CVPR 2026 Papers — Page 27
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
PointCNN++: Performant Convolution on Native Points
Lihan Li (Peking University), Yangyan Li (Ant Group)
SegmentationComputational EfficiencyConvolutional Neural NetworkPoint Cloud
🎯 What it does: Propose PointCNN++, an efficient operator that performs convolution directly on raw point clouds, addressing the issues of precision loss caused by voxelization and low computational efficiency in point cloud methods.
PointCSP: Cross-Sample Semantic Propagation and Stability Preservation in Self-Supervised Point Cloud Learning
Xinxing Yu (Macau University of Science and Technology), Yanyan Liang (Macau University of Science and Technology)
SegmentationKnowledge DistillationRecurrent Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: Proposing a self-supervised point cloud learning framework that combines cross-sample semantic propagation with semantic-preserving distillation to construct a unified and transferable semantic space
Pointer-CAD: Unifying B-Rep and Command Sequences via Pointer-based Edges & Faces Selection
Dacheng Qi (Transcengram), Shenghua Gao (Transcengram)
GenerationGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextMeshGraph
🎯 What it does: Utilizing large language models (LLM) to generate CAD models through pointer-based command sequences enables explicit selection of B-rep edges/faces during generation, achieving advanced editing operations such as chamfer and fillet, while ensuring semantic consistency and topological integrity through multi-step generation.
PointGS: Semantic-Consistent Unsupervised 3D Point Cloud Segmentation with 3D Gaussian Splatting
Yixiao Song (Beijing Jiaotong University), Zhicheng Yan (Beijing Jiaotong University)
SegmentationContrastive LearningGaussian SplattingPoint Cloud
🎯 What it does: Construct a unified intermediate representation using 3D Gaussian Splatting to achieve unsupervised 3D point cloud semantic segmentation.
Pointing at Parts: Training-Free Few-Shot Grounding in Multimodal LLMs
Shiang-Feng Tsai (National Tsing Hua University), Min Sun (National Tsing Hua University)
SegmentationTransformerLarge Language ModelVision Language ModelContrastive LearningImage
🎯 What it does: Propose a training-free few-shot part-level localization method called POP to enhance the precise pointing capability of multimodal large language models (MLLMs) on object parts.
PointNSP: Autoregressive 3D Point Cloud Generation with Next-Scale Level-of-Detail Prediction
Ziqiao Meng (National University of Singapore), Peilin Zhao (Shanghai Jiao Tong University)
GenerationTransformerAuto EncoderPoint Cloud
🎯 What it does: This paper proposes PointNSP, a self-recursive 3D point cloud generation framework based on coarse-to-fine level-of-detail (LoD) prediction;
POINTS-Long: Adaptive Dual-Mode Visual Reasoning in MLLMs
Haicheng Wang (Shanghai Jiao Tong University), Yanfeng Wang (Tencent)
CompressionComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoText
🎯 What it does: Proposes POINTS-Long, a dual-mode multi-modal large language model capable of switching between high precision (focus mode) and efficiency (standby mode), addressing computational bottlenecks in long video visual reasoning.
Points-to-3D: Structure-Aware 3D Generation with Point Cloud Priors
Jiatong Xia (University of Adelaide), Lingqiao Liu (University of Adelaide)
GenerationDiffusion modelFlow-based ModelAuto EncoderPoint Cloud
🎯 What it does: Proposed a diffusion framework called Points-to-3D, which achieves geometry-controllable 3D generation by leveraging visible point cloud priors, directly mapping point clouds to complete 3D models within a sparse structural latent space.
PointThinker: Point-Incentivized Parallel Thinking for Multimodal Large Language Model
Zhengdong Hu (University of Technology Sydney), Yi Yang (Zhejiang University)
TransformerSupervised Fine-TuningReinforcement LearningVision Language ModelImageVideoMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose the PointThinker framework, which utilizes a point-induced parallel thinking approach to enable multimodal large language models (MLLMs) to first list key visual points in visual reasoning tasks, then conduct separate reasoning for each point, and finally synthesize all reasoning to derive an answer.
PointTPA: Dynamic Network Parameter Adaptation for 3D Scene Understanding
Siyuan Liu (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
SegmentationSupervised Fine-TuningPoint Cloud
🎯 What it does: In the 3D scene-level point cloud semantic segmentation task, a test-time parameter adaptation framework called PointTPA is introduced, dynamically generating input-aware network parameters;
PointWorld: Scaling 3D World Models for In-The-Wild Robotic Manipulation
Wenlong Huang (Stanford University), Li Fei-Fei (Stanford University)
Robotic IntelligenceTransformerWorld ModelPoint Cloud
🎯 What it does: Built and trained a large-scale 3D world model called POINTWORLD, which can predict panoramic 3D point flows from a single RGB-D image and a robot's low-level action sequence, enabling zero-shot field robot manipulation without demonstrations.
POLAR: A Portrait OLAT Dataset and Generative Framework for Illumination-Aware Face Modeling
Zhuo Chen (Shanghai Jiao Tong University), Yichao Yan (Shanghai Jiao Tong University)
GenerationData SynthesisFlow-based ModelAuto EncoderImage
🎯 What it does: Proposed and released the large-scale, physically calibrated Portrait OLAT (POLAR) dataset, and developed a flow-based generative model POLARNet based on this dataset, enabling the rapid synthesis of OLAT responses under arbitrary lighting directions from a single uniformly illuminated portrait, achieving controllable and physically consistent facial relighting.
PolarGuide-GSDR: 3D Gaussian Splatting Driven by Polarization Priors and Deferred Reflection for Real-World Reflective Scenes
Derui Shan (North China University of Technology), Peng Lu (Beijing University of Posts and Telecommunications)
GenerationOptimizationGaussian SplattingImagePhysics Related
🎯 What it does: Designed a 3D Gaussian Splatting framework based on polarization information, achieving high-quality novel view synthesis for complex reflective scenes under real-time rendering.
