CVPR 2025 Papers — Page 20
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Perturb-and-Revise: Flexible 3D Editing with Generative Trajectories
Susung Hong (University of Washington), Ira Kemelmacher-Shlizerman (Google Research)
GenerationData SynthesisOptimizationDiffusion modelScore-based ModelNeural Radiance FieldMesh
🎯 What it does: A NeRF-based 3D editing framework (Perturb-and-Revise) is proposed, which can achieve various edits such as color, texture, shape, and pose.
pFedMxF: Personalized Federated Class-Incremental Learning with Mixture of Frequency Aggregation
Yifei Zhang (Nanyang Technological University), Han Yu (Nanyang Technological University)
Federated LearningImage
🎯 What it does: A framework for class-incremental learning in a federated learning environment is proposed—pFedMxF, which can alleviate spatial, temporal, and resource heterogeneity simultaneously.
PGC: Physics-Based Gaussian Cloth from a Single Pose
Michelle Guo (Stanford University), Egor Larionov (Meta Reality Labs)
GenerationData SynthesisPose EstimationGaussian SplattingMeshPhysics Related
🎯 What it does: Reconstructing clothing models that can be used for physical simulation and have a realistic appearance through single-frame multi-view capture, integrating mesh-embedded 3D Gaussian splats and physically-based rendering;
PhD: A ChatGPT-Prompted Visual Hallucination Evaluation Dataset
Jiazhen Liu (Renmin University of China), Xirong Li (Renmin University of China)
GenerationData SynthesisRetrievalTransformerLarge Language ModelPrompt EngineeringImageTextMultimodalityBenchmark
🎯 What it does: A large-scale multimodal visual illusion assessment benchmark PhD has been developed, which includes four assessment modes (base, sec, icc, ccs) and five visual tasks, constructed through a ChatGPT-assisted pipeline.
PHGC: Procedural Heterogeneous Graph Completion for Natural Language Task Verification in Egocentric Videos
Xun Jiang (University of Electronic Science and Technology of China), Heng Tao Shen (University of Electronic Science and Technology of China)
RecognitionData SynthesisGraph Neural NetworkVision Language ModelVideoTextMultimodality
🎯 What it does: This paper proposes a model based on Program Heterogeneous Graph Completion (PHGC) to verify whether multi-step tasks are completed in first-person perspective videos according to natural language instructions.
Phoenix: A Motion-based Self-Reflection Framework for Fine-grained Robotic Action Correction
Wenke Xia (Renmin University of China), Di Hu (Renmin University of China)
Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningMultimodality
🎯 What it does: A self-reflection framework based on motion commands, named Phoenix, is proposed, which utilizes a multimodal large language model to transform semantic reflection into fine-grained action corrections and achieve lifelong learning.
PhyS-EdiT: Physics-aware Semantic Image Editing with Text Description
Ziqi Cai (Peking University), Boxin Shi (Peking University)
Image TranslationGenerationDiffusion modelImageTextPhysics Related
🎯 What it does: A unified model named PhyS‑EdiT is proposed, achieving fine-grained, local, and continuous editing of low-level physical properties (roughness, metallicity, albedo, transparency) and high-level semantic information (text instructions) simultaneously.
PhysAnimator: Physics-Guided Generative Cartoon Animation
Tianyi Xie (University of California Los Angeles), Chenfanfu Jiang (University of California Los Angeles)
GenerationData SynthesisDiffusion modelOptical FlowVideoPhysics Related
🎯 What it does: This study proposes the PhysAnimator framework, which can generate dynamic animations with physical consistency and anime style from a single static anime illustration, and supports user interactive control through energy pens and binding points.
PhysGen3D: Crafting a Miniature Interactive World from a Single Image
Boyuan Chen (Tsinghua University), Shenlong Wang (Columbia University)
SegmentationGenerationDepth EstimationLarge Language ModelDiffusion modelImageVideo
🎯 What it does: Reconstruct an interactive 3D world from a single image and generate future dynamic videos through physical simulation.
Physical Plausibility-aware Trajectory Prediction via Locomotion Embodiment
Hiromu Taketsugu (Toyota Technological Institute), Norimichi Ukita (Toyota Technological Institute)
Pose EstimationRobotic IntelligenceTime SeriesPhysics Related
🎯 What it does: A pedestrian trajectory prediction framework based on physical feasibility assessment (Locomotion Embodiment) is proposed, which generates gait through physical simulation and evaluates trajectory feasibility using the LocoVal function, combined with EmLoco loss to train the prediction network.
PhysicsGen: Can Generative Models Learn from Images to Predict Complex Physical Relations?
