CVPR 2026 Papers — Page 26
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
ParTY: Part-Guidance for Expressive Text-to-Motion Synthesis
KunHo Heo (Kyung Hee University), MyeongAh Cho (Kyung Hee University)
GenerationData SynthesisGraph Neural NetworkTransformerLarge Language ModelVision-Language-Action ModelAuto EncoderContrastive LearningTextMultimodality
🎯 What it does: This paper proposes the ParTY framework, which generates full-body motion from text while enhancing the coordination between part-level expressions and overall motion through part guidance.
PAS: A Training-Free Stabilizer for Temporal Encoding in Video LLMs
Bowen Sun (University of California Merced), Yiwei Wang (University of California Merced)
TransformerLarge Language ModelVision Language ModelVideoMultimodality
🎯 What it does: Proposed a training-agnostic Phase Aggregated Smoothing (PAS) mechanism to enhance the temporal encoding stability of Video LLMs.
PAS: Prelim Attention Score for Detecting Object Hallucinations in Large Vision-Language Models
Nhat Hoang (University of Florida), Manish Bhattarai (University of Florida)
Anomaly DetectionExplainability and InterpretabilityTransformerVision Language ModelImage
🎯 What it does: Propose using the attention distribution of pre-output tokens during the generation process of a large vision-language model (LVLM) to detect object hallucinations, and introduce a training-free, reference-agnostic pre-attention score (PAS).
PatchAlign3D: Local Feature Alignment for Dense 3D Shape Understanding
Souhail Hadgi (cole polytechnique), Maks Ovsjanikov (cole polytechnique)
SegmentationTransformerVision Language ModelContrastive LearningTextPoint Cloud
🎯 What it does: This paper proposes PatchAlign3D, an encoder using only point cloud Transformers, which directly generates locally aligned features with language through two-stage pre-training, achieving dense segmentation of zero-shot 3D shapes.
PatchScene: Patch-based Voxel Diffusion Model for Large-Scale Scene Completion
Qingdong Xu (MEGVII Technology), Jiyao Zhang (Peking University)
GenerationAutonomous DrivingDiffusion modelPoint Cloud
🎯 What it does: Propose PatchScene, a sparse LiDAR scene completion framework based on block-wise voxel diffusion, capable of generating high-precision, boundary-free point clouds in large-scale, spatiotemporally continuous environments;
PAUL: Uncertainty-Guided Partition and Augmentation for Robust Cross-View Geo-Localization under Noisy Correspondence
Zheng Li (National University of Defense Technology), Mingrui Lao (National University of Defense Technology)
RetrievalImage
🎯 What it does: This paper proposes a cross-view geolocation method called PAUL, designed to address the spatial misalignment (noisy correspondences) problem caused by GPS drift in UAV and satellite image pairings;
PAVAS: Physics-Aware Video-to-Audio Synthesis
Oh Hyun-Bin (POSTECH), Yuki Mitsufuji (Sony AI)
GenerationData SynthesisVision Language ModelDiffusion modelVideoTextMultimodalityPhysics RelatedAudio
🎯 What it does: Proposed a physics-aware image-to-audio synthesis framework called PAVAS, which can generate audio that is more consistent with visual content and physically more credible by utilizing object mass and velocity information from videos.
PC-Talk: Precise Facial Animation Control for Audio-Driven Talking Face Generation
Baiqin Wang (Chinese Academy of Sciences), Zhen Lei (Chinese Academy of Sciences)
GenerationTransformerVideoMultimodalityAudio
🎯 What it does: This paper proposes an audio-driven speaker animation generation framework called PC-Talk, which can finely control lip synchronization alignment and emotional expression.
PCA-Seg: Revisiting Cost Aggregation for Open-Vocabulary Semantic and Part Segmentation
Jianjian Yin (Nanjing University of Science and Technology), Fumin Shen (University of Electronic Science and Technology of China)
SegmentationTransformerMixture of ExpertsVision Language ModelImage
🎯 What it does: Propose the PCA-Seg model, which addresses the knowledge interference between class-level semantics and spatial context in open-vocabulary semantics and part segmentation through mechanisms such as parallel cost aggregation, expert-driven perceptual learning (EPL), and feature orthogonal decoupling (FOD).
PDCR: Perception-Decomposed Confidence Reward for Vision-Language Reasoning
Hee Suk Yoon (Korea Advanced Institute of Science and Technology), Chang D. Yoo (Korea Advanced Institute of Science and Technology)
Large Language ModelReinforcement LearningMultimodality
🎯 What it does: Proposed the Perception-Decomposed Confidence Reward (PDCR), which classifies multimodal reasoning steps into visual perception and text reasoning categories using unsupervised visual dependency scores. It normalizes confidence improvements within each category to provide reinforcement learning with more stable, non-diluted step-level rewards, significantly enhancing multimodal reasoning performance.
PDD: Manifold-Prior Diverse Distillation for Medical Anomaly Detection
Xijun Lu (Tianjin University), Liang Wan (Tianjin University)
Anomaly DetectionKnowledge DistillationConvolutional Neural NetworkTransformerMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a dual-teacher dual-student manifold distillation framework called PDD for medical image anomaly detection, which integrates the global context from VMamba-Tiny with the local structural priors from Wide-ResNet50 to construct a unified high-dimensional manifold, achieving more robust normal pattern learning through multi-scale fusion and diversity distillation.