Polarization State Tracing for Reflection Removal and Color-Consistent Reconstruction
Dongyue Wang (Shenyang Institute of Automation, Chinese Academy of Sciences), Jiandong Tian (Shenyang Institute of Automation, Chinese Academy of Sciences)
RestorationTransformerImagePhysics Related
🎯 What it does: Proposed a new optical model called PSTM based on polarization imaging theory to address reflection and color deviation issues under colored glass, and designed a Transformer network with a physics-guided CRA mechanism to achieve reflection removal and color-consistent reconstruction.
Polyphony: Diffusion-based Dual-Hand Action Segmentation with Alternating Vision Transformer and Semantic Conditioning
Hao Zheng (New York University Abu Dhabi), Tuka Alhanai (New York University Abu Dhabi)
SegmentationTransformerDiffusion modelVideo
🎯 What it does: This study investigates hand action segmentation, proposing the Polyphony three-stage model (alternating trained bimanual Vision Transformer, semantic feature alignment, and diffusion-based segmentation) to address challenges such as mutual hand dependency, visual asymmetry, representation conflicts, and semantic ambiguity.
PolySLGen: Online Multimodal Speaking-Listening Reaction Generation in Polyadic Interaction
Zhi-Yi Lin (Delft University of Technology), Xucong Zhang (Delft University of Technology)
GenerationTransformerLarge Language ModelSupervised Fine-TuningMultimodality
🎯 What it does: Propose PolySLGen, an online multimodal speaking-listening response generation framework capable of generating future speech, body motion, and speaking state scores for target participants based on past multi-party interaction information.
Portable Active Learning for Object Detection
Rashi Sharma (Panasonic R&D Center Singapore), Karthikk Subramanian (Panasonic R&D Center Singapore)
Object DetectionTransformerImage
🎯 What it does: This paper proposes a portable active learning framework called PAL, which selects the most valuable unlabeled images for annotation by leveraging instance-level logistic uncertainty (LIUS) and image-level global uncertainty and diversity estimation (GUIDE) based solely on inference outputs, without modifying the detector or training pipeline.
PortraitDirector: A Hierarchical Disentanglement Framework for Controllable and Real-time Facial Reenactment
Chaonan Ji (Tongyi Lab), Bang Zhang (Tongyi Lab)
Image TranslationGenerationTransformerDiffusion modelAuto EncoderVideo
🎯 What it does: Propose the PortraitDirector framework to achieve controllable real-time facial reenactment.
Pose-Free Omnidirectional Gaussian Splatting for 360-Degree Videos with Consistent Depth Priors
Chuanqing Zhuang (University of Chinese Academy of Sciences), Jun Xiao (University of Chinese Academy of Sciences)
Pose EstimationDepth EstimationGaussian SplattingVideo
🎯 What it does: This paper proposes a pose-free global spherical 3D Gaussian Splatting framework for reconstructing high-quality scenes and camera poses from unlabelled 360-degree videos.
Pose-guided Enriched Feature Learning for Federated-by-camera Person Re-identification
JooHyung Oh (Ulsan National Institute of Science and Technology), Jae-Young Sim (Ulsan National Institute of Science and Technology)
Pose EstimationRetrievalFederated LearningKnowledge DistillationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This study addresses the single-camera scenario in federated person re-identification by proposing a pose-guided feature enhancement framework. It employs a Pose-Extraction Module (PEM) to decompose local features into pose-related and unrelated components, and synthesizes multi-pose features through pose swapping, significantly enhancing contrastive learning performance;
PoseAnything: General Pose-guided Video Generation with Part-aware Temporal Coherence
Ruiyan Wang (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderVideo
🎯 What it does: Proposed the PoseAnything framework, achieving general pose-driven video generation that supports arbitrary (including non-human) skeleton inputs, and enhances temporal coherence through the Part-aware Temporal Coherence Module; also, it first实现了 the decoupling of subject motion and camera motion control via CFG.
PoseD-Flow: Versatile and Guided Flow Matching Model of Human Pose
Jebastin Nadar (Imperial College London), Tolga Birdal (Imperial College London)
GenerationPose EstimationFlow-based ModelImageOrdinary Differential Equation
🎯 What it does: This paper proposes the PoseD-Flow framework, achieving unsupervised inverse generation in the 3D human pose space;
PoseGAM: Robust Unseen Object Pose Estimation via Geometry-Aware Multi-View Reasoning
Jianqi Chen (King Abdullah University of Science and Technology), Peter Wonka (King Abdullah University of Science and Technology)
Pose EstimationTransformerImageMultimodalityPoint Cloud
🎯 What it does: PoseGAM predicts the 6D pose of a target object directly from a query image and several template images with known poses through a multi-view feedforward network, without requiring explicit feature matching;
PoseGaussian: 6D Pose Estimation for Unseen Objects via Sparse-View Object-Level 3D Gaussian Splatting
Wubin Shi (Southeast University), Feipeng Da (Southeast University)
Pose EstimationGaussian SplattingImagePoint Cloud
🎯 What it does: Propose PoseGaussian, which initializes 3D Gaussian Splatting (3DGS) for object-level reconstruction using sparse viewpoint RGB-D image priors, and achieves 6D pose estimation based on this.
PoseMaster: A Unified 3D Native Framework for Stylized Pose Generation
Hongyu Yan (Hong Kong University of Science and Technology), Ping Tan (Hong Kong University of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderPoint CloudMesh
🎯 What it does: Propose the PoseMaster framework, unifying pose stylization and 3D generation into an end-to-end process, directly utilizing 3D skeletons to guide the generation of high-fidelity 3D meshes.