Martin Spitznagel (Offenburg University), Janis Keuper (Mannheim University)
GenerationData SynthesisOptimizationDiffusion modelAuto EncoderGenerative Adversarial NetworkImageBenchmarkPhysics Related
🎯 What it does: This study proposes the PhysicsGen benchmark to explore whether generative models can learn complex physical relationships from image pairs and systematically evaluates three types of physical simulation tasks (urban sound propagation, lens distortion, and bouncing motion).
PhysVLM: Enabling Visual Language Models to Understand Robotic Physical Reachability
Weijie Zhou (Beijing Jiaotong University), Jinqiao Wang (Guangdong Polytechnic Normal University)
Robotic IntelligenceTransformerVision Language ModelMultimodalityBenchmark
🎯 What it does: A unified model called PhysVLM is proposed, which combines visual, linguistic, and robot reachability information to perform visual reasoning and task planning under reachability constraints in multi-robot environments.
PhyT2V: LLM-Guided Iterative Self-Refinement for Physics-Grounded Text-to-Video Generation
Qiyao Xue (University of Pittsburgh), Wei Gao (University of Pittsburgh)
GenerationData SynthesisOptimizationTransformerLarge Language ModelDiffusion modelVideoTextPhysics RelatedChain-of-Thought
🎯 What it does: This paper proposes PhyT2V, an iterative self-correction framework based on LLM, which improves the compliance of existing text-to-video models with physical rules through chain reasoning and backtracking reasoning.
PI-HMR: Towards Robust In-bed Temporal Human Shape Reconstruction with Contact Pressure Sensing
Ziyu Wu (University of Science and Technology of China), Xiaohui Cai (University of Science and Technology of China)
Pose EstimationKnowledge DistillationConvolutional Neural NetworkTransformerMultimodality
🎯 What it does: For the bed environment, an integrated framework from data annotation to model design is proposed to reconstruct the three-dimensional shape of the human body using time-sequenced pressure images from pressure-sensitive bedsheets.
PIAD: Pose and Illumination agnostic Anomaly Detection
Kaichen Yang (Dalian University of Technology), Andrea Tagliasacchi (Simon Fraser University)
Anomaly DetectionGaussian SplattingImage
🎯 What it does: Proposes the problem of pose and illumination insensitive anomaly detection (PIAD) and provides a complete baseline method based on 3D Gaussian Splatting.
PICD: Versatile Perceptual Image Compression with Diffusion Rendering
Tongda Xu (Tsinghua University), Yan Lu (Microsoft Research Asia)
CompressionDiffusion modelImageText
🎯 What it does: This paper proposes a universal perceptual image compression framework called PICD, which achieves high text accuracy and high visual quality for both screen content and natural images. The framework first utilizes OCR for lossless compression of text information, then incorporates the text as conditional input during image encoding; during decoding, a diffusion model synthesizes the compressed image with the text to create the final image, achieving seamless integration of text and image.
PICO: Reconstructing 3D People In Contact with Objects
Alpár Cseke (Max Planck Institute for Intelligent Systems), Dimitrios Tzionas (University of Amsterdam)
Object DetectionPose EstimationOptimizationImage
🎯 What it does: A framework called PICO has been developed for recovering human-object interaction (HOI) 3D reconstruction from a single natural image, which includes the dataset PICO-db and the method PICO-fit.
PIDLoc: Cross-View Pose Optimization Network Inspired by PID Controllers
Wooju Lee (Korea Advanced Institute of Science and Technology), Hyun Myung (Korea Advanced Institute of Science and Technology)
Pose EstimationAutonomous DrivingOptimizationImagePoint Cloud
🎯 What it does: A cross-view pose optimization network called PIDLoc, based on the idea of PID controllers, is proposed for precise vehicle positioning through satellite view images in GNSS-constrained environments.
PIDSR: Complementary Polarized Image Demosaicing and Super-Resolution
Shuangfan Zhou (Beijing University of Posts and Telecommunications), Imari Sato (National Institute of Informatics)
RestorationSuper ResolutionImage
🎯 What it does: By using a jointly designed two-stage recursive network, we achieve demosaicing and super-resolution of polarization images directly from CPFA raw images, resulting in high-quality, high-resolution polarization images along with their Degree of Polarization (DoP) and Angle of Polarization (AoP).
PillarHist: A Quantization-aware Pillar Feature Encoder based on Height-aware Histogram
Sifan Zhou (Southeast University), Ziyu Zhao (Southeast University)
Object DetectionAutonomous DrivingPoint Cloud
🎯 What it does: A pillar feature encoding module called PillarHist based on height histograms is proposed to address the issues of height information loss and activation value distribution differences in traditional PFE.
