robust adversarial training noise generation model to generate noise without learning, effectively resisting the damage caused by adversarial training.
REACTO: Reconstructing Articulated Objects from a Single Video
Chaoyue Song (Nanyang Technological University), Fayao Liu (Institute for Inforcomm Research A*STAR)
GenerationPose EstimationNeural Radiance FieldVideo
🎯 What it does: This paper presents REACTO, which reconstructs 3D models of general movable parts from single-segment videos, taking into account shape, texture, and motion.
READ: Retrieval-Enhanced Asymmetric Diffusion for Motion Planning
Takeru Oba (Toyota Technological Institute), Norimichi Ukita (Toyota Technological Institute)
Robotic IntelligenceTransformerReinforcement LearningDiffusion modelImageRetrieval-Augmented GenerationStochastic Differential Equation
🎯 What it does: A retrieval-based image-driven robot motion planning framework called READ (Retrieval-Enhanced Asymmetric Diffusion) is proposed, which utilizes retrieval to obtain initial motion and refines it through asymmetric diffusion.
Readout Guidance: Learning Control from Diffusion Features
Grace Luo (Google Research), Aleksander Holynski (Google Research)
Pose EstimationDepth EstimationDiffusion modelContrastive LearningImageVideo
🎯 What it does: A Readout Guidance framework is designed, which extracts image attributes by training a lightweight readout head on a frozen diffusion model and guides during sampling to achieve various control tasks such as pose, depth, correspondence, appearance similarity, and identity preservation.
Real Acoustic Fields: An Audio-Visual Room Acoustics Dataset and Benchmark
Ziyang Chen (University of Michigan), Alexander Richard (Codec Avatars Lab, Pittsburgh, Meta)
Neural Radiance FieldMultimodalityBenchmarkAudio
🎯 What it does: The Real Acoustic Fields (RAF) dataset is proposed, which contains dense multimodal indoor acoustic recordings and high-quality visual data.
Real-IAD: A Real-World Multi-View Dataset for Benchmarking Versatile Industrial Anomaly Detection
Chengjie Wang (Youtu Lab Tencent), Lizhuang Ma (Shanghai Jiao Tong University)
Anomaly DetectionImageBenchmark
🎯 What it does: This paper presents the Real-IAD industrial defect detection dataset, which consists of 150K samples, 30 categories, and 5 perspectives, and introduces a completely unsupervised FUIAD setting for the first time.
Real-time 3D-aware Portrait Video Relighting
Ziqi Cai (Beijing Jiaotong University), Lin Gao (Institute of Computing Technology Chinese Academy of Sciences)
GenerationData SynthesisConvolutional Neural NetworkTransformerNeural Radiance FieldGenerative Adversarial NetworkVideo
🎯 What it does: A real-time 3D perception portrait video relighting method is proposed, capable of generating high-quality, temporally consistent facial videos under arbitrary viewpoints and lighting conditions.
Real-time Acquisition and Reconstruction of Dynamic Volumes with Neural Structured Illumination
Yixin Zeng (Zhejiang University), Hongzhi Wu (Zhejiang University)
Data SynthesisOptimizationGenerative Adversarial NetworkVideo
🎯 What it does: A framework for joint optimization of lighting and depth networks is proposed, capable of capturing and reconstructing dynamic 3D volumes in real-time.
Real-Time Exposure Correction via Collaborative Transformations and Adaptive Sampling
Ziwen Li (Huazhong University of Science and Technology), Nong Sang (Huazhong University of Science and Technology)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A collaborative transformation framework CoTF is proposed to achieve real-time exposure correction by combining global 3D LUT with local pixel-level transformations.
Real-Time Neural BRDF with Spherically Distributed Primitives
Yishun Dou (Shanghai Jiao Tong University), Junxiang Ke (Huawei)
Mesh
🎯 What it does: This paper presents NeuBRDF, a lightweight neural reflection model that decomposes 4D BRDF into two low-dimensional hemispherical feature grids, and achieves real-time rendering evaluation at millisecond speeds through quantifiable neural reflection primitives (codebook) and polar coordinate grids (HEALPix).