PositionIC: Unified Position and Identity Consistency for Image Customization
Junjie Hu (Meituan), Wenqiang Zhang (Fudan University)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelNeural Radiance FieldAuto EncoderImageTextMultimodalityBenchmark
🎯 What it does: Propose the PositionIC framework to achieve high-fidelity identity consistency and fine-grained spatial controllability during multi-subject image customization.
Post-training Feature Pruning for Fundus Images Classification
Van-Nguyen Pham (Sungkyunkwan University), Hyunseung Choo (SKAI X Inc.)
ClassificationComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed a post-training feature pruning framework (Greedy Feature Pruning, GFP), which removes redundant or weak feature dimensions by greedily selecting output feature vectors after the frozen feature extractor.
PosterIQ: A Design Perspective Benchmark for Poster Understanding and Generation
Yuheng Feng (Hong Kong Polytechnic University), Xingxing Zou (Hong Kong Polytechnic University)
ClassificationGenerationLarge Language ModelDiffusion modelImageTextMultimodalityBenchmark
🎯 What it does: Established the PosterIQ benchmark, containing 7,765 poster annotation instances and 822 generation prompts, covering multi-task scenarios such as OCR, font, layout, style, and intent; provides fine-grained evaluation metrics.
PosterOmni: Generalized Artistic Poster Creation via Task Distillation and Unified Reward Feedback
Sixiang Chen (Hong Kong University of Science and Technology), Lei Zhu (Hong Kong University of Science and Technology)
Image TranslationSegmentationGenerationTransformerSupervised Fine-TuningReinforcement LearningMixture of ExpertsDiffusion modelImageTextMultimodality
🎯 What it does: Propose a unified artistic poster generation framework called PosterOmni, which combines two modes: local editing and global creation, and achieves multi-task image-to-poster generation through task distillation and unified reward feedback.
PosterReward: Unlocking Accurate Evaluation for High-Quality Graphic Design Generation
Jianyu Lai (Hong Kong University of Science and Technology), Lei Zhu (Hong Kong University of Science and Technology)
GenerationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningImageMultimodalityBenchmark
🎯 What it does: Proposed the PosterReward reward model for multi-dimensional evaluation and scoring in poster generation tasks;
POUR: A Provably Optimal Method for Unlearning Representation via Neural Collapse
Anjie Le (University of Oxford), J. Alison Noble (University of Oxford)
Safty and PrivacyRepresentation LearningImageTextBiomedical Data
🎯 What it does: Propose a provably optimal representation layer unlearning method called POUR based on projection, which can completely forget specified classes or concepts without retraining.
PowerCLIP: Powerset Alignment for Contrastive Pre-Training
Masaki Kawamura (Institute of Science Tokyo), Rio Yokota (Institute of Science Tokyo)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Propose the PowerCLIP framework, which performs contrastive pre-training by aligning the power set of image regions with text parse trees to enhance the combinatoriality and robustness of vision-language models.
PP-Brep: Few-Shot B-rep Classification with Hybrid Graph Representation
Jiacheng Hao (Northwestern Polytechnical University), Yilei Shi (Northwestern Polytechnical University)
ClassificationMeta LearningGraph Neural NetworkReinforcement LearningContrastive LearningGraph
🎯 What it does: Study few-shot B-rep (3D CAD boundary representation) classification and propose the PP-Brep framework.
PP-OCRv5: A Specialized 5M-Parameter Model Rivaling Billion-Parameter Vision-Language Models on OCR Tasks
Cheng Cui (Baidu Inc.), Yi Liu (Baidu Inc.)
RecognitionObject DetectionConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This work proposes PP-OCRv5, a lightweight OCR system with only 5 M parameters;
PPISP: Physically-Plausible Compensation and Control of Photometric Variations in Radiance Field Reconstruction
Isaac Deutsch (NVIDIA), Zan Gojcic (NVIDIA)
RestorationNeural Radiance FieldGaussian SplattingImagePhysics Related
🎯 What it does: Propose a differentiable, physically plausible ISP processing pipeline and controller to address photometric inconsistencies in multi-view reconstruction caused by camera optical characteristics and image signal processing (ISP).
PPM-CLIP: Probabilistic Prompt Modeling for Generalizable AI-Generated Image Detection
Xinyuan Wang (Xiamen University), Zhihui Liu (Truesight Technology Co. Ltd)
Object DetectionAnomaly DetectionTransformerPrompt EngineeringVision Language ModelContrastive LearningImageBenchmark
🎯 What it does: Proposed a PPM-CLIP framework based on CLIP, utilizing a generative rather than discriminative approach to detect AI-generated images.
PQDT: Pseudo-Query Dual Transformer for Robust Point Cloud Restoration
Haoqing Wu (Mercedes-Benz AG), Jochen Garcke (University of Bonn)
RestorationTransformerPoint Cloud
🎯 What it does: Proposes a unified Pseudo-Query Dual Transformer (PQDT) for point cloud restoration, addressing multiple degradations such as missing data, noise, and deformation.
PR-IQA: Partial-Reference Image Quality Assessment for Diffusion-Based Novel View Synthesis
Inseong Choi (Dongguk University), Soohwan Song (Dongguk University)
GenerationDiffusion modelGaussian SplattingImage
🎯 What it does: This paper proposes the PR-IQA framework, which evaluates the quality of sparsely viewed new views generated by diffusion models using cross-perspective reference images, and introduces quality maps into 3D Gaussian Splatting (3DGS) training, significantly improving the quality of sparse-view 3D reconstruction and novel view synthesis.