PE3R: Perception-Efficient 3D Reconstruction
Jie Hu (National University of Singapore), Xinchao Wang (National University of Singapore)
SegmentationGenerationDepth EstimationTransformerNeural Radiance FieldImagePoint CloudRetrieval-Augmented Generation
🎯 What it does: Propose the PE3R framework to achieve efficient 3D semantic reconstruction and open-vocabulary interaction without scene fine-tuning or camera parameters;
PEARL: Geometry Aligns Semantics for Training-Free Open-Vocabulary Semantic Segmentation
Gensheng Pei (Sungkyunkwan University), Byeungwoo Jeon (Sungkyunkwan University)
SegmentationTransformerVision Language ModelImageTextBenchmark
🎯 What it does: Proposes the PEARL framework, achieving training-free open-vocabulary semantic segmentation by inserting Procrustes alignment into the final self-attention of ViT and performing text-aware Laplacian propagation on a small grid without additional training.
PECCVAI: Overcoming the Brittleness of AI Image Watermarking Under Visual Paraphrasing Attacks
Shreyas Dixit (VIIT Pune), Amitava Das (BITS Pilani)
Safty and PrivacyDiffusion modelImage
🎯 What it does: Proposed and implemented a robust image watermarking method called PECCAVI against visual pseudowrite attacks.
Perceiving the Near, Reasoning the Distant: Coherent Long-Horizon Trajectory Prediction for Autonomous Driving
Hua Hu (City University of Hong Kong), Jianping Wang (City University of Hong Kong)
Autonomous DrivingRecurrent Neural NetworkTransformerTime SeriesSequentialBenchmark
🎯 What it does: Propose a two-stage trajectory prediction network, NDPNet, which models near-term and distant time windows separately and achieves smooth transition between the two stages through a time bridging module; meanwhile, introduce a motion-aware consistency (MAC) loss that directly constrains the velocity and orientation consistency of predicted trajectories during training, enhancing the physical plausibility of trajectories.
Percept-WAM: Perception-Enhanced World-Awareness-Action Model for Robust End-to-End Autonomous Driving
Jianhua Han (Yinwang Intelligent Technology Co. Ltd.), Hang Xu (Yinwang Intelligent Technology Co. Ltd.)
Object DetectionSegmentationAutonomous DrivingTransformerLarge Language ModelVision Language ModelImageTextMultimodalityPoint Cloud
🎯 What it does: Built a unified vision-language model (VLM) by fusing 2D/3D perception and planning tasks through 'World-PV' and 'World-BEV' tokens, achieving end-to-end autonomous driving.
Perception Characteristics Distance: Measuring Stability and Robustness of Perception System in Dynamic Conditions under a Certain Decision Rule
Boyu Jiang (Virginia Tech), Feng Guo (Virginia Tech)
Object DetectionSegmentationAutonomous DrivingConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: Propose a new perception evaluation metric called Perception Characteristics Distance (PCD) to quantify the reliable detection range of perception systems under different distances and environmental conditions, and calculate the average PCD (aPCD) based on this metric; meanwhile, construct the SensorRainFall dataset, which provides real distance, bounding box, and segmentation annotations under controllable rainy weather and lighting conditions.
Perceptual 3D Simulation With Physical World Modeling
Wanhee Lee, Daniel LK Yamins
GenerationData SynthesisTransformerWorld ModelOptical FlowImageVideoMultimodality
🎯 What it does: This paper proposes the P3Sim system, which can simulate future scene states through perceptual inputs under partial visual observations and incomplete 3D transformation signals, achieving various 3D transformation tasks such as novel view synthesis, object manipulation, and dynamic scene prediction;
Perceptual Neural Video Compression with Color Separation and Rank Chain
Xiongzhuang Liang (University of Science and Technology of China), Dong Liu (University of Science and Technology of China)
CompressionGenerative Adversarial NetworkVideo
🎯 What it does: This paper proposes a variable-rate neural video compression framework PNVC-C/PNVC-CR based on color separation and grade chain GAN to enhance perceptual and objective compression quality.
Perceptual-Evidence Anchored Reinforced Learning for Multimodal Reasoning
Chi Zhang (Wuhan University), Jing Zhang (Wuhan University)
Reinforcement LearningVision Language ModelMultimodality
🎯 What it does: Introduce a perceptual inspection mechanism in reinforcement learning for vision-language models (VLMs), constructing a perceptual checklist and using its results as rewards and gating to enhance the model's协同能力 (synergy) between perception and reasoning.
PercHead: Perceptual Head Model for Single-Image 3D Head Reconstruction & Editing
Antonio Oroz (Technical University of Munich), Tobias Kirschstein (Technical University of Munich)
GenerationTransformerVision Language ModelContrastive LearningGaussian SplattingImage
🎯 What it does: Designed and implemented PercHead, a single-image 3D head reconstruction and editable model based on Vision Transformer.
PerformRecast: Expression and Head Pose Disentanglement for Portrait Video Editing
Jiadong Liang (HUJING Digital Media & Entertainment Group), Huan Fu (HUJING Digital Media & Entertainment Group)
Image TranslationGenerationPose EstimationTransformerGenerative Adversarial NetworkVideo
🎯 What it does: Propose PerformRecast, an algorithm capable of editing facial expressions in source videos or static images, supporting two inference modes: source video expression replacement and enhancement.
PerpetualWonder: Long-horizon Action-conditioned 4D Scene Generation
Jiahao Zhan (Stanford University), Jiajun Wu (Stanford University)
GenerationData SynthesisDiffusion modelGaussian SplattingSimultaneous Localization and MappingImageVideoPoint CloudMesh
🎯 What it does: Proposes PerpetualWonder, a hybrid generative simulator that can generate 4D scenes over long action sequences starting from a single image, achieving continuous interaction and arbitrary view rendering through closed-loop bidirectional updates between physical and visual systems.