Real-Time Simulated Avatar from Head-Mounted Sensors
Zhengyi Luo (Reality Labs Research), Weipeng Xu (Reality Labs Research)
Pose EstimationKnowledge DistillationRobotic IntelligenceConvolutional Neural NetworkReinforcement LearningImageVideo
🎯 What it does: Developed the SimXR framework, which utilizes multi-view camera images from XR head-mounted devices and head pose to control a simulated humanoid avatar following user movements in a real-time physics simulation environment.
Real-World Efficient Blind Motion Deblurring via Blur Pixel Discretization
Insoo Kim (Samsung Advanced Institute of Technology), Hyong-Euk Lee (Samsung Advanced Institute of Technology)
RestorationComputational EfficiencyTransformerImage
🎯 What it does: An efficient framework is proposed that decomposes the blind motion deblurring task into the discretization of blurry pixels and the discrete-to-continuous transformation, significantly reducing computational load.
Real-World Mobile Image Denoising Dataset with Efficient Baselines
Roman Flepp (ETH Zurich), Luc Van Gool (ETH Zurich)
RestorationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper constructs a large-scale multi-sensor mobile image denoising dataset MIDD and proposes an efficient segmentation network SplitterNet, which supports processing 8MP images on mobile GPUs/NPUs in less than 1 second.
RealCustom: Narrowing Real Text Word for Real-Time Open-Domain Text-to-Image Customization
Mengqi Huang (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
GenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: The paper proposes the RealCustom method, which enables real-time open-domain text-image customization for a single image.
Realigning Confidence with Temporal Saliency Information for Point-Level Weakly-Supervised Temporal Action Localization
Ziying Xia (University of Electronic Science and Technology of China), Liwan Dang (Civil Aviation Administration of China)
RecognitionObject DetectionConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: A pluggable proposal learning framework (TSP-Net) is proposed, which aligns confidence and proposal quality in point-level weakly supervised temporal action localization through center score learning and alignment boundary adaptation.
RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection
Ximiao Zhang (Capital Normal University), Xiuzhuang Zhou (Beijing University of Posts and Telecommunications)
Anomaly DetectionConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A feature reconstruction-based unsupervised defect detection framework, RealNet, is proposed, which achieves high-precision defect detection and localization using only normal images.
RecDiffusion: Rectangling for Image Stitching with Diffusion Models
Tianhao Zhou (University of Electronic Science and Technology of China), Shuaicheng Liu (University of Electronic Science and Technology of China)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes RecDiffusion, which utilizes motion diffusion models and content diffusion models for the rectification of stitched images.
ReconFusion: 3D Reconstruction with Diffusion Priors
Rundi Wu (Google Research), Aleksander Ho?y?ski
Diffusion modelNeural Radiance FieldImage
🎯 What it does: Using a diffusion model prior to assist NeRF in 3D reconstruction under limited view conditions.
Reconstructing Hands in 3D with Transformers
Georgios Pavlakos (University of California Berkeley), Jitendra Malik (University of California Berkeley)
GenerationPose EstimationTransformerMesh
🎯 What it does: This paper proposes a monocular hand 3D mesh reconstruction method based on Transformer, named HaMeR, which can generate accurate hand 3D meshes in complex scenarios such as multiple viewpoints, occlusions, and object interactions.
Reconstruction-free Cascaded Adaptive Compressive Sensing
Chenxi Qiu (Nanjing University), Xuemei Hu (Nanjing University)
RestorationCompressionConvolutional Neural NetworkScore-based ModelImage
🎯 What it does: A reconstruction-free cascade adaptive compressed sensing (ACS) framework is proposed, which utilizes a lightweight ScoreNet to directly allocate sampling rates based on the previous measurement, and completes adaptive sampling through a differentiable linear programming sampling module, ultimately performing a single image reconstruction only after all samples have been collected.
ReCoRe: Regularized Contrastive Representation Learning of World Model
Rudra P.K. Poudel (Toshiba Europe Limited), Roberto Cipolla (University of Cambridge)
Representation LearningRobotic IntelligenceReinforcement Learning from Human FeedbackContrastive LearningWorld ModelImage
🎯 What it does: This paper proposes the use of world models combined with contrastive learning in everyday tasks such as visual navigation to enhance sample efficiency and robustness to the environment.
Referring Expression Counting
Siyang Dai (Singapore University of Technology and Design), Ngai-Man Cheung (Singapore University of Technology and Design)
Object DetectionTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: The Referring Expression Counting (REC) task is proposed, and the REC-8K dataset is constructed, which includes 8,011 images and 17,122 descriptions, using the GroundingREC model for training and evaluation.