PR-MaGIC: Prompt Refinement Via Mask Decoder Gradient Flow For In-Context Segmentation
Minjae Lee (Pohang University of Science and Technology), Won Hwa Kim (Pohang University of Science and Technology)
SegmentationPrompt EngineeringImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Propose a training-agnostic test-time prompt refinement method, PR-MaGIC, which iteratively updates prompts by leveraging the gradient flow from SAM's mask decoder, thereby improving few-shot/one-shot segmentation performance.
Precise Object and Effect Removal with Adaptive Target-Aware Attention
Jixin Zhao (Nanyang Technological University), Shangchen Zhou (Nanyang Technological University)
Image HarmonizationRestorationVision Language ModelDiffusion modelImage
🎯 What it does: Propose the ObjectClear framework to precisely remove the target object and its visual effects such as shadows and reflections.
Predict Before You Explore: Predictive Planning with Specialized Memory for Embodied Question Answering
Bowen Yuan (Nanjing University of Posts and Telecommunications), Bing-Kun Bao (Hefei University of Technology)
Robotic IntelligenceTransformerReinforcement LearningVision Language ModelVision-Language-Action ModelImageTextMultimodality
🎯 What it does: Propose a prediction-driven EQA framework called Pred-EQA, combining hierarchical prediction planning with structured memory to achieve coherent long-term exploration and efficient evidence collection.
Predicting Spatial Transcriptomics from Histology Images via High-Order Multi-Cell Interaction Modeling
Youhan Sun (Sun Yat-sen University), Yuedong Yang (Sun Yat-sen University)
TransformerFlow-based ModelContrastive LearningImageBiomedical Data
🎯 What it does: Predicting spatial transcriptome expression using H&E slice images, the core idea is to explicitly model multi-cell, multi-order interactions to improve the accuracy of microenvironment expression prediction.
Predictive Regularization Against Visual Representation Degradation in Multimodal Large Language Models
Enguang Wang (NKIARI), Ming-Ming Cheng (NKIARI)
Representation LearningTransformerImageMultimodalityBenchmark
🎯 What it does: This paper diagnoses and addresses the degradation of visual representations in multimodal large language models, proposing a Predictive Regularization (PRe) mechanism to maintain the intrinsic integrity of visual features.
Preference-Aligned LoRA Merging: Preserving Subspace Coverage and Addressing Directional Anisotropy
Wooseong Jeong (KAIST), Kuk-Jin Yoon (KAIST)
OptimizationRepresentation LearningTransformerImageTextBenchmark
🎯 What it does: Studying how to merge multiple LoRA adapters to build a general-purpose multi-task model
Prefill-Time Intervention for Mitigating Hallucination in Large Vision-Language Models
Chengsheng Zhang (University of Science and Technology of China), Xinmei Tian (University of Science and Technology of China)
Representation LearningTransformerContrastive LearningMultimodality
🎯 What it does: This paper proposes a prefill-time intervention (Prefill-Time Intervention, PTI), which adjusts key/value vectors on the KV cache of an LVLM in a one-time, modality-aware manner before generation, thereby correcting the deviation between visual and textual representations in advance and significantly reducing the model's hallucination behavior.
Premier: Personalized Preference Modulation with Learnable User Embedding in Text-to-Image Generation
Zihao Wang (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
GenerationTransformerPrompt EngineeringDiffusion modelFlow-based ModelContrastive LearningImageText
🎯 What it does: Design and train learnable user embeddings to achieve personalized control in text-image generation through prompt preference modulation and dispersion loss, and adopt a linear combination strategy for new users.
Preserving Source Video Realism: High-Fidelity Face Swapping for Cinematic Quality
Zekai Luo (Zhejiang University), Chunhua Shen (Zhejiang University)
Image TranslationGenerationTransformerDiffusion modelFlow-based ModelRectified FlowAuto EncoderVideoBenchmark
🎯 What it does: Proposes LIVINGSWAP, a high-fidelity facial replacement model based on video reference and keyframe injection, capable of maintaining identity consistency and visual realism in long videos.
Pressure2Motion: Hierarchical Human Motion Reconstruction from Ground Pressure with Text Guidance
Zhengxuan Li (Nanjing University), Hao Zhu (Nanjing University)
GenerationPose EstimationConvolutional Neural NetworkRecurrent Neural NetworkVision-Language-Action ModelDiffusion modelImageTextBenchmark
🎯 What it does: Utilizing ground pressure sensors and text prompts, achieving full-body 3D motion capture without the need for cameras or wearable devices.
Prime Once, then Reprogram Locally: An Efficient Alternative to Black-Box Service Model Adaptation
Yunbei Zhang (Tulane University), Jihun Hamm (Tulane University)
ClassificationComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerSupervised Fine-TuningVision Language ModelImageText
🎯 What it does: Propose a new service model reprogramming method called AReS, which primes the local encoder through a one-time API call, followed by local visual reprogramming to achieve efficient migration of closed-box service models
PRIMU: Uncertainty Estimation for Novel Views in Gaussian Splatting from Primitive-Based Representations of Error and Coverage
Thomas Gottwald (University of Wuppertal), Matthias Rottmann (University of Osnabruck)
GenerationGaussian SplattingImagePoint Cloud
🎯 What it does: Propose PRIMU, a post-processing Gaussian splat uncertainty estimation framework: extract error and coverage statistics from training views, project them onto primitives to generate orientation-dependent primitive-level uncertainty feature maps, then use gradient boosting regression to obtain pixel-level error prediction, which can be directly applied to active view selection.