PersonaLive! Expressive Portrait Image Animation for Live Streaming
Zhiyuan Li (University of Macau), Xiaodong Cun (Great Bay University)
GenerationDiffusion modelGenerative Adversarial NetworkImageVideo
🎯 What it does: Propose PersonaLive, a diffusion model framework for real-time streaming portrait animation;
Personalized Federated Training of Diffusion Models with Privacy Guarantees
Kumar Kshitij Patel (Yale FDS), Lingxiao Wang (New Jersey Institute of Technology)
Federated LearningSafty and PrivacyDiffusion modelImage
🎯 What it does: Propose a federated learning framework PFDM for training diffusion models on decentralized and private datasets, while simultaneously providing a global shared model and personalized models for each client;
Personalized Image Descriptions from Attention Sequences
Ruoyu Xue (Stony Brook University), Dimitris Samaras (Stony Brook University)
GenerationContrastive LearningImageSequential
🎯 What it does: Developed the DEPER model, which generates personalized image descriptions by modeling individual gaze sequences.
Personalized Longitudinal Medical Report Generation via Temporally-Aware Federated Adaptation
He Zhu (Hokkaido University), Miki Haseyama (Hokkaido University)
GenerationFederated LearningMeta LearningTransformerBiomedical DataElectronic Health Records
🎯 What it does: Propose the FedTAR framework, which utilizes federated learning to achieve personalized, time-aware medical report generation;
PersonaVLM: Long-Term Personalized Multimodal LLMs
Chang Nie (Nanjing University), Caifeng Shan (Nanjing University)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the PersonaVLM framework, enabling multimodal large language models to achieve long-term personalized interaction, capable of remembering, reasoning, and dynamically responding to user personalities;
PET-DINO: Unifying Visual Cues into Grounding DINO with Prompt-Enriched Training
Weifu Fu, Chengjie Wang
Object DetectionTransformerPrompt EngineeringContrastive LearningImageText
🎯 What it does: Propose PET-DINO, a universal object detection framework that simultaneously supports text prompts and visual prompts, modified from Grounding DINO with added modules including Alignment-Friendly Visual Prompt Generation (AFVPG), Intra-Batch Parallel Prompting (IBP), and Dynamic Memory-Driven Prompting (DMD).
PETAR: Localized Findings Generation with Mask-Aware Vision-Language Modeling for PET Automated Reporting
Danyal Maqbool (University of Wisconsin-Madison), Tyler J. Bradshaw (Microsoft)
SegmentationGenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelMultimodalityBiomedical DataComputed TomographyPositron Emission Tomography
🎯 What it does: Created the first large-scale PET/CT dataset, PETARSeg-11K, and developed a 3D mask-aware vision-language model, PETAR-4B, based on this dataset to generate accurate tumor localization reports.
PFGNet: A Fully Convolutional Frequency-Guided Peripheral Gating Network for Efficient Spatiotemporal Predictive Learning
Xinyong Cai (Sichuan University), Yuankai Wu (Sichuan University)
Computational EfficiencyConvolutional Neural NetworkVideoTime Series
🎯 What it does: Propose a fully convolutional PFGNet that achieves adaptive receptive fields via pixel-level frequency-guided peripheral frequency gating, addressing spatial and temporal modeling in spatiotemporal prediction learning.
PG-VTON: Single-Pass Training-Free Virtual Try-On via Patch-Guided Reference Alignment
Guohao Zhao (Peking University), Yuxin Peng (Peking University)
Image HarmonizationTransformerDiffusion modelImage
🎯 What it does: Proposes a single-channel, training-agnostic virtual try-on framework called PG-VTON, which can naturally fit reference clothing onto a target person in a single diffusion inference.
PGA: Prior-free Generative Attack for Practical No-box Scenario
Hongyu Peng (Northwestern Polytechnical University), Gong Cheng (Northwestern Polytechnical University)
Adversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: In practical black-box attack scenarios with no prior information and only a small number of unlabeled images, we propose a prior-agnostic generative attack method called PGA to generate transferable adversarial examples.
PGR-Net: Prior-Guided ROI Reasoning Network for Brain Tumor MRI Segmentation
Jiacheng Lu, Guoping Huo (Capital Normal University)
SegmentationTransformerBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose a ROI Reasoning Network (PGR-Net) based on spatial prior for precise segmentation of brain tumor MRI.
pH-Strips for Selective Forgetting: A Blunt but Fast Diagnostic Baseline for Machine Unlearning
Chengyao Qian (Monash University), Mehrtash Harandi (Monash University)
ClassificationGenerationSafty and PrivacyComputational EfficiencyImageTextMultimodality
🎯 What it does: Proposes a training-free and reserved-set-free machine forgetting method called MUpHT, which achieves rapid 'forgetting' by eliminating low-dimensional subspaces in the model weight space associated with undesirable concepts through closed-form projection.
PHAC: Promptable Human Amodal Completion
Seung Young Noh (Kwangwoon University), Ju Yong Chang (Kwangwoon University)
RestorationGenerationPose EstimationPrompt EngineeringVision Language ModelDiffusion modelAuto EncoderImage
🎯 What it does: Propose a method for generating occluded human images through user point prompts (pose points or bounding boxes), balancing visible appearance preservation and pose alignment.
Phantom: Physical Object Interactions as Dynamic Triggers for NMS-Exploited Backdoors
Tianlin Huo (Tsinghua University), Yuan He (Tsinghua University)
Object DetectionAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Propose a backdoor attack named Phantom, which leverages real-object interaction dynamics and manipulates the NMS process. It can maliciously cause misclassification, mislocalization, object disappearance, or emergence in detection results while maintaining the model's performance on normal inputs.
