CVPR 2023 Papers — Page 16
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
OmniVidar: Omnidirectional Depth Estimation From Multi-Fisheye Images
Sheng Xie (Harbin Institute of Technology), Yun-Hui Liu
Depth EstimationTransformerOptical FlowImage
🎯 What it does: The OmniVidar system is proposed, utilizing four 250° FOV fisheye cameras to achieve omnidirectional depth estimation, transforming multi-fisheye imaging into four stereo systems for depth inference.
On Calibrating Semantic Segmentation Models: Analyses and an Algorithm
Dongdong Wang, Liqiang Wang
SegmentationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new method to address a specific computer vision problem.
On Data Scaling in Masked Image Modeling
Zhenda Xie (Microsoft Research Asia), Han Hu (Microsoft Research Asia)
Object DetectionSegmentationTransformerAuto EncoderImage
🎯 What it does: The system studied the scaling behavior of Masked Image Modeling (MIM) under different model sizes, data scales, and training lengths.
On Distillation of Guided Diffusion Models
Chenlin Meng (Stanford University), Tim Salimans (Google Research)
GenerationComputational EfficiencyKnowledge DistillationDiffusion modelImage
🎯 What it does: This paper proposes a two-stage distillation method that compresses a diffusion model guided by classifier-free guidance into a model capable of generating high-quality images in 1-4 steps, significantly improving sampling efficiency.
On the Benefits of 3D Pose and Tracking for Human Action Recognition
Jathushan Rajasegaran (University of California Berkeley), Jitendra Malik (University of California Berkeley)
RecognitionObject TrackingPose EstimationTransformerVideo
🎯 What it does: This paper proposes a Lagrangian perspective-based action recognition framework called LART, which utilizes 3D pose tracking information and appearance features to model human trajectories in videos for action detection.
On the Convergence of IRLS and Its Variants in Outlier-Robust Estimation
Liangzu Peng (Johns Hopkins University), René Vidal (University of Pennsylvania)
OptimizationPoint Cloud
🎯 What it does: A robust estimation algorithm based on IRLS and graduated non-convexity (GNC) called GNC-IRLS is proposed, and an improved MS-GNC-TLS method is presented based on it.
On the Difficulty of Unpaired Infrared-to-Visible Video Translation: Fine-Grained Content-Rich Patches Transfer
Zhenjie Yu (Beijing Institute of Technology), Shuigen Wang (Yantai IRay Technologies)
Image TranslationGenerationGenerative Adversarial NetworkContrastive LearningVideo
🎯 What it does: For unpaired infrared-visible video translation, the CPTrans framework is proposed to achieve fine-grained content-rich patch transfer.
On the Effectiveness of Partial Variance Reduction in Federated Learning With Heterogeneous Data
Bo Li (Technical University of Denmark), Sebastian U. Stich (CISPA Helmholtz Center for Information Security)
OptimizationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes FedPVR, an algorithm for partially reducing variance in the last few layers of deep neural networks within federated learning, improving traditional FedAvg to alleviate client drift caused by data heterogeneity.
On the Effects of Self-Supervision and Contrastive Alignment in Deep Multi-View Clustering
Daniel J. Trosten (UiT The Arctic University of Norway), Michael C. Kampffmeyer
Representation LearningContrastive LearningImage
🎯 What it does: This paper proposes the DeepMVC framework to unify deep multi-view clustering methods and analyzes the impact of self-supervised learning (especially contrastive alignment) on clustering performance, pointing out that alignment reduces the number of distinguishable clusters as the number of views increases.
On the Importance of Accurate Geometry Data for Dense 3D Vision Tasks
HyunJun Jung (Technical University of Munich), Benjamin Busam (3Dwe.ai)
Depth EstimationRobotic IntelligenceNeural Radiance FieldMultimodalityMesh
🎯 What it does: A multi-modal high-precision dataset is proposed, and based on this, the impact of noise from different depth sensors (I-ToF, D-ToF, active stereo, RGB+P, etc.) on dense 3D vision tasks (monocular depth estimation, implicit reconstruction, etc.) is evaluated.
On the Pitfall of Mixup for Uncertainty Calibration
Deng-Bao Wang (Southeast University), Min-Ling Zhang (Southeast University)
ClassificationData SynthesisConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper systematically studies the impact of mixup training on uncertainty calibration, pointing out that it makes the model harder to calibrate and provides improvement methods.
On the Stability-Plasticity Dilemma of Class-Incremental Learning
Dongwan Kim (Seoul National University), Bohyung Han (Seoul National University)
ClassificationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: This paper studies the stability-plasticity dilemma in class-incremental learning by analyzing the changes in feature representations to evaluate the balance effects of various algorithms. Based on the analysis results, two improvement methods are proposed: pDER and Exploit.
On-the-Fly Category Discovery
Ruoyi Du (Beijing University of Posts and Telecommunications), Zhanyu Ma (Beijing University of Posts and Telecommunications)
ClassificationRecognitionContrastive LearningImage
🎯 What it does: In the open, online visual classification scenario, an Instant Category Discovery (OCD) task is proposed, utilizing a scalable recognition model with hash coding to achieve instant recognition of new and old categories.