Principled Steering via Null-space Projection for Jailbreak Defense in Vision-Language Models
Xingyu Zhu (MoE Key Lab of BIPC University of Science and Technology of China), Xiangnan He (MoE Key Lab of BIPC University of Science and Technology of China)
Adversarial AttackImageTextMultimodality
🎯 What it does: Proposes a null-space projection-based activation-oriented defense framework called NullSteer for defending against multi-modal jailbreak attacks in vision-language models
PRISM: Learning a Shared Primitive Space for Transferable Skeleton Action Representation
Di Yang (University of Science and Technology of China), Jiangtao Wang (Teesside University)
ClassificationObject DetectionGenerationRepresentation LearningAuto EncoderVideo
🎯 What it does: Propose the PRISM framework, which learns a shared physics-informed motion primitive space, representing skeleton actions as primitive activation trajectories and migrating this structured representation to classification, detection, and generation tasks.
PRISM: Prototype-based Reasoning with Inter-modal Semantic Mining for Interpretable Image Recognition
Anni Yu (Nanjing University), Yu-Bin Yang (Nanjing University)
RecognitionExplainability and InterpretabilityConvolutional Neural NetworkVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Proposes the PRISM framework, which leverages auxiliary information from natural language to guide prototype learning, achieving interpretable image recognition.
PRISM: Video Dataset Condensation with Progressive Refinement and Insertion for Sparse Motion
Jaehyun Choi (Korea AI Safety Institute, ETRI), Junmo Kim (KAIST)
RecognitionCompressionConvolutional Neural NetworkVideo
🎯 What it does: Proposes PRISM, a sparse motion video dataset compression method based on progressive refinement and insertion;
PrITTI: Primitive-based Generation of Controllable and Editable 3D Semantic Urban Scenes
Christina Ourania Tze (University of Tübingen), Andreas Geiger (University of Tübingen)
GenerationData SynthesisAutonomous DrivingTransformerDiffusion modelAuto EncoderPoint Cloud
🎯 What it does: Designed and implemented a primitive-based latent diffusion model called PrITTI for generating editable and controllable 3D semantic urban scenes.
PrivateEyes: Gaze-Preserving Anonymization for Data Sharing
Surabhi Gupta (Samsung R&D Institute-Bangalore), Donghwan Seo (Samsung Electronics)
SegmentationGenerationData SynthesisPose EstimationSafty and PrivacyConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Proposed the PrivateEyes framework, which utilizes a three-stage pipeline to de-identify ocular images collected by wearable devices while preserving the gaze direction, enabling secure data sharing.
PriVi: Towards a General-Purpose Video Model for Primate Behavior in the Wild
Felix B. Mueller (University of Göttingen), Alexander S. Ecker (University of Göttingen)
ClassificationRecognitionContrastive LearningVideoBenchmark
🎯 What it does: Constructed a large-scale primate video pre-training dataset called PriVi, and continued pre-training the V-JEPA video model on this dataset, achieving improved cross-dataset action recognition performance.
PrivSynth: Alternating and Control-Based Optimization for Privacy and Utility in Synthetic Data
Xinyuan Zhao (Tsinghua University), Yuxing Han (Tsinghua University)
ClassificationObject DetectionData SynthesisOptimizationDiffusion modelImageBiomedical Data
🎯 What it does: Propose the PrivSynth framework, which treats synthetic data generation as a dual-objective optimization problem, employing alternating optimization and discrete-time optimal control (Pontryagin's maximum principle) to simultaneously satisfy privacy and downstream task performance requirements.
ProactiveMobile: A Comprehensive Benchmark for Boosting Proactive Intelligence On Mobile Devices
Dezhi Kong (Xiaomi Corporation), Ying Huang (Xiaomi Corporation)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextTabularSequentialBenchmark
🎯 What it does: The paper proposes a comprehensive benchmark called ProactiveMobile for mobile proactive intelligence.
Probabilistic Concept Graph Reasoning for Multimodal Misinformation Detection
Ruichao Yang (University of Science and Technology Beijing), Xu-Cheng Yin (University of Science and Technology Beijing)
ClassificationExplainability and InterpretabilityGraph Neural NetworkTransformerLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Propose the Probabilistic Concept Graph Reasoning (PCGR) framework, which transforms multi-modal fake news detection into soft probability reasoning on an interpretable concept graph, incorporating automatic concept expansion, hierarchical attention, and multi-layer concept graph construction.
Probabilistic Discrepancy Learning for Roadside LiDAR Scene Completion
Xiaogang Wu (Ningxia University), Yiqiang Wu (Ningxia University)
GenerationPose EstimationAutonomous DrivingConvolutional Neural NetworkDiffusion modelImageMultimodalityPoint CloudBenchmark
🎯 What it does: This paper proposes a Probability Difference Learning (PDL) framework that jointly optimizes the noisy poses from visual detection and sparse LiDAR point clouds to generate pseudo-real poses as supervision, subsequently completing road-side LiDAR scenes using a diffusion model.
Probabilistic Precipitation Nowcasting with Rectified Flow Transformers
Johannes Schusterbauer (LMU Munich), Björn Ommer (LMU Munich)
TransformerFlow-based ModelRectified FlowVideo
🎯 What it does: Propose the FREUD framework, implementing a Rectified Flow Transformer for frame-level encoding and unified video decoding, used for probabilistic rainfall nowcasting.
Probabilistic Prompt Adaptation for Unified Image Aesthetics and Quality Assessment
Takayuki Hara (LY Corporation), Yuya Otsuka (LY Corporation)
Large Language ModelPrompt EngineeringVision Language ModelImageText
🎯 What it does: Propose Probabilistic Prompt Adaptation (PPA), a unified probabilistic framework that dynamically infers image aesthetics and quality assessment scores using text prompts, supporting dual control at the task-level and prompt-level.