PHANTOM: Physics-Infused Video Generation via Joint Modeling of Visual and Latent Physical Dynamics
Ying Shen (University of Illinois Urbana Champaign), Ismini Lourentzou (University of Illinois Urbana Champaign)
GenerationDiffusion modelContrastive LearningVideoPhysics Related
🎯 What it does: Propose a Physics-Infused Video Generation (PHANTOM) model that can simultaneously generate video content and potential physical dynamics;
PHASE-Net: Physics-Grounded Harmonic Attention System for Efficient Remote Photoplethysmography Measurement
Bo Zhao, Zitong Yu (Harbin Institute of Technology)
Computational EfficiencyRepresentation LearningConvolutional Neural NetworkVideoBiomedical DataPhysics Related
🎯 What it does: PHASE-Net achieves temporal modeling of remote photoplethysmography (rPPG) through a physics-driven damped resonator model, and proposes a zero-FLOPs ZAS (Zero-FLOPs Axial Swapper) and Adaptive Spatial Filter (ASF) to enhance spatial feature interaction and attention selection.
Phased DMD: Few-step Distribution Matching Distillation via Score Matching within Subintervals
Xiangyu Fan (SenseTime Research), Lei Yang (SenseTime Research)
Knowledge DistillationMixture of ExpertsScore-based ModelImageVideo
🎯 What it does: Propose Phased DMD, a phased multi-step distribution matching distillation framework, for distilling large diffusion models into efficient MoE generators.
PhaseWin Search Framework Enable Efficient Object-Level Interpretation
Zihan Gu (Institute of Information Engineering, Chinese Academy of Sciences), Xiaochun Cao (School of Cyber Science and Technology, Sun Yat-sen University)
Explainability and InterpretabilityComputational EfficiencyImage
🎯 What it does: Propose the PhaseWin algorithm, which achieves efficient region attribution for object-level multimodal models through phase window search;
PhaSR: Generalized Image Shadow Removal with Physically Aligned Priors
Chia-Ming Lee (National Yang Ming Chiao Tung University), Chih-Chung Hsu (National Yang Ming Chiao Tung University)
RestorationDepth EstimationTransformerImage
🎯 What it does: Proposes a dual-layer physical alignment shadow removal framework called PhaSR, which achieves efficient, mask-free shadow removal by leveraging global normalization and cross-modal attention.
PhenoYieldNet: Learning Crop-Aware Phenological Responses for Multi-Crop Yield Prediction
Yu Luo (University of Sydney), Kun Hu (Edith Cowan University)
TransformerContrastive LearningImageMultimodalityTime SeriesAgriculture Related
🎯 What it does: Propose a unified multi-crop yield prediction framework, PhenoYieldNet, capable of learning crop-specific growth cycle responses and associating them with meteorological changes;
Photo-Guided Tooth Segmentation on 3D Oral Scan Model
Shaojie Zhuang (Shandong University), Yuanfeng Zhou (Shandong University)
SegmentationConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningImageMultimodalityMesh
🎯 What it does: Align oral photographs with 3D oral scan models and enhance tooth segmentation accuracy by leveraging texture information from multi-view photographs.
Photo3D: Advancing Photorealistic 3D Generation through Structure-Aligned Detail Enhancement
Xinyue Liang (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
GenerationData SynthesisLarge Language ModelDiffusion modelGaussian SplattingImage
🎯 What it does: Construct a structure-aligned multi-view dataset named Photo3D-MV, and design a detail enhancement scheme to improve the lighting, texture details, and structural consistency of 3D-native generative models.
PhotoFramer: Multi-modal Image Composition Instruction
Zhiyuan You (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences), Zhoutong Zhang (Adobe NextCam)
Image TranslationGenerationTransformerLarge Language ModelReinforcement LearningPrompt EngineeringVision Language ModelDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: Propose PhotoFramer, a multi-modal framework that provides natural language composition guidance along with corresponding example images during the photography process, helping users improve photo composition.
Phrase-grounded APO for Improving Chest X-ray Report Generation
Raziuddin Mahmood (Rensselaer Polytechnic Institute), Tanveer Syeda-Mahmood (Stanford University)
GenerationOptimizationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningImageTextBiomedical Data
🎯 What it does: Propose an automatic preference optimization (APO) for chest X-ray report generation models during the inference phase, which automatically generates pairs of superior and inferior examples through a fact-checking model and LLM error correction, combined with lightweight fine-tuning of model parameters using a phrase localization loss.
Phrase-Grounding-Aware Supervised Fine-Tuning for Chart Recognition via Side-Masked Attention
Koichiro Ito (Japan Aerospace Exploration Agency)
RecognitionTransformerSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: Proposes a chart recognition method that introduces phrase alignment supervision during the fine-tuning stage of visual language models.
PhyCo: Learning Controllable Physical Priors for Generative Motion
Sriram Narayanan (Carnegie Mellon University), Manmohan Chandraker (NEC Labs America)
GenerationData SynthesisReinforcement LearningVision Language ModelDiffusion modelVideoPhysics Related
🎯 What it does: Designed a physics-prior-based controllable video generation framework called PhyCo, which can generate physically consistent and controllable motion videos based on input physical attributes such as friction, recovery, deformation, and external forces.
PhyCritic: Multimodal Critic Models for Physical AI
Tianyi Xiong (NVIDIA), Zhiding Yu (NVIDIA)
Reinforcement Learning from Human FeedbackReinforcement LearningMultimodalityBenchmarkPhysics Related
🎯 What it does: This paper proposes PhyCritic, a multimodal evaluation model designed for physical AI tasks.