One-Shot High-Fidelity Talking-Head Synthesis With Deformable Neural Radiance Field
Weichuang Li (Shanghai AI Laboratory), Xuelong Li (Northwestern Polytechnical University)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkVideo
🎯 What it does: We propose HiDe-NeRF, a single-image high-fidelity, free-viewpoint speaker avatar synthesis framework based on Deformable NeRF.
One-Shot Model for Mixed-Precision Quantization
Ivan Koryakovskiy (Huawei Technologies Co. Ltd), Gleb Odinokikh (Huawei Technologies Co. Ltd)
Super ResolutionCompressionNeural Architecture SearchConvolutional Neural NetworkImage
🎯 What it does: A One-Shot MPS method is proposed, which can search for various mixed-precision quantization schemes in one round, aimed at simultaneously finding the width allocation on the Pareto front of performance and hardware resources.
One-Stage 3D Whole-Body Mesh Recovery With Component Aware Transformer
Jing Lin (International Digital Economy Academy), Yu Li (Tsinghua University)
Pose EstimationTransformerImageMesh
🎯 What it does: A single-stage 3D full-body mesh recovery framework called OSX is proposed, which utilizes a Component Aware Transformer to directly regress SMPL-X parameters from a single image, achieving unified modeling of the body, hands, and face.
One-to-Few Label Assignment for End-to-End Dense Detection
Shuai Li (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: A one-to-few (label assignment) strategy is proposed, utilizing soft anchors to dynamically adjust positive and negative weights, enabling end-to-end dense detection without NMS in fully convolutional networks.
OneFormer: One Transformer To Rule Universal Image Segmentation
Jitesh Jain (IIT Roorkee), Humphrey Shi (University of Oregon)
SegmentationTransformerContrastive LearningImage
🎯 What it does: This paper proposes OneFormer, a unified Transformer-based image segmentation framework that can simultaneously perform semantic, instance, and panoptic segmentation in a single model with a single training session.
OPE-SR: Orthogonal Position Encoding for Designing a Parameter-Free Upsampling Module in Arbitrary-Scale Image Super-Resolution
Gaochao Song (Tianjin University), Ying He (Nanyang Technological University)
RestorationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a parameter-free upsampling module called OPE-Upscale, which utilizes Orthogonal Position Encoding (OPE) to achieve continuous reconstruction of images at arbitrary scales, completely replacing traditional MLP-based implicit networks.
Open Set Action Recognition via Multi-Label Evidential Learning
Chen Zhao (Kitware), Christopher Funk (Kitware)
RecognitionVideo
🎯 What it does: The MULE framework is proposed to achieve open-set action recognition and novel action detection for multiple actors and actions in videos.
Open Vocabulary Semantic Segmentation With Patch Aligned Contrastive Learning
Jishnu Mukhoti (University of Oxford), Ser-Nam Lim (Meta AI)
ClassificationSegmentationTransformerContrastive LearningImageTextMultimodality
🎯 What it does: Proposes Patch Aligned Contrastive Learning (PACL), which modifies the contrastive loss compatibility function of CLIP to achieve alignment at the visual patch level and text CLS level, thereby enabling unsupervised open-vocabulary semantic segmentation and zero-shot image classification.
Open-Category Human-Object Interaction Pre-Training via Language Modeling Framework
Sipeng Zheng (Renmin University of China), Qin Jin (Renmin University of China)
RecognitionObject DetectionKnowledge DistillationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes the OpenCat model, which rephrases human-object interaction (HOI) learning as a sequence generation task, achieving open-category HOI prediction through large-scale weakly supervised data and various proxy tasks for pre-training.
Open-Set Fine-Grained Retrieval via Prompting Vision-Language Evaluator
Shijie Wang (Dalian University of Technology), Qi Tian (Huawei Cloud and AI)
RetrievalKnowledge DistillationPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes the Prompting Vision-Language Evaluator (PLEor), which uses the pre-trained CLIP model as an evaluator to mine and transfer category-specific differences through learnable visual and textual prompts, enhancing the performance of open fine-grained retrieval.
Open-Set Likelihood Maximization for Few-Shot Learning
Malik Boudiaf (ÉTS Montreal), Ismail Ben Ayed (ÉTS Montreal)
ClassificationAnomaly DetectionOptimizationMeta LearningImage
🎯 What it does: A transfer method for few-shot open set recognition, OSLO, is proposed, which introduces a latent inlierness score within a maximum likelihood framework to simultaneously perform category prediction and anomaly detection.
Open-Set Representation Learning Through Combinatorial Embedding
Geeho Kim (Seoul National University), Bohyung Han (Seoul National University)
Representation LearningContrastive LearningImage
🎯 What it does: A combination embedding framework that integrates multiple coarse-grained meta-classifiers is proposed for representation learning under open sets.
Open-Set Semantic Segmentation for Point Clouds via Adversarial Prototype Framework
Jianan Li (University of Chinese Academy of Sciences), Qiulei Dong (Chinese Academy of Sciences)
Object DetectionSegmentationGenerative Adversarial NetworkPoint Cloud
🎯 What it does: This paper proposes an Adversarial Prototype Framework (APF) for open-set point cloud semantic segmentation, which not only retains the segmentation accuracy for known category point clouds but also identifies point clouds of unseen categories.