Probing and Bridging Geometry-Interaction Cues for Affordance Reasoning in Vision Foundation Models
Qing Zhang (Australian National University), Jing Zhang (Australian National University)
SegmentationRepresentation LearningTransformerDiffusion modelContrastive LearningImage
🎯 What it does: Systematically explore the two core capabilities of visual foundation models (VFM), geometric perception and interactive perception, and achieve zero-shot functional inference through unsupervised fusion.
ProcessMaker: A Generalized Process Visualization Framework with Adaptive Sequence Steps on Diffusion Transformers
Mengling Xu, Bing-Kun Bao (Nanjing University of Posts and Telecommunications)
GenerationDomain AdaptationTransformerSupervised Fine-TuningDiffusion modelContrastive LearningImageSequential
🎯 What it does: Propose the ProcessMaker framework, which utilizes Diffusion Transformers to generate cross-domain process sequences with self-adaptive steps.
ProFocus: Proactive Perception and Focused Reasoning in Vision-and-Language Navigation
Wei Xue (Fudan University), Lihua Zhang (Fudan University)
Autonomous DrivingTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Proposes a training-agnostic ProFocus framework that enhances the efficiency and accuracy of vision-language navigation through collaboration between LLM and VLM for active perception and focused reasoning.
Progress by Pieces: Test-Time Scaling for Autoregressive Image Generation
Joonhyung Park (Korea Advanced Institute of Science and Technology), Eunho Yang (Korea Advanced Institute of Science and Technology)
GenerationTransformerPrompt EngineeringVision Language ModelAuto EncoderImageBenchmark
🎯 What it does: This paper studies the scaling issue in visual autoregressive image generation models during testing, proposing the GridAR framework to achieve grid-partitioned progressive generation and candidate pruning, combined with a zero-shot validator and layout-specified prompt rewriting.
Progress-Think: Semantic Progress Reasoning for Vision-Language Navigation
Shuo Wang (School of Information, Renmin University of China), Zhaoxin Fan (Horizon Robotics)
Explainability and InterpretabilityReinforcement LearningVision Language ModelVision-Language-Action ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposed the Progress-Think framework, achieving long-term consistency and interpretability in Vision-Language Navigation (VLN) through semantic progress reasoning.
Progressive Cross-Modal Causal Intervention for Long-Term Action Recognition
Shaowu Xu (Beijing University of Technology), Qianmei Sun (Capital Medical University)
RecognitionTransformerVision Language ModelContrastive LearningVideo
🎯 What it does: Proposed the progressive cross-modal causal intervention framework PCMCI for long-term action recognition.
Progressive Guessing to Fixed Point: Rethinking Human Motion Prediction with Deep Equilibrium Models
Dong Wei (Hohai University), Yuhui Zheng (Qinghai Normal University)
Pose EstimationGraph Neural NetworkTime SeriesSequential
🎯 What it does: Convert the multi-stage progressive guessing framework for human motion prediction into a fixed-point problem of Deep Equilibrium Models (DEQ), achieving implicit layers, no explicit layers, and O(1) training memory.
Progressive Mask Distillation for Self-supervised Video Representation
Kewei Wu (Hefei University of Technology), Dan Guo (Hefei University of Technology)
Knowledge DistillationRepresentation LearningTransformerVideo
🎯 What it does: This paper proposes a Progressive Mask Distillation (PMD) method that performs self-supervised learning on videos using dynamic multi-stage mask ratios, and improves video representation quality through three modules: knowledge distillation, difficulty-aware region enhancement, and cross-layer feature alignment.
Progressive Multi-cue Alignment for Unaligned RGBT Tracking
Jiandong Jin (State Key Laboratory of Opto-Electronic Information Acquisition and Protection Technology), Jin Tang (Anhui University)
Object TrackingTransformerMixture of ExpertsMultimodality
🎯 What it does: Propose PMATrack, a phased multi-clue cross-modal alignment framework for unaligned RGB-T tracking.
Progressive Neural Architecture Generation
Caiyang Yu (Wenzhou University), Jiancheng Lv (Sichuan University)
ClassificationNeural Architecture SearchTransformerAuto EncoderImage
🎯 What it does: Propose a neural network architecture generation framework PNAG based on coarse-to-fine recursive generation, leveraging multi-scale sub-architecture quantization and step-by-step consistency constraints to significantly improve generation efficiency and effectiveness
Progressive Supernet Training for Efficient Visual Autoregressive Modeling
Xiaoyue Chen, Mingbao Lin
GenerationTransformerAuto EncoderImage
🎯 What it does: Proposed the VARiant super network framework, achieving efficient inference of visual autoregressive models (VAR) through adjustable depth, balancing quality and resource consumption;
ProgressiveAvatars: Progressive Animatable 3D Gaussian Avatars
Kaiwen Song (University of Science and Technology of China), Juyong Zhang (University of Science and Technology of China)
GenerationGaussian SplattingVideo
🎯 What it does: Propose ProgressiveAvatars, a 3D high-fidelity animated avatar representation that supports progressive transmission and rendering.
ProgTrack: A Multi-Object Tracking Algorithm with Progressive Matching Strategy
Chenhui Zhang (Xi'an University of Electronic Science and Technology), Weijie Peng (Xi'an University of Electronic Science and Technology)
Object TrackingConvolutional Neural NetworkVideo
🎯 What it does: Propose a multi-object tracking algorithm for UAV videos called ProgTrack, which adopts a three-stage progressive matching strategy (LMI, CE-Feature, GMI) and introduces CE-ReID, GRNED, and PKF modules.
ProjFlow: Projection Sampling with Flow Matching for Zero-Shot Exact Spatial Motion Control
Akihisa Watanabe (Waseda University), Kent Fujiwara (LY Corporation)
GenerationOptimizationFlow-based ModelRectified FlowTextSequential
🎯 What it does: Propose a zero-training, zero-loop iteration ProjFlow framework for precisely satisfying linear space motion control constraints.