Pioneering 4-Bit FP Quantization for Diffusion Models: Mixup-Sign Quantization and Timestep-Aware Fine-Tuning
Maosen Zhao (Fudan University), Tao Chen (Fudan University)
GenerationData SynthesisComputational EfficiencyDiffusion modelImage
🎯 What it does: A framework for training and fine-tuning diffusion models based on 4-bit floating-point quantization is proposed, which significantly reduces memory and computational requirements while maintaining generation quality almost identical to that of full-precision models.
Pippo: High-Resolution Multi-View Humans from a Single Image
Yash Kant (Meta Reality Labs), Timur Bagautdinov (Meta Reality Labs)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: Pippo is a model based on diffusion transformers that can generate high-resolution (1K) and viewpoint-consistent multi-view human images from a single photo.
Pixel-aligned RGB-NIR Stereo Imaging and Dataset for Robot Vision
Jinnyeong Kim, Seung-Hwan Baek
Object DetectionRobotic IntelligenceConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a target detection method based on multi-scale convolutional neural networks, aimed at improving detection accuracy and robustness in complex scenes.
Pixel-level and Semantic-level Adjustable Super-resolution: A Dual-LoRA Approach
Lingchen Sun (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
RestorationSuper ResolutionDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a pixel-level and semantic-level adjustable super-resolution method called PiSA-SR, which utilizes the LoRA module to achieve decoupled and controllable inference for detail recovery and semantic enhancement on the pre-trained Stable Diffusion model.
PlanarSplatting: Accurate Planar Surface Reconstruction in 3 Minutes
Bin Tan (Ant Group), Nan Xue (Ant Group)
OptimizationGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes PlanarSplatting, a fast and high-precision indoor plane reconstruction method based on differentiable planar projection, which optimizes directly on 3D planar primitives.
Playing the Fool: Jailbreaking LLMs and Multimodal LLMs with Out-of-Distribution Strategy
Joonhyun Jeong (NAVER Cloud), Eunho Yang (Korea Advanced Institute of Science and Technology)
Adversarial AttackTransformerLarge Language ModelReinforcement LearningImageTextMultimodality
🎯 What it does: This paper proposes an attack framework called JOOD that breaks through the security alignment of large language models and multimodal models by transforming harmful inputs through out-of-distribution (OOD) modification.
PLeaS - Merging Models with Permutations and Least Squares
Anshul Nasery (University of Washington), Sewoong Oh (University of Washington)
OptimizationTransformerImage
🎯 What it does: A two-step model merging method called PLeaS is proposed, which can merge models with the same architecture but different pre-trained bases and tasks into a single model.
Plug-and-Play Interpretable Responsible Text-to-Image Generation via Dual-Space Multi-facet Concept Control
Basim Azam (University of Melbourne), Naveed Akhtar (University of Melbourne)
GenerationExplainability and InterpretabilityKnowledge DistillationDiffusion modelImageText
🎯 What it does: A pluggable and interpretable dual-space multi-faceted concept control method is proposed to simultaneously constrain the text embedding space and the latent space of the diffusion model during the text-to-image generation process, thereby achieving fair and safe image generation.
Plug-and-Play PPO: An Adaptive Point Prompt Optimizer Making SAM Greater
Xueyu Liu (Taiyuan University of Technology), Wen Zheng (Taiyuan University of Technology)
SegmentationOptimizationGraph Neural NetworkReinforcement LearningPrompt EngineeringImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a pluggable point prompt optimizer (PPO) that utilizes deep reinforcement learning to automatically adjust the point prompts of SAM on a dual-space heterogeneous graph, thereby enhancing unsupervised/one-shot segmentation performance.
Plug-and-Play Versatile Compressed Video Enhancement
Huimin Zeng (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationSegmentationSuper ResolutionCompressionConvolutional Neural NetworkMixture of ExpertsOptical FlowVideo
🎯 What it does: A pluggable compression video enhancement framework based on bitstream information (Codec-Aware Enhancement Framework) is proposed, which can adaptively enhance videos at different compression levels and directly support various downstream visual tasks.
PMA: Towards Parameter-Efficient Point Cloud Understanding via Point Mamba Adapter
Yaohua Zha (Tsinghua University), Shu-Tao Xia (Tsinghua University)
ClassificationSegmentationTransformerSupervised Fine-TuningPrompt EngineeringPoint Cloud
🎯 What it does: A parameter-efficient fine-tuning framework called Point Mamba Adapter (PMA) is proposed, which integrates all intermediate layer features on pre-trained point cloud models to enhance downstream task performance.
PMNI: Pose-free Multi-view Normal Integration for Reflective and Textureless Surface Reconstruction
Mingzhi Pei (Beijing University of Posts and Telecommunications), Zhanyu Ma
RestorationDepth EstimationNeural Radiance FieldImage
🎯 What it does: This paper proposes a posture-free 3D reconstruction method based on multi-view surface normals, PMNI, which can simultaneously recover high-quality surfaces of reflective and texture-less invalid surfaces along with camera poses without prior knowledge of camera poses.