Open-Vocabulary Attribute Detection
María A. Bravo (University of Freiburg), Thomas Brox (University of Freiburg)
RecognitionObject DetectionVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: This paper proposes the Open Vocabulary Attribute Detection (OVAD) task and its dense annotation benchmark, along with baseline methods.
Open-Vocabulary Panoptic Segmentation With Text-to-Image Diffusion Models
Jiarui Xu (University of California San Diego), Shalini De Mello (NVIDIA)
Object DetectionSegmentationDiffusion modelImage
🎯 What it does: Proposes the ODISE method, which combines frozen internal features of the text-to-image diffusion model with CLIP to achieve open vocabulary panoptic segmentation for any category.
Open-Vocabulary Point-Cloud Object Detection Without 3D Annotation
Yuheng Lu (Peking University), Shanghang Zhang (University of California Berkeley)
Object DetectionTransformerVision Language ModelContrastive LearningPoint Cloud
🎯 What it does: Utilize a 2D pre-trained detector to generate pseudo 3D boxes and learn a 3D detector, then achieve text-point cloud cross-modal alignment through CLIP, enabling open vocabulary point cloud detection without 3D annotations.
Open-Vocabulary Semantic Segmentation With Mask-Adapted CLIP
Feng Liang (University of Texas at Austin), Diana Marculescu (Meta Reality Labs)
SegmentationTransformerPrompt EngineeringContrastive LearningImageText
🎯 What it does: For the open vocabulary semantic segmentation task, this paper first generates category-free mask candidates, then classifies the masked images using CLIP, and proposes methods for fine-tuning CLIP on masked images and optimizing mask prompts.
Open-World Multi-Task Control Through Goal-Aware Representation Learning and Adaptive Horizon Prediction
Shaofei Cai (Peking University), Yitao Liang (Peking University)
Representation LearningRobotic IntelligenceReinforcement LearningImage
🎯 What it does: This study investigates the problem of learning goal-conditioned policies in the Minecraft environment, proposing the Goal-Sensitive Backbone (GSB) and an adaptive horizon prediction module to address the challenges posed by task indistinguishability and the non-stationarity of environmental dynamics.
OpenGait: Revisiting Gait Recognition Towards Better Practicality
Chao Fan (Southern University of Science and Technology), Shiqi Yu (Southern University of Science and Technology)
RecognitionConvolutional Neural NetworkImageVideo
🎯 What it does: A unified and scalable gait recognition codebase, OpenGait, has been constructed, and existing methods have been fairly reproduced and comprehensively ablated on four mainstream datasets, ultimately proposing a structurally simple and powerful baseline model, GaitBase.
OpenMix: Exploring Outlier Samples for Misclassification Detection
Fei Zhu (Institute of Automation, Chinese Academy of Sciences), Cheng-Lin Liu (Institute of Automation, Chinese Academy of Sciences)
ClassificationAnomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: Using unlabeled external samples in an open world to help deep networks detect misclassifications and erroneous predictions of OOD samples.
OpenScene: 3D Scene Understanding With Open Vocabularies
Songyou Peng (Google Research), Thomas Funkhouser (Google Research)
SegmentationRetrievalConvolutional Neural NetworkVision Language ModelContrastive LearningImagePoint Cloud
🎯 What it does: A zero-shot 3D scene understanding framework called OpenScene has been constructed, which utilizes CLIP visual-language embeddings to jointly map 3D points, image pixels, and text into the same feature space, supporting various tasks such as semantic segmentation, material, functionality, and posture through arbitrary text queries.
Optimal Proposal Learning for Deployable End-to-End Pedestrian Detection
Xiaolin Song (Beijing University of Posts and Telecommunications), Honggang Zhang (Beijing University of Posts and Telecommunications)
Object DetectionImage
🎯 What it does: A deployable NMS-free pedestrian detection framework OPL is constructed, achieving end-to-end training and inference by combining the Coarse-to-Fine learning strategy and Completed Proposal Network.
Optimal Transport Minimization: Crowd Localization on Density Maps for Semi-Supervised Counting
Wei Lin (City University of Hong Kong), Antoni B. Chan (City University of Hong Kong)
OptimizationImage
🎯 What it does: This paper proposes a parameter-free Optimal Transport Minimization (OT-M) algorithm that converts density maps into hard point annotations and is used for semi-supervised counting;
Optimization-Inspired Cross-Attention Transformer for Compressive Sensing
Jiechong Song (Peking University), Jian Zhang (Peking University)
RestorationCompressionOptimizationTransformerImage
🎯 What it does: An optimized heuristic cross-attention Transformer module (OCT) is proposed, and a lightweight deep unfolding framework (OCTUF) is constructed based on this module for the reconstruction of image compressed sensing.
ORCa: Glossy Objects As Radiance-Field Cameras
Kushagra Tiwary (Massachusetts Institute of Technology), Ramesh Raskar (Massachusetts Institute of Technology)
GenerationDepth EstimationNeural Radiance FieldImage
🎯 What it does: This paper transforms unknown geometric smooth objects into light field cameras, utilizing multi-view images to jointly learn the implicit surface of the object, diffuse radiance, and a 5D environment light field, thereby achieving depth and radiance estimation of the surrounding environment and supporting super-viewpoint synthesis.