ProM3E: Probabilistic Masked MultiModal Embedding Model for Ecology
Srikumar Sastry (Washington University in St. Louis), Nathan Jacobs (Washington University in St. Louis)
ClassificationRetrievalRepresentation LearningTransformerAuto EncoderContrastive LearningMultimodality
🎯 What it does: Proposes ProM3E, a multimodal embedding model based on a probabilistic masking variational autoencoder, capable of generating and inferring multimodal representations even when any modality is missing, applied to ecological tasks.
PROMO: Promptable Outfitting for Efficient High-Fidelity Virtual Try-On
Haohua Chen (Xiaohongshu Inc.), Zhiyong Wu (Tsinghua University)
Image TranslationSegmentationGenerationPose EstimationTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelDiffusion modelFlow-based ModelImageTextMultimodality
🎯 What it does: Propose PROMO, a high-fidelity virtual try-on framework that can control multiple garments through text prompts, capable of automatically generating required conditions even in the absence of a model or garment images.
Prompt Yourself: Awakening Textual Semantics in 1D Visual Tokenizers
Hualiang Wang (Hong Kong University of Science and Technology), Xiaomeng Li (Zhejiang University)
GenerationRepresentation LearningConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelAuto EncoderContrastive LearningImageTextMultimodality
🎯 What it does: Propose a 1D visual tokenizer, VLTok, which can learn through self-prompted alignment without relying on external text, encoding image information into both visual and text subsequences simultaneously to achieve finer-grained visual content representation.
Prompt-Anchored Vision-Text Distillation for Lifelong Person Re-identification
Wen Wen, Shiliang Zhang (University Of Electronic Science And Technology Of China)
RecognitionDomain AdaptationKnowledge DistillationMeta LearningTransformerPrompt EngineeringVision Language ModelContrastive LearningImageText
🎯 What it does: Propose a lifelong person re-identification framework without examples, PAD (Prompt-Anchored Vision-Text Distillation), which uses a frozen text encoder as a cross-domain semantic anchor. It achieves cross-domain semantic alignment and adaptive learning through learnable text prompts (TA-Prompt) and visual prompts (VA-Prompt), and employs knowledge distillation using a teacher model generated by EMA in the visual branch, maintaining overall model stability and plasticity.
Prompt-Free Universal Region Proposal Network
Qihong Tang (Nanjing University), Yang Gao (Nanjing University)
Object DetectionTransformerMixture of ExpertsContrastive LearningImage
🎯 What it does: Proposes a Prompt-Free Universal Region Proposal Network (PF-RPN) that can adaptively generate potential target boxes using visual features without any textual or visual prompts;
Prompt-Free Unknown Label Generation for Open World Detection in Remote Sensing
Abdullah Azeem (Shenzhen University), Abubakar Siddique (Chongqing University)
Object DetectionTransformerVision Language ModelImage
🎯 What it does: Propose the HSGDet framework to achieve open-world object detection in remote sensing images, capable of automatically discovering and assigning semantic labels to unknown targets during testing.
PromptDepth: Efficient and Promptable Geometric 3D Vision Model for Embodied Intelligence
Xianyun Wang (Harbin Institute of Technology), Jun Yu (Harbin Institute of Technology)
Object TrackingData SynthesisDepth EstimationAutonomous DrivingTransformerPrompt EngineeringContrastive LearningImageVideo
🎯 What it does: Propose PromptDepth, a feedforward network controllable by prompts, capable of generating scene, instance, or tracking depth maps in real-time, enabling multi-task dense prediction.
PromptEnhancer: Taming Your Rewriter for Text-to-Image Generation via Fine-Grained Reward
Linqing Wang, Qinglin Lu (Tencent Hunyuan)
GenerationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Designed PromptEnhancer, a general-purpose prompt rewriting framework that rewrites user prompts through Chain-of-Thought (CoT) to enable pre-trained text-to-image (T2I) models to better understand and generate images aligned with user intent;
PromptLoop: Plug-and-Play Prompt Refinement via Latent Feedback for Diffusion Model Alignment
Suhyeon Lee (KAIST), Jong Chul Ye (KAIST)
GenerationReinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningPrompt EngineeringDiffusion modelImageTextMultimodality
🎯 What it does: Achieving reward alignment for diffusion models by iteratively optimizing input prompts through feedback from intermediate latent states during the sampling process, using a multi-modal large language model (MLLM) for reinforcement learning, without directly modifying model weights.
PROMPTMINER: Black-Box Prompt Stealing against Text-to-Image Generative Models via Reinforcement Learning and VLM-Guided Optimization
Mingzhe Li (University of Massachusetts Amherst), Shiqing Ma (University of Massachusetts Amherst)
GenerationReinforcement LearningPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Propose a two-stage black-box prompt stealing framework called PROMPTMINER, which first recovers the main content through reinforcement learning and then optimizes style and modifiers using VLM-guided search.
PromptMoE: A Segmentation Refinement Framework Leveraging Mixture of Experts for Improved Prompting
Stephen Price, Elke A. Rundensteiner (Worcester Polytechnic Institute)
SegmentationPrompt EngineeringMixture of ExpertsImageBenchmark
🎯 What it does: This paper proposes PromptMoE, a segmentation refinement framework based on Mixture-of-Experts, which generates multi-point prompts using image information to enhance segmentation quality.
PromptStereo: Zero-Shot Stereo Matching via Structure and Motion Prompts
Xianqi Wang (Huazhong University of Science and Technology), Xin Yang (Optics Valley Laboratory)
Depth EstimationAutonomous DrivingRecurrent Neural NetworkSupervised Fine-TuningPrompt EngineeringImage
🎯 What it does: This paper proposes PromptStereo, a zero-shot stereo matching framework that achieves stronger iterative refinement by replacing GRU with Prompt Recurrent Unit (PRU) and combining structural prompts (SP) and motion prompts (MP).