PhyGaP: Physically-Grounded Gaussians with Polarization Cues
Jiale Wu (Zhejiang University), Yifan Peng (The University Of Hong Kong)
GenerationGaussian SplattingImagePhysics Related
🎯 What it does: Propose a physics-based 3D Gaussian splatting method called PhyGaP, which utilizes polarization information to achieve reflection decomposition and relightable rendering of glossy objects.
PhyOceanCast: Global Ocean Forecasting with Physics-Informed Diffusion
Qixiu Li, Xiaolong Xu (Nanjing University of Information Science and Technology)
Recurrent Neural NetworkGraph Neural NetworkDiffusion modelTime SeriesPhysics Related
🎯 What it does: Proposed a physics-informed diffusion model called PhyOceanCast for global ocean prediction, capable of performing probabilistic spatiotemporal forecasting on a sphere for 145 ocean variables.
PhysGaia: A Physics-aware Benchmark with Multi-Body Interactions for Dynamic Novel View Synthesis
Mijeong Kim (Seoul National University), Bohyung Han (Seoul National University)
GenerationData SynthesisGaussian SplattingImageVideoBenchmarkPhysics Related
🎯 What it does: Propose PhysGaia, a dynamic novel view synthesis benchmark that includes multi-object, multi-material (liquid, gas, viscoelastic, textile) physics simulations.
PhysGen: Physically Grounded 3D Shape Generation for Industrial Design
Yingxuan You (EPFL), Pascal Fua (EPFL)
GenerationOptimizationTransformerFlow-based ModelRectified FlowAuto EncoderMeshPhysics Related
🎯 What it does: Propose PhysGen, a 3D shape generation framework that combines flow matching with physics-guided methods, capable of meeting specified aerodynamic performance while maintaining geometric feasibility.
PhysGM: Large Physical Gaussian Model for Feed-Forward 4D Synthesis
Chunji Lv (Beijing Institute of Technology), Changsheng Li (Beijing Institute of Technology)
GenerationData SynthesisTransformerGaussian SplattingImageVideoMultimodalityPhysics Related
🎯 What it does: This paper proposes an end-to-end forward inference framework called PhysGM, which can directly predict 3D Gaussian Splatting (3DGS) and its physical properties (e.g., elastic modulus, Poisson's ratio) from a single image. These parameters are then used in Material Point Method (MPM) physics simulations to generate a realistic 4D animation;
PhysGS: Bayesian-Inferred Gaussian Splatting for Physical Property Estimation
Samarth Chopra (University of Maryland), Dinesh Manocha (University of Maryland)
SegmentationVision Language ModelGaussian SplattingImageTextPhysics Related
🎯 What it does: Proposes PhysGS, a 3D Gaussian Splatting method based on Bayesian inference, for inferring dense physical properties (friction, hardness, density, elasticity, etc.) from RGB images and vision-language priors, along with uncertainty modeling.
PhysHead: Simulation-Ready Gaussian Head Avatars
Berna Kabadayi (Max Planck Institute for Intelligent Systems), Gerard Pons-Moll (University of Tbingen)
GenerationData SynthesisGaussian SplattingImageVideoPhysics Related
🎯 What it does: Construct animatable head avatars using multi-view videos, and achieve separated control of dynamic hairstyles and facial expressions through physics simulation;
PhysHO: Physics-Based Dynamic 3D Gaussian Human and Object from Monocular Video
Suyi Jiang (National University of Singapore), Gim Hee Lee (National University of Singapore)
GenerationOptimizationGaussian SplattingOptical FlowVideoPhysics Related
🎯 What it does: This paper proposes PhysHO, a framework for reconstructing physically plausible human-object dynamic interactions from monocular videos, combining SMPL-driven linear blend skinning (LBS) with Material Point Method (MPM) simulation, achieving dynamic reconstruction under physical constraints and future motion prediction.
Physical Adversarial Clothing Evades Visible-Thermal Detectors via Non-Overlapping RGB-T Pattern
Xiaopei Zhu (Tsinghua University), Xiaolin Hu (Tsinghua University)
Object DetectionAdversarial AttackMultimodality
🎯 What it does: This paper designs a non-overlapping visible-thermal infrared pattern (NORP) physical adversarial clothing, leveraging 3D full-perspective RGB-T modeling and space discrete-continuous optimization (SDCO) to realize wearable adversarial attacks, successfully deceiving multiple RGB-T object detectors with fused architectures in both digital and real-world scenarios.
Physical Object Understanding with a Physically Controllable World Model
Rahul Venkatesh (Stanford University), Daniel LK Yamins (Stanford University)
Robotic IntelligenceTransformerLarge Language ModelWorld ModelOptical FlowImageVideoPhysics Related
🎯 What it does: Construct a probabilistic world model based on autoregressive sequence models, capable of predicting the distribution of visual variables in scenarios under arbitrary local variable conditions and generating controllable physical future states from it.
Physical Simulator In-the-Loop Video Generation
Lin Geng Foo (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)
GenerationDiffusion modelOptical FlowVideoTextMeshPhysics Related
🎯 What it does: By embedding a physics simulator into the text-to-video diffusion generation loop, the generated videos adhere to real-world physical laws while maintaining high visual quality.
Physically Ground Commonsense Knowledge for Articulated Object Manipulation with Analytic Concepts
Jiude Wei (Shanghai Jiao Tong University), Jianhua Sun (Shanghai Jiao Tong University)
Robotic IntelligenceTransformerLarge Language ModelVision-Language-Action ModelGenerative Adversarial NetworkMultimodalityPoint CloudMesh
🎯 What it does: Propose converting semantic-level common-sense knowledge generated by multi-modal large language models (MLLMs) into computable knowledge at the physical level through 'analytic concepts,' enabling more precise manipulation of articulated objects;
Physically Inspired Gaussian Splatting for HDR Novel View Synthesis
Huimin Zeng (Northeastern University), Yun Fu (Northeastern University)
GenerationNeural Radiance FieldGaussian SplattingImagePhysics Related
🎯 What it does: Propose the PhysHDR-GS framework, combining an image exposure branch and a Gaussian illumination branch, using 3D Gaussian Splatting for HDR novel view synthesis from multi-exposure LDR images.