OReX: Object Reconstruction From Planar Cross-Sections Using Neural Fields
Haim Sawdayee (Blavatnik School of Computer Science Tel Aviv University), Amit H. Bermano (Blavatnik School of Computer Science Tel Aviv University)
GenerationData SynthesisNeural Radiance FieldPoint CloudMeshBiomedical DataMagnetic Resonance Imaging
🎯 What it does: The research focuses on reconstructing 3D shapes from sparse planar slices and proposes the OReX method.
OrienterNet: Visual Localization in 2D Public Maps With Neural Matching
Paul-Edouard Sarlin (ETH Zurich), Vasileios Balntas (Meta Reality Labs)
Pose EstimationAutonomous DrivingConvolutional Neural NetworkSimultaneous Localization and MappingImage
🎯 What it does: This paper presents OrienterNet, an end-to-end deep network capable of achieving sub-meter level 3D pose localization using only 2D public maps (OpenStreetMap) and a single image.
Orthogonal Annotation Benefits Barely-Supervised Medical Image Segmentation
Heng Cai (Nanjing University), Yang Gao (Nanjing University)
SegmentationImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes an 'orthogonal labeling' method that only requires labeling two mutually perpendicular slices in each 3D medical image volume, and based on this, constructs a Dense-Sparse Co-Training (DeSCO) model to achieve efficient semi-supervised segmentation under sparse labeling.
OSAN: A One-Stage Alignment Network To Unify Multimodal Alignment and Unsupervised Domain Adaptation
Ye Liu (Tencent Youtu Lab), Bo Ren (Tencent Youtu Lab)
Domain AdaptationContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes a One-Stage Alignment Network (OSAN) that unifies the challenges of multimodal alignment and unsupervised domain adaptation.
OSRT: Omnidirectional Image Super-Resolution With Distortion-Aware Transformer
Fanghua Yu (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), Chao Dong (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences)
RestorationSuper ResolutionTransformerImage
🎯 What it does: For the super-resolution task of panoramic (360°) images, the authors propose fisheye downsampling to simulate the real imaging process and design a distortion-aware Transformer (OSRT) that adaptively modulates ERP distortion through two modules, DAAB and DACB; at the same time, an augmentation strategy is employed to generate pseudo ERP data from planar images to alleviate overfitting in large models.
OT-Filter: An Optimal Transport Filter for Learning With Noisy Labels
Chuanwen Feng (University of Science and Technology of China), Xike Xie (University of Science and Technology of China)
ClassificationData-Centric LearningImage
🎯 What it does: This paper proposes OT-Filter, which uses optimal transport (Wasserstein barycenter and sparse OT) methods to automatically filter clean samples from noisy labeled data and combines it with semi-supervised training to achieve robust learning.
OTAvatar: One-Shot Talking Face Avatar With Controllable Tri-Plane Rendering
Zhiyuan Ma (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
GenerationData SynthesisPose EstimationGenerative Adversarial NetworkVideo
🎯 What it does: This paper proposes a one-shot 3D consistent talking head system called OTAvatar, which utilizes controllable three-plane rendering to generate high-quality animated avatars from a single portrait, allowing for any viewpoint, expression, and pose.
Out-of-Candidate Rectification for Weakly Supervised Semantic Segmentation
Zesen Cheng (Peking University), Jie Chen (Peking University)
SegmentationTransformerImage
🎯 What it does: Proposes the Out-of-Candidate Rectification (OCR) mechanism to correct erroneous pixels where the predicted categories are inconsistent with the image label set in weakly supervised semantic segmentation, thereby improving segmentation accuracy.
Out-of-Distributed Semantic Pruning for Robust Semi-Supervised Learning
Yu Wang (Peking University), Jie Chen (Peking University)
ClassificationAnomaly DetectionContrastive LearningImage
🎯 What it does: This paper proposes a framework called OOD Semantic Pruning (OSP) for removing OOD information at the semantic level, aimed at improving ID classification and OOD detection performance in robust semi-supervised learning.
OvarNet: Towards Open-Vocabulary Object Attribute Recognition
Keyan Chen (Beihang University), Weidi Xie (Shanghai AI Laboratory)
RecognitionObject DetectionKnowledge DistillationTransformerVision Language ModelImageText
🎯 What it does: This paper proposes an open vocabulary framework called OvarNet that can simultaneously perform object detection and attribute recognition, accurately locating and describing novel objects during testing with only partial categories and attributes seen during training.
Overcoming the Trade-Off Between Accuracy and Plausibility in 3D Hand Shape Reconstruction
Ziwei Yu (National University of Singapore), Angela Yao (National University of Singapore)
GenerationPose EstimationKnowledge DistillationGraph Neural NetworkTransformerAuto EncoderPoint CloudMesh
🎯 What it does: A weakly supervised framework is proposed, combining non-parametric grid fitting with the MANO statistical model to achieve accurate and physically feasible 3D hand reconstruction.