PO3AD: Predicting Point Offsets toward Better 3D Point Cloud Anomaly Detection
Jianan Ye (Xi'an Jiaotong-Liverpool University), Kaizhu Huang (Duke Kunshan University)
Anomaly DetectionConvolutional Neural NetworkPoint Cloud
🎯 What it does: A 3D point cloud anomaly detection framework PO3AD based on point offset prediction is proposed, along with a pseudo-anomaly generation method Norm-AS based on normal vectors.
Point Cloud Upsampling Using Conditional Diffusion Module with Adaptive Noise Suppression
Boqian Zhang (Xidian University), Guang Jiang (Xidian University)
Diffusion modelPoint Cloud
🎯 What it does: This paper proposes a point cloud upsampling network PDANS based on a conditional diffusion model, which can suppress noise while preserving geometric details.
Point Clouds Meets Physics: Dynamic Acoustic Field Fitting Network for Point Cloud Understanding
Changshuo Wang (Nanyang Technological University), Prayag Tiwari (Halmstad University)
RecognitionSegmentationConvolutional Neural NetworkPoint CloudPhysics Related
🎯 What it does: This paper proposes a dynamic adaptive convolutional network for point clouds, DAFNet, based on the principles of acoustic fields, to improve local structure representation.
Point-Cache: Test-time Dynamic and Hierarchical Cache for Robust and Generalizable Point Cloud Analysis
Hongyu Sun (Renmin University of China), Jianfei Cai (Monash University)
ClassificationRecognitionDomain AdaptationPoint Cloud
🎯 What it does: A training-free, hierarchical caching model called Point-Cache is proposed, constructed solely using point cloud data at the testing stage, to achieve adaptive and generalized point cloud recognition under distribution drift.
Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting
Wei Lin (City University of Hong Kong), Antoni B. Chan (City University of Hong Kong)
Object DetectionDomain AdaptationExplainability and InterpretabilityConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a semi-supervised point-based crowd counting framework that utilizes point-to-region (P2R) matching to replace traditional point-to-point (P2P) matching, and visualizes the credibility of pseudo-labels through Point-Specific Activation Maps (PSAM), enabling the training of a high-accuracy crowd counter using only a small number of labeled samples and a large number of unlabeled samples.
Point2RBox-v2: Rethinking Point-supervised Oriented Object Detection with Spatial Layout Among Instances
Yi Yu (Southeast University), Xue Yang (Shanghai Jiao Tong University)
Object DetectionImage
🎯 What it does: This paper proposes a single-point supervised inclined object detection method using instance spatial layout—Point2RBox-v2.
PointLoRA: Low-Rank Adaptation with Token Selection for Point Cloud Learning
Song Wang (Zhejiang University), Xinchao Wang (National University of Singapore)
RecognitionSegmentationTransformerPrompt EngineeringPoint Cloud
🎯 What it does: A parameter-efficient fine-tuning method for point cloud pre-training models, PointLoRA, is proposed, achieving fine-tuning with only 3.43% of trainable parameters by combining low-rank adaptation and multi-scale token selection.
PointSR: Self-Regularized Point Supervision for Drone-View Object Detection
Weizhuo Li (Xidian University), Qiguang Miao (Xidian University)
Object DetectionImage
🎯 What it does: A self-regularized point supervision object detection framework called PointSR is proposed, which can automatically generate high-quality pseudo boxes from single point annotations and significantly improve the performance of object detection from a drone's perspective.
PolarFree: Polarization-based Reflection-Free Imaging
Mingde Yao (Chinese University of Hong Kong), Jinwei Gu (Chinese University of Hong Kong)
RestorationDiffusion modelImage
🎯 What it does: This paper proposes a reflection removal method called PolarFree that utilizes polarization information and diffusion models, and constructs the PolaRGB dataset containing 6500 pairs of RGB-polarized images.
Polarized Color Screen Matting
Kenji Enomoto (Adobe Research), TJ Rhodes (Adobe Research)
RestorationSegmentationImageVideo
🎯 What it does: This paper proposes a single-frame, pixel-wise method for matting transparent objects by combining hue and polarization information—Polarized Color Screen Matting (pCSM). It can directly recover the alpha transparency and foreground color of the target foreground under the conditions of known polarized background and calibrated light source color.
PolarNeXt: Rethink Instance Segmentation with Polar Representation
Jiacheng Sun (Shanghai University), Xiaomao Li (Shanghai Artificial Intelligence Laboratory)
Object DetectionSegmentationImage
🎯 What it does: This paper proposes the PolarNeXt framework, which improves Polar Representation for efficient and lightweight instance segmentation.
Poly-Autoregressive Prediction for Modeling Interactions