Proof-of-Perception: Certified Tool-Using Multimodal Reasoning with Compositional Conformal Guarantees
Arya Fayyazi (University of Southern California), Haleh Akrami (Nuro)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelReinforcement LearningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Propose the Proof‑of‑Perception (PoP) framework, modeling multi-modal reasoning as executing a directed acyclic graph with conformal certificates, and dynamically allocating computational resources based on node-level uncertainty using a controller.
ProOOD: Prototype-Guided Out-of-Distribution 3D Occupancy Prediction
Yuheng Zhang, Kailun Yang (Hunan University)
Anomaly DetectionAutonomous DrivingContrastive LearningPoint Cloud
🎯 What it does: Propose the ProOOD framework, integrating prototype-guided semantic imputation, long-tail mining, and EchoOOD OOD detection, to achieve lightweight, pluggable 3D occupancy prediction and anomaly detection.
Property-Informed Diffusion-Based Text-to-Microstructure Generation
Bingxuan Dai (Southeast University), Jie Gui (Southeast University)
GenerationData SynthesisReinforcement LearningDiffusion modelGenerative Adversarial NetworkContrastive LearningTextMeshPhysics Related
🎯 What it does: Generating 3D Metamaterial Microstructures under Dual Conditions of Text and Physical Properties Based on Diffusion Models
PropFly: Learning to Propagate via On-the-Fly Supervision from Pre-trained Video Diffusion Models
Wonyong Seo (KAIST), Munchurl Kim (KAIST)
GenerationData SynthesisDiffusion modelFlow-based ModelAuto EncoderVideoText
🎯 What it does: Propose PropFly, a propagation-based video editing method that instantly generates supervised data during training by leveraging pre-trained video diffusion models.
ProPhy: Progressive Physical Alignment for Dynamic World Simulation
Zijun Wang (Sun Yat-sen University), Xiaodan Liang (Lenovo Research)
GenerationData SynthesisTransformerMixture of ExpertsVision Language ModelDiffusion modelVideoMultimodalityPhysics Related
🎯 What it does: This work proposes the ProPhy framework, which achieves step-by-step physical alignment of videos through a stage-wise physical expert mechanism, thereby generating physically consistent videos.
ProSoftArena: Benchmarking Hierarchical Capabilities of Multi-modal Agents in Professional Software Environments
Jiaxin Ai (Wuhan University), Kaipeng Zhang (BIT)
Large Language ModelAgentic AIVision Language ModelTextMultimodalityBenchmark
🎯 What it does: Constructed a benchmark for evaluating multimodal agents in professional software environments, named ProSoftArena. It proposed a five-tier capability hierarchy ranging from basic GUI operations to cross-software workflows, and finally to creation and project-level tasks. The benchmark pre-installs 13 core professional software packages in virtual machines, along with executable automated evaluation scripts and human-agent collaborative assessment modes.
Prospective Dynamic 3D MRI Reconstruction via Latent-Space Motion Tracking from Single Measurement
Lixuan Chen (University of Michigan), Liyue Shen (University of Michigan)
RestorationOptimizationAuto EncoderBiomedical DataMagnetic Resonance Imaging
🎯 What it does: To address the need for real-time, low-latency 3D dynamic imaging in MRI-guided therapy, this paper proposes a prospective reconstruction framework called PDMR. It leverages an offline-learned motion manifold and a geometry-aware tri-plane mapping network to rapidly estimate the current displacement field and reconstruct images from a sparse set of k-space measurements by optimizing a low-dimensional latent vector.
Protect to Adapt: Orthogonal Subspace Control with Ranked Negative-Prompt Curriculum for Few-Shot Action Recognition
Hantao Qi (Xiamen University), Hanzi Wang (Xiamen University)
RecognitionTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningVideo
🎯 What it does: Proposed a parameter-efficient adaptive framework called P2A for few-shot action recognition, which combines Orthogonal Subspace Control (OSC) and Difficulty-Sorted Negative Prompt Curriculum (RNC) to improve the adaptation of Vision-Language Models (VLMs).
Protego: User-Centric Pose-Invariant Privacy Protection Against Face Recognition-Induced Digital Footprint Exposure
Ziling Wang (University of Hong Kong), Ka-Ho Chow (University of Hong Kong)
RecognitionSafty and PrivacyGenerative Adversarial NetworkImageVideo
🎯 What it does: Propose a user-centric, pose-agnostic privacy protection method called Protego, which generates 2D protective textures from users' 3D facial features applicable across arbitrary poses, dynamically generates 3D masks to obscure faces before image upload, preventing facial recognition searches from exposing digital footprints.
Prototype-as-Prompt: Multimodal Sentiment Prototypes Endowing Large Language Models the Capability to Perform Multimodal Sentiment Analysis
Xianbing Zhao (Jiangnan University), Buzhou Tang (Harbin Institute of Technology)
ClassificationTransformerLarge Language ModelPrompt EngineeringMultimodality
🎯 What it does: Propose the Prototype-as-Prompt (PaP) framework, which maps audio/visual features into large language models (LLM) via a fixed number of emotional prototypes as soft prompts to accomplish multimodal sentiment analysis.
Prototype-based Causal Intervention for Multi-Label Image Classification
Yanmin Li (National University of Defense Technology), Weidong Bao (National University of Defense Technology)
ClassificationImageBenchmark
🎯 What it does: Propose the ProCI framework, which reduces confusion associations in multi-label image classification under image-level labels through dynamic confusion memory and adaptive causal intervention.