Overlooked Factors in Concept-Based Explanations: Dataset Choice, Concept Learnability, and Human Capability
Vikram V. Ramaswamy (Princeton University), Olga Russakovsky (Princeton University)
Explainability and InterpretabilityImage
🎯 What it does: This paper systematically evaluates three key factors of concept explanation methods: the selection of probe datasets, concept learnability, and the impact of explanation complexity on human interpretability.
OVTrack: Open-Vocabulary Multiple Object Tracking
Siyuan Li (Computer Vision Lab, ETH Zurich), Fisher Yu (Computer Vision Lab, ETH Zurich)
Object TrackingKnowledge DistillationVision Language ModelDiffusion modelContrastive LearningImageVideo
🎯 What it does: We propose OVTrack, an open-vocabulary multi-object tracker capable of tracking objects of any category in videos.
PA&DA: Jointly Sampling Path and Data for Consistent NAS
Shun Lu (Chinese Academy of Sciences), Chengru Song (Kuaishou Technology)
Neural Architecture SearchImage
🎯 What it does: Proposes the PA&DA method, which jointly optimizes the path and data sampling distribution in supernet training to reduce gradient variance and improve ranking consistency in one-shot NAS.
PaCa-ViT: Learning Patch-to-Cluster Attention in Vision Transformers
Ryan Grainger (North Carolina State University), Tianfu Wu (North Carolina State University)
ClassificationObject DetectionSegmentationExplainability and InterpretabilityComputational EfficiencyTransformerImage
🎯 What it does: The Patch-to-Cluster Attention (PaCa) module is proposed, which replaces the patch-to-patch attention of the Vision Transformer with cluster-based key/value computation, reducing the complexity from O(N²) to O(NM) while obtaining interpretable visual tokens.
PACO: Parts and Attributes of Common Objects
Vignesh Ramanathan (Meta AI), Dhruv Mahajan (Meta AI)
Object DetectionSegmentationRetrievalTransformerImageVideoBenchmark
🎯 What it does: A unified large-scale dataset called PACO has been constructed, which includes objects, object parts, and attributes, and three benchmark tasks are provided: part segmentation, object/part attribute prediction, and zero-shot instance retrieval.
Paint by Example: Exemplar-Based Image Editing With Diffusion Models
Binxin Yang (University of Science and Technology of China), Fang Wen (Microsoft Research Asia)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: A framework for example image editing based on diffusion models, called Paint by Example, is proposed, which can integrate the target from the reference image into the target image in a controllable manner while maintaining the consistency of the original background.
Painting 3D Nature in 2D: View Synthesis of Natural Scenes From a Single Semantic Mask
Shangzhan Zhang (Zhejiang University), Xiaowei Zhou (Zhejiang University)
SegmentationGenerationData SynthesisDepth EstimationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a method for generating 3D perspective synthesis of natural scenes using only a single semantic segmentation map, utilizing semantic fields and SPADE networks to generate multi-view consistent RGB images.
Paired-Point Lifting for Enhanced Privacy-Preserving Visual Localization
Chunghwan Lee (Hanyang University), Je Hyeong Hong (Hanyang University)
Pose EstimationSafty and PrivacySimultaneous Localization and MappingPoint Cloud
🎯 What it does: Designed and implemented the Paired-Point Lifting (PPL) scheme, which pairs sparse point clouds and connects them into line clouds, thereby enhancing privacy protection while maintaining visual localization accuracy.
PaletteNeRF: Palette-Based Appearance Editing of Neural Radiance Fields
Zhengfei Kuang (Stanford University), Kalyan Sunkavalli (Adobe Research)
Image TranslationGenerationNeural Radiance FieldImage
🎯 What it does: A NeRF appearance editing framework based on PaletteNeRF has been developed, enabling intuitive and controllable re-coloring, style transfer, and lighting modification through a 3D color reference.
PanelNet: Understanding 360 Indoor Environment via Panel Representation
Haozheng Yu (Tencent America), Shan Liu (Tencent America)
SegmentationDepth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: Design the PanelNet framework, which achieves 360° indoor depth estimation, semantic segmentation, and layout prediction by slicing ERP images into vertical panels and incorporating geometric embeddings and a Local2Global Transformer.
PAniC-3D: Stylized Single-View 3D Reconstruction From Portraits of Anime Characters
Shuhong Chen (ByteDance), Matthias Zwicker (University of Maryland)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImageMeshBenchmark
🎯 What it does: Automatically reconstructing stylized 3D character head models from a single anime character portrait.
PanoHead: Geometry-Aware 3D Full-Head Synthesis in 360deg
Sizhe An (ByteDance Inc.), Linjie Luo (ByteDance Inc.)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Proposes PanoHead, achieving 3D perception consistency synthesis of 360° full head images.
Panoptic Compositional Feature Field for Editable Scene Rendering With Network-Inferred Labels via Metric Learning
Xinhua Cheng (Peking University), Jian Zhang (Peking University)
SegmentationGenerationNeural Radiance FieldContrastive LearningPoint Cloud
🎯 What it does: Proposes the PCFF framework, which utilizes labels inferred from a 2D panoptic segmentation network to learn editable object-level neural implicit representations, enabling instance-level scene editing and visualization.