Physically-Grounded Turbulence Mitigation with Frame-Shared Degradation Parameters
Dongxin Xie (South China University of Technology), Hui Ji (National University of Singapore)
RestorationImagePhysics Related
🎯 What it does: Propose an unsupervised multi-frame atmospheric turbulence image recovery method called TMFS, which utilizes the physical model of tilt-then-blur and models distortion and blur parameters as frame-related stochastic processes, significantly improving recovery performance.
Physics-Consistent Diffusion for Efficient Fluid Super-Resolution via Multiscale Residual Correction
Zhihao Li (Hong Kong University of Science and Technology (Guangzhou)), Guangtao Zhang (SandGold AI Research)
Super ResolutionDiffusion modelPhysics Related
🎯 What it does: Propose a fluid super-resolution method ReMD based on physical consistency diffusion, enhancing high-frequency details through multi-scale residual correction.
Physics-Guided Multistep Deformation Reversal for Ancient Bamboo Slip Restoration
Qianqian Tang (Wuhan University), Yongchao Xu (Wuhan University)
RestorationConvolutional Neural NetworkImagePhysics Related
🎯 What it does: Propose a physics-guided multi-step inverse deformation framework for digital restoration of deformed bamboo slips.
PhysInOne: Visual Physics Learning and Reasoning in One Suite
Siyuan Zhou (Hong Kong Polytechnic University), Bo Yang (Hong Kong Polytechnic University)
Supervised Fine-TuningDiffusion modelNeural Radiance FieldOptical FlowVideoBenchmarkPhysics Related
🎯 What it does: Constructed and released the PhysInOne dataset, containing 153,810 3D scenes with multi-object, multi-background, and 71 fundamental physical phenomena, along with 2 million dynamic videos and complete geometric, semantic, motion, physical attributes, and textual annotations; subsequently conducted baseline evaluations on four tasks (video generation, future frame prediction, physical attribute estimation, and motion transfer), and improved the physical consistency of large models through fine-tuning.
PhysIR-Splat: Physically Consistent Thermal Infrared Radiative Transfer in 3D Gaussian Splatting
Jingyuan Gao (Xidian University), Mingjin Zhang (Xidian University)
GenerationData SynthesisTransformerNeural Radiance FieldGaussian SplattingImageMultimodalityPhysics Related
🎯 What it does: This paper proposes the PhysIR Splat framework, combining 3D Gaussian Splatting with thermal infrared radiative transfer models, achieving physically consistent thermal infrared 3D reconstruction and novel view synthesis.
PhysSkin: Real-Time and Generalizable Physics-Based Animation via Self-Supervised Neural Skinning
Yuanhang Lei (Zhejiang University), Zhaopeng Cui (Zhejiang University)
GenerationTransformerPoint CloudMeshPhysics Related
🎯 What it does: Developed a real-time, cross-shape, cross-discretization physics-driven neural skinning framework called PhysSkin for generating continuous, orthogonal, physically consistent skinning fields.
PhysVid: Physics Aware Local Conditioning for Generative Video Models
Saurabh Pathak, Bahram Zonooz
GenerationTransformerVision Language ModelFlow-based ModelVideoTextPhysics Related
🎯 What it does: Propose PhysVid, which utilizes video clip-level physical descriptions as local conditions to guide the model to better adhere to physical laws during video generation.
PhysX-Anything: Simulation-Ready Physical 3D Assets from Single Image
Ziang Cao (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelDiffusion modelImageTextMeshBenchmarkPhysics Related
🎯 What it does: Proposed PhysX-Anything, a single-image generation system based on VLM that produces simulation-ready physical 3D assets, capable of generating URDF/XML files with geometry, joints, and physical properties.
Pico-Banana-400K: A Large-Scale Dataset for Text-Guided Image Editing
Yusu Qian (Apple), Zhe Gan (Apple)
Image TranslationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Constructed a text-guided image editing dataset named Pico-Banana-400K with approximately 400K entries, utilizing Nano-Banana to perform various edits on real images from OpenImages and automatically evaluate quality.
PiLoT: Neural Pixel-to-3D Registration for UAV-based Ego and Target Geo-localization
Xiaoya Cheng (National University of Defense Technology), Shen Yan (National University of Defense Technology)
Pose EstimationOptimizationConvolutional Neural NetworkSimultaneous Localization and MappingVideoPoint CloudMesh
🎯 What it does: A unified framework named PiLoT was constructed, directly aligning the real-time video stream of UAVs with a global 3D map through pixel-to-3D registration, achieving drift-free pose estimation for UAVs and geolocation regression for arbitrary target pixels.
PinPoint: Evaluation of Composed Image Retrieval with Explicit Negatives, Multi-Image Queries, and Paraphrase Testing
Rohan Mahadev (Pinterest), Dmitry Kislyuk (Pinterest)
RetrievalLarge Language ModelVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: Propose PinPoint, a large-scale zero-shot synthetic image retrieval benchmark covering multi-answer, explicit hard negatives, multi-image queries, and sentence diversity.
Pip-Stereo: Progressive Iterations Pruner for Iterative Optimization based Stereo Matching
Jintu Zheng (Aridge), Zhuojie Chen (Aridge)
Depth EstimationOptimizationComputational EfficiencyNeural Architecture SearchRecurrent Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Proposed Pip-Stereo, combining iterative compression optimization, collaborative monocular depth prior transfer, and hardware-aware FlashGRU, achieving real-time high-precision stereo matching on edge devices.