Referring Image Editing: Object-level Image Editing via Referring Expressions
Chang Liu (Fudan University), Henghui Ding (Institute for Infocomm Research)
Image TranslationSegmentationGenerationTransformerDiffusion modelImage
🎯 What it does: This paper proposes the Referring Image Editing (RIE) task, which involves editing specified objects in an image using only referring expressions, and introduces the ReferDiffusion method based on diffusion models to accomplish this task.
Reg-PTQ: Regression-specialized Post-training Quantization for Fully Quantized Object Detector
Yifu Ding (Beihang University), Xianglong Liu (Beihang University)
Object DetectionImage
🎯 What it does: A full post-training quantization framework Reg-PTQ for object detection regression tasks is proposed.
ReGenNet: Towards Human Action-Reaction Synthesis
Liang Xu (Shanghai Jiao Tong University), Wenjun Zeng (Ningbo Institute of Digital Twin)
GenerationData SynthesisPose EstimationTransformerDiffusion modelVideoMultimodalityBenchmark
🎯 What it does: This study investigates how to generate corresponding human reaction actions in real-time based on a given sequence of human actions, thereby achieving human action-reaction synthesis.
Region-Based Representations Revisited
Michal Shlapentokh-Rothman (University of Illinois), Derek Hoiem (University of Illinois)
SegmentationRetrievalTransformerContrastive LearningImageVideo
🎯 What it does: This paper proposes to average pool the semantic regions generated by SAM (and SLIC) with strong self-supervised features like DINOv2 to obtain a sparse yet semantically rich 'region representation', and uses simple linear/MLP/Transformer decoders to accomplish tasks such as semantic segmentation, object retrieval, multi-view segmentation, and video action classification.
RegionGPT: Towards Region Understanding Vision Language Model
Qiushan Guo (University of Hong Kong), Sifei Liu (NVIDIA)
ClassificationObject DetectionRetrievalTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: A general visual language model named RGPT has been constructed, supporting complex descriptions, reasoning, classification, and pointing tasks for arbitrary shaped regions, and performance is enhanced through region-level instruction fine-tuning.
RegionPLC: Regional Point-Language Contrastive Learning for Open-World 3D Scene Understanding
Jihan Yang (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
Object DetectionSegmentationVision Language ModelContrastive LearningPoint Cloud
🎯 What it does: This paper proposes RegionPLC, which utilizes a 2D visual language model to generate dense region-level 3D-language pairs and achieves open-world 3D scene understanding through region-aware point-level contrastive learning.
Regressor-Segmenter Mutual Prompt Learning for Crowd Counting
Mingyue Guo (University of Chinese Academy of Sciences), Qixiang Ye (University of Chinese Academy of Sciences)
Object DetectionSegmentationConvolutional Neural NetworkPrompt EngineeringImage
🎯 What it does: A mutual prompting learning framework is proposed, combining density regressors with head segmenters, utilizing point prompts and context prompts to correct label variance and improve crowd counting accuracy.
Regularized Parameter Uncertainty for Improving Generalization in Reinforcement Learning
Pehuen Moure (Institute of Neuroinformatics ETH Zurich and University of Zurich), Shih-Chii Liu (Institute of Neuroinformatics ETH Zurich and University of Zurich)
Reinforcement LearningSequential
🎯 What it does: This paper proposes a parameter uncertainty network based on signal-to-noise ratio (SNR) (SNR PUN), which enhances the model's generalization ability in unseen environments by regularizing the ratio of the mean to the standard deviation of each parameter in reinforcement learning.
Relation Rectification in Diffusion Model
Yinwei Wu (National University of Singapore), Xinchao Wang (National University of Singapore)
GenerationGraph Neural NetworkDiffusion modelImageBenchmark
🎯 What it does: The RRNet performs relationship correction on the text embeddings of Stable Diffusion, enabling the model to accurately generate images of specified object relationships in sentences.
Relational Matching for Weakly Semi-Supervised Oriented Object Detection
Wenhao Wu (City University of Hong Kong), Tianyou Zhang (South China University of Technology)
Object DetectionKnowledge DistillationImage
🎯 What it does: A weakly semi-supervised directional object detection framework is proposed, utilizing point-level annotations and full box annotations to enhance detection performance.