Panoptic Lifting for 3D Scene Understanding With Neural Fields
Yawar Siddiqui (Technical University of Munich), Peter Kontschieder (Meta Reality Labs)
SegmentationDepth EstimationNeural Radiance FieldImage
🎯 What it does: By elevating the only available RGB images and the poorly consistent machine-generated 2D panoramic segmentation results to 3D space, a panoramic radiance field is constructed that can generate color, depth, semantic, and instance information from any viewpoint.
Panoptic Video Scene Graph Generation
Jingkang Yang (Nanyang Technological University), Ziwei Liu (SenseTime Research)
Object DetectionObject TrackingSegmentationGenerationTransformerVideoBenchmark
🎯 What it does: This paper proposes a panoramic scene graph generation task for videos (PVSG) and constructs a corresponding dataset and benchmark model.
PanoSwin: A Pano-Style Swin Transformer for Panorama Understanding
Zhixin Ling (Fudan University), Guichun Zhou (Fudan University)
ClassificationObject DetectionTransformerImage
🎯 What it does: A Swin Transformer architecture specifically designed for equidistant cylindrical projection panoramic images, called PanoSwin, has been designed and implemented to address the issues of boundary discontinuity and spatial distortion in panoramic images.
Parallel Diffusion Models of Operator and Image for Blind Inverse Problems
Hyungjin Chung (KAIST), Jong Chul Ye (KAIST)
RestorationDiffusion modelScore-based ModelImageStochastic Differential Equation
🎯 What it does: This paper studies the blind inverse problem and proposes the BlindDPS parallel diffusion backward sampling method to achieve joint estimation of the operator and the image.
Parameter Efficient Local Implicit Image Function Network for Face Segmentation
Mausoom Sarkar (Adobe), Balaji Krishnamurthy (Adobe)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: A lightweight face segmentation network based on Local Implicit Image Function (FP-LIIF) is proposed, capable of achieving pixel-level semantic segmentation of faces with extremely low parameters.
Parametric Implicit Face Representation for Audio-Driven Facial Reenactment
Ricong Huang (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
GenerationData SynthesisTransformerGenerative Adversarial NetworkVideoAudio
🎯 What it does: This paper proposes an audio-driven facial reproduction framework, which combines interpretable 3DMM expression parameters with a high-expressiveness implicit three-plane representation (PIR) to achieve controllable and high-quality speaker reproduction.
PartDistillation: Learning Parts From Instance Segmentation
Jang Hyun Cho (University of Texas at Austin), Vignesh Ramanathan (Meta AI)
Object DetectionSegmentationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A scalable self-training framework called PartDistillation is proposed, which learns object part segmentation unsupervisedly from the intermediate embedding information of instance segmentation models.
Partial Network Cloning
Jingwen Ye (National University of Singapore), Xinchao Wang (National University of Singapore)
OptimizationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A Partial Network Cloning method is proposed to clone transferable modules from pre-trained models and directly insert them into target models, allowing the target model to acquire new functionalities without altering the original parameters.
PartManip: Learning Cross-Category Generalizable Part Manipulation Policy From Point Cloud Observations
Haoran Geng (Peking University), He Wang (Peking University)
Domain AdaptationKnowledge DistillationRobotic IntelligenceReinforcement LearningPoint CloudBenchmark
🎯 What it does: This paper proposes PartManip, a cross-category, part-based object manipulation benchmark, and designs a two-stage distillation framework from expert strategies to visual strategies. It utilizes part coordinate mapping, part-aware rewards, point cloud augmentation, a Sparse-UNet backbone, and domain adversarial learning to achieve general manipulation in both simulation and reality.
PartMix: Regularization Strategy To Learn Part Discovery for Visible-Infrared Person Re-Identification
Minsu Kim (Yonsei University), Kwanghoon Sohn (Korea Institute of Science and Technology)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes an enhancement method based on part descriptor mixing called PartMix, which is used for regularization of visible-infrared person re-identification (VI-ReID) models, and improves model robustness through contrastive learning and entropy-based mining.
Parts2Words: Learning Joint Embedding of Point Clouds and Texts by Bidirectional Matching Between Parts and Words
Chuan Tang (Jilin University), Yi Chang (Jilin University)
SegmentationRetrievalRecurrent Neural NetworkTextPoint Cloud
🎯 What it does: An end-to-end framework is proposed to learn joint embeddings of point clouds and text, achieving high-precision shape-text retrieval through bidirectional matching between segments obtained from point cloud segmentation and text vocabulary.
PartSLIP: Low-Shot Part Segmentation for 3D Point Clouds via Pretrained Image-Language Models
Minghua Liu (University of California San Diego), Hao Su (University of California San Diego)
Object DetectionSegmentationTransformerSupervised Fine-TuningVision Language ModelPoint Cloud
🎯 What it does: This paper proposes a method for low-shot (zero/few-shot) 3D point cloud part segmentation using the pre-trained image-language model GLIP.
Passive Micron-Scale Time-of-Flight With Sunlight Interferometry
Alankar Kotwal (Carnegie Mellon University), Ioannis Gkioulekas (Carnegie Mellon University)
Depth EstimationImage
🎯 What it does: A movable optical system was constructed to perform full-field interference measurements outdoors using direct sunlight, achieving passive time-of-flight imaging at the micron level through the temporal and spatial incoherence of sunlight, and enabling the reconstruction of three-dimensional structures of targets such as circuit boards and metal coins.
Patch-Based 3D Natural Scene Generation From a Single Example
Weiyu Li (Shandong University), Baoquan Chen (Peking University)
GenerationData SynthesisNeural Radiance FieldPoint Cloud
🎯 What it does: This paper proposes a single-example based 3D scene generation method that synthesizes diverse and realistic natural scenes using a multi-scale Patch-Nearest-Neighbor framework on Plenoxels representation.
Patch-Craft Self-Supervised Training for Correlated Image Denoising
Gregory Vaksman (Technion), Michael Elad (Technion)
RestorationConvolutional Neural NetworkImageVideo
🎯 What it does: An unsupervised image denoising framework called Patch-Craft is proposed, which generates artificially synthesized 'cropped' target images by performing nearest neighbor matching (excluding itself) on overlapping patches of each input image within a group of images. These targets are then used as training labels to learn the denoiser; during training, statistical analysis of the target noise is conducted, and low covariance samples are trimmed to reduce the correlation between target noise and input noise.
Patch-Mix Transformer for Unsupervised Domain Adaptation: A Game Perspective
Jinjing Zhu (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
Domain AdaptationTransformerImage
🎯 What it does: This paper proposes an unsupervised domain adaptation framework based on a visual Transformer called PMTrans, where the core is the PatchMix module that constructs an intermediate domain by learning to mix patches from the source and target domains, thereby achieving domain alignment.
PATS: Patch Area Transportation With Subdivision for Local Feature Matching
Junjie Ni (Zhejiang University), Guofeng Zhang (Zhejiang University)
Object DetectionPose EstimationTransformerImage
🎯 What it does: Proposes Patch Area Transportation with Subdivision (PATS) which significantly improves the quality of local feature matching under large-scale differences through self-supervised multi-to-multi matching and scale alignment.
PC2: Projection-Conditioned Point Cloud Diffusion for Single-Image 3D Reconstruction
Luke Melas-Kyriazi (University of Oxford), Andrea Vedaldi (University of Oxford)
RestorationGenerationDiffusion modelImagePoint Cloud
🎯 What it does: A point cloud diffusion model based on projection conditions, PC2, is proposed to recover 3D shapes and colors from a single RGB image.
pCON: Polarimetric Coordinate Networks for Neural Scene Representations
Henry Peters (University of California), Achuta Kadambi (University of California)
RestorationNeural Radiance FieldImage
🎯 What it does: A new neural field architecture called pCON is proposed for image reconstruction while preserving polarization information.
PCR: Proxy-Based Contrastive Replay for Online Class-Incremental Continual Learning
Huiwei Lin (Harbin Institute of Technology), Yunming Ye (Harbin Institute of Technology)
ClassificationRepresentation LearningContrastive LearningImage
🎯 What it does: A novel online class-incremental learning framework called PCR, which combines proxy and contrastive learning, is proposed to alleviate the problem of catastrophic forgetting.
PCT-Net: Full Resolution Image Harmonization Using Pixel-Wise Color Transformations
Julian Jorge Andrade Guerreiro (University of Tokyo), Björn Stenger (Rakuten Group)
Image HarmonizationConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposes PCT-Net, a method for harmonizing full-resolution images using pixel-level color transformations.
PD-Quant: Post-Training Quantization Based on Prediction Difference Metric
Jiawei Liu (Huazhong University of Science and Technology), Wenyu Liu (Huazhong University of Science and Technology)
CompressionOptimizationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A post-training quantization method based on prediction difference measurement, PD-Quant, is developed to optimize quantization parameters and reduce overfitting through distribution correction.
PDPP:Projected Diffusion for Procedure Planning in Instructional Videos
Hanlin Wang (Nanjing University), Limin Wang (Nanjing University)
GenerationRobotic IntelligenceDiffusion modelVideo
🎯 What it does: The study generates goal-oriented action sequences given the initial and target visual states in instructional videos, proposing to view process planning as conditional distribution fitting, using a projection diffusion model to output the complete action sequence in one go.
PeakConv: Learning Peak Receptive Field for Radar Semantic Segmentation
Liwen Zhang (Intelligent Science and Technology Academy of CASIC), Zhe Ma (Intelligent Science and Technology Academy of CASIC)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: PeakConv is proposed to achieve radar semantic segmentation using Peak Receptive Field (PRF);
PEAL: Prior-Embedded Explicit Attention Learning for Low-Overlap Point Cloud Registration
Junle Yu (Hangzhou Dianzi University), Guojun Dai (Hangzhou Dianzi University)
Pose EstimationOptimizationTransformerPoint Cloud
🎯 What it does: A point cloud registration method called PEAL is proposed, which utilizes overlapping prior explicit unidirectional attention learning, significantly improving the matching accuracy of low-overlap point clouds.
PEFAT: Boosting Semi-Supervised Medical Image Classification via Pseudo-Loss Estimation and Feature Adversarial Training
Qingjie Zeng (Northwestern Polytechnical University), Yong Xia (Northwestern Polytechnical University)
ClassificationConvolutional Neural NetworkContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A semi-supervised medical image classification framework called PEFAT is proposed, which combines pseudo-loss estimation and feature-level adversarial training to improve classification performance with limited labeled data.
Perception and Semantic Aware Regularization for Sequential Confidence Calibration
Zhenghua Peng (South China University of Technology), Shuangping Huang (Guangdong University of Technology)
RecognitionRecurrent Neural NetworkLarge Language ModelTextSequentialAudio
🎯 What it does: This paper proposes a Perceptual and Semantic Perception Sequence Regularization framework (PSSR), which enhances the confidence calibration of deep sequence recognition models by introducing sequences that are visually similar and semantically related to the target sequence as additional supervision.
Perception-Oriented Single Image Super-Resolution Using Optimal Objective Estimation
Seung Ho Park (Seoul National University), Nam Ik Cho (Seoul National University)
RestorationSuper ResolutionConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A single image super-resolution framework is proposed, which utilizes Optimal Objective Estimation (OOE) to dynamically select the optimal loss combination for each region, and achieves variable target super-resolution through a generative model.
PermutoSDF: Fast Multi-View Reconstruction With Implicit Surfaces Using Permutohedral Lattices
Radu Alexandru Rosu (University of Bonn), Sven Behnke (University of Bonn)
GenerationData SynthesisComputational EfficiencyNeural Radiance FieldPoint CloudMesh
🎯 What it does: An implicit surface model is constructed using a hashable Permutohedral grid and SDF representation, achieving efficient multi-view reconstruction.
Persistent Nature: A Generative Model of Unbounded 3D Worlds
Lucy Chai (Massachusetts Institute of Technology), Noah Snavely (Cornell Tech)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: Based on single-view natural landscape images without poses, an unsupervised generative model was trained that can synthesize infinite, persistent 3D natural scenes and supports arbitrary camera trajectories.
Person Image Synthesis via Denoising Diffusion Model
Ankan Kumar Bhunia (Mohamed bin Zayed University of AI), Fahad Shahbaz Khan (Mohamed bin Zayed University of AI)
GenerationData SynthesisPose EstimationDiffusion modelImage
🎯 What it does: A pose-guided portrait generation framework PIDM based on a denoising diffusion model is proposed, which can synthesize high-quality portraits under the conditions of a given pose and source image.
PersonNeRF: Personalized Reconstruction From Photo Collections
Chung-Yi Weng (University of Washington), Ira Kemelmacher-Shlizerman (University of Washington)
GenerationPose EstimationNeural Radiance FieldImage
🎯 What it does: Construct a 3D portrait model that can freely switch poses, viewpoints, and appearances using only a few personal photos with different poses and outfits.
Perspective Fields for Single Image Camera Calibration
Linyi Jin (University of Michigan), David F. Fouhey (University of Michigan)
Data SynthesisKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes Perspective Fields as a representation of local perspective attributes in images, training a network to predict this representation and recover camera parameters for image synthesis and matching.
PET-NeuS: Positional Encoding Tri-Planes for Neural Surfaces
Yiqun Wang (Chongqing University), Peter Wonka (KAUST)
GenerationData SynthesisNeural Radiance FieldPoint Cloud
🎯 What it does: A PET-NeuS framework based on tri-plane representation is proposed for multi-view surface reconstruction.
PHA: Patch-Wise High-Frequency Augmentation for Transformer-Based Person Re-Identification
Guiwei Zhang (Beihang University), Shiliang Pu (Hikvision Research Institute)
RecognitionRetrievalTransformerContrastive LearningImage
🎯 What it does: In the pedestrian re-identification task, the authors propose Patch-wise High-frequency Augmentation (PHA), which enhances the Vision Transformer's ability to express high-frequency details and improves its recognition performance through frequency domain decomposition and self-attention analysis.
Phase-Shifting Coder: Predicting Accurate Orientation in Oriented Object Detection
Yi Yu (Southeast University), Feipeng Da (Southeast University)
Object DetectionImage
🎯 What it does: A phase shift encoder (PSC) and its dual-frequency version (PSCD) are proposed for angle regression in inclined target detection, addressing boundary discontinuity and square issues;
Phone2Proc: Bringing Robust Robots Into Our Chaotic World
Matt Deitke (Allen Institute for AI), Aniruddha Kembhavi (Allen Institute for AI)
Domain AdaptationRobotic IntelligenceRecurrent Neural NetworkReinforcement LearningImage
🎯 What it does: Quickly scan the target environment using a mobile phone, and generate diverse training scenarios that are semantically similar through conditional procedural generation, achieving a high success rate on real robots after training visual navigation agents.
Photo Pre-Training, but for Sketch
Ke Li (Beijing University of Posts and Telecommunications), Yi-Zhe Song (University of Surrey)
RetrievalConvolutional Neural NetworkTransformerSupervised Fine-TuningContrastive LearningImage
🎯 What it does: Improving fine-grained sketch-based image retrieval (FG-SBIR) performance by utilizing the neighborhood topology of photo pre-trained models as additional supervision.