PIX-TAB: Efficient PIXel-Precise TABle Structure Recognition Approach with Speculative Decoding and Region-Based Image Segmentation
Viktor Zaytsev (Samsung R&D Institute Ukraine), Artem Shcherbina (BrilliantFlux)
RecognitionSegmentationData SynthesisComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose a lightweight end-to-end table structure recognition method called PIX-TAB, which can parse table structures on mobile devices with pixel-level accuracy and support multiple languages;
PixARMesh: Autoregressive Mesh-Native Single-View Scene Reconstruction
Xiang Zhang (University Of California San Diego), Zhuowen Tu (University Of California San Diego)
GenerationTransformerImagePoint CloudMesh
🎯 What it does: Proposed a single-view indoor scene reconstruction framework called PixARMesh, which directly generates complete 3D meshes in an autoregressive manner, avoiding the need for SDF decoding and layout optimization;
PixDLM: A Dual-Path Multimodal Language Model for UAV Reasoning Segmentation
Shuyan Ke (Xiamen University), Rongrong Ji (Xiamen University)
SegmentationLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Proposed the UAV Reasoning Segmentation task, constructed a large-scale high-resolution dataset named DRSeg, and designed the dual-path pixel-level multimodal language model PixDLM as the baseline.
Pixel Motion Diffusion is What We Need for Robot Control
E-Ro Nguyen (Stony Brook University), Michael S. Ryoo (Stony Brook University)
Robotic IntelligenceTransformerVision-Language-Action ModelDiffusion modelOptical FlowMultimodality
🎯 What it does: This paper proposes DAWN, a two-stage diffusion framework that bridges language instructions with robot low-level actions by leveraging explicit pixel motion representation, achieving efficient robot manipulation under language conditions.
Pixel2Phys: Distilling Governing Laws from Visual Dynamics
Ruikun Li (Tsinghua University), Yan Lu (Shanghai Artificial Intelligence Laboratory)
Explainability and InterpretabilityLarge Language ModelAuto EncoderVideoPhysics RelatedOrdinary Differential Equation
🎯 What it does: Propose a multi-agent based Pixel2Phys framework that automatically extracts physical variables from high-dimensional videos and derives interpretable governing equations, covering three types of visual dynamics: object motion, field evolution, and complex experimental videos.
PixelDiT: Pixel Diffusion Transformers for Image Generation
Yongsheng Yu (NVIDIA), Jiebo Luo (University of Rochester)
GenerationTransformerDiffusion modelRectified FlowImageMultimodality
🎯 What it does: Propose PixelDiT, a Diffusion Transformer trained and sampled entirely in pixel space, achieving end-to-end learning without an Autoencoder.
PixelRush: Ultra-Fast, Training-Free High-Resolution Image Generation via One-step Diffusion
Hong-Phuc Lai (Qualcomm Vietnam Company Limited), Anh Tran (Qualcomm Vietnam Company Limited)
GenerationDiffusion modelImageText
🎯 What it does: Propose the PixelRush framework to achieve ultra-high-resolution image generation without optimization and single-step diffusion;
Pixels Don't Lie (But Your Detector Might): Bootstrapping MLLM-as-a-Judge for Trustworthy Deepfake Detection and Reasoning Supervision
Kartik Kuckreja (Mohamed bin Zayed University of Artificial Intelligence), Abhinav Dhall (Monash University)
Anomaly DetectionTransformerVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: Proposes the DeepfakeJudge framework, integrating an OOD visual deception detection dataset, human-annotated visual reasoning labels, and a multi-modal judge based on a generate-evaluate cycle, to supervise and evaluate the inference quality of deepfake detection models on a large scale.
PLACID: Identity-Preserving Multi-Object Compositing via Video Diffusion with Synthetic Trajectories
Gemma Canet Tarrés (Amazon), Yumeng Li (Amazon)
GenerationData SynthesisTransformerDiffusion modelFlow-based ModelImageVideoBenchmark
🎯 What it does: Propose the PLACID framework to achieve seamless synthesis of multiple objects while preserving identity, color, and background.
PlanaReLoc: Camera Relocalization in 3D Planar Primitives via Region-Based Structure Matching
Hanqiao Ye (University of Chinese Academy of Sciences), Shuhan Shen (University of Chinese Academy of Sciences)
Pose EstimationDepth EstimationTransformerImageMesh
🎯 What it does: This paper proposes a camera localization method called PlanaReLoc based on planar primitives, which achieves 6-DoF camera pose estimation by cross-modal matching between planes recovered from a single image and texture-free 3D plane maps.
PlannerRFT: Reinforcing Diffusion Planners through Closed-Loop and Sample-Efficient Fine-Tuning
Hongchen Li (Tongji University), Hongyang Li (University of Hong Kong)
Autonomous DrivingOptimizationComputational EfficiencyTransformerReinforcement LearningDiffusion modelScore-based ModelSequentialBenchmark
🎯 What it does: Fine-tune a diffusion-based autonomous driving trajectory planner using reinforcement learning, propose a dual-branch strategy for guided denoising and scene-adaptive exploration, and build a GPU parallel simulator, nuMax, to accelerate closed-loop training.
Planning in 8 Tokens: A Compact Discrete Tokenizer for Latent World Model
Dongwon Kim (KAIST), Suha Kwak (POSTECH)
CompressionTransformerAuto EncoderContrastive LearningWorld ModelImageVideo
🎯 What it does: Propose an extremely compressed discrete image tokenizer, CompACT, which compresses each image into 8-16 tokens, and trains a world model in this low-dimensional space to enable fast decision-making planning.
Plant Taxonomy Meets Plant Counting: A Fine-Grained, Taxonomic Dataset for Counting Hundreds of Plant Species
Jinyu Xu (Huazhong University of Science and Technology), Hao Lu (Huazhong University of Science and Technology)
Object DetectionConvolutional Neural NetworkTransformerPrompt EngineeringImageBenchmarkAgriculture Related
🎯 What it does: Constructed the TPC-268 dataset containing 268 countable plant species, 678,050 point annotations, and complete Latin taxonomic hierarchies, and benchmarked multiple multi-class unsupervised counting models
Plenoptic Video Generation
Xiao Fu (Nvidia), Chen-Hsuan Lin (Nvidia)
GenerationData SynthesisRobotic IntelligenceTransformerDiffusion modelVideo
🎯 What it does: Proposed the PlenopticDreamer framework, achieving generative rerendering of videos with camera control under multi-view settings while maintaining long-term spatiotemporal consistency.
Plug-and-Play Incomplete Multi-View Clustering via Janus-Faced Affinity Learning with Topology Harmonization
Shengju Yu (National University of Defense Technology), Xinwang Liu (National University of Defense Technology)
OptimizationRepresentation LearningGraph Neural NetworkMultimodality
🎯 What it does: Propose an end-to-end framework named PJFTH for incomplete multi-view clustering, which utilizes Janus-faced affinity learning to suppress view-specific noise and achieves unified anchor point ordering through topological harmonization, ultimately obtaining a high-quality consensus similarity matrix.
Plug-and-Play PDE Optimization for 3D Gaussian Splatting: Toward High-Quality Rendering and Reconstruction
Yifan Mo (University of Science and Technology of China), Ligang Liu (University of Science and Technology of China)
OptimizationGaussian SplattingPoint CloudOrdinary Differential Equation
🎯 What it does: Propose a PDE-based optimization component called PDEO for stabilizing the optimization of 3D Gaussian point clouds, achieving high-quality view synthesis and surface reconstruction.
Pluggable Pruning with Contiguous Layer Distillation for Diffusion Transformers
Jian Ma (OPPO AI Center), Haonan Lu (OPPO AI Center)
GenerationComputational EfficiencyKnowledge DistillationTransformerDiffusion modelMultimodality
🎯 What it does: A pluggable structured pruning framework named PPCL is proposed for Diffusion Transformer (DiT), which can flexibly prune depth and width without requiring retraining.
PMRNet: Physics-informed Multi-scale Refinement Network for Medical Image Segmentation
Boce Kang (Northwestern Polytechnical University)
SegmentationConvolutional Neural NetworkImageBiomedical DataPhysics Related
🎯 What it does: Proposes PMRNet, a lightweight multi-scale improved network designed specifically for medical image segmentation
PnP-CM: Consistency Models as Plug-and-Play Priors for Inverse Problems
Merve Gulle (University of Minnesota), Mehmet Akcakaya (University of Minnesota)
RestorationDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose a framework (PnP-CM) that uses consistency models as a Plug-and-Play prior to solve both linear and nonlinear inverse problems.
POCA: Pareto-Optimal Curriculum Alignment for Visual Text Generation
Yaohou Fan (Tohoku University), Shinichiro Omachi (Tohoku University)
GenerationOptimizationReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Propose a Pareto-optimal multi-objective reinforcement learning framework called POCA, which jointly uses text accuracy, image alignment, and aesthetic rewards for post-training in visual text generation;
POGA: Paraphrased and Oppositional Graph Alignment for Fine-Grained Cross-Modal Retrieval
Junfeng Zhang (Beijing University of Posts and Telecommunications), Ming-Hsuan Yang (UC Merced)
RetrievalGraph Neural NetworkLarge Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposed a framework named POGA to address the issues of fine-grained alignment and fact discrimination in long-text cross-modal retrieval.
PoInit-of-View: Poisoning Initialization of Views Transfers Across Multiple 3D Reconstruction Systems
Weijie Wang (University of Trento), Bruno Lepri (Fondazione Bruno Kessler)
Adversarial AttackGaussian SplattingImage
🎯 What it does: Propose the PoInit-of-View attack, which adversarially corrupts input views during the initialization phase of structured light reconstruction in 3D reconstruction systems.
Point Cloud as a Foreign Language for Multi-modal Large Language Model
Sneha Paul (Concordia University), Nizar Bouguila (Concordia University)
ClassificationGenerationTransformerLarge Language ModelReinforcement LearningMultimodalityPoint Cloud
🎯 What it does: Propose a multi-modal large language model called SAGE, which is end-to-end and does not rely on pre-trained 3D encoders. It can directly convert raw point clouds into discrete tokens and input them into LLMs for natural language generation and reasoning.
Point4Cast: Streaming Dynamic Scene Reconstruction and Forecasting
Xinhang Liu (Hong Kong University of Science and Technology), Moitreya Chatterjee (Mitsubishi Electric Research Laboratories)
GenerationRepresentation LearningTransformerVideoPoint Cloud
🎯 What it does: Proposes a unified framework called Point4Cast that can real-time reconstruct and predict dynamic 3D scene point clouds and camera parameters from continuous 2D video streams.
PointAlign: Feature-Level Alignment Regularization for 3D Vision-Language Models
Yuanhao Su (University Of Science And Technology Of China), Qi Fan (University Of Science And Technology Of China)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningTextPoint Cloud
🎯 What it does: This paper proposes PointAlign, a regularization method for feature-level alignment of intermediate point cloud tokens in 3D vision-language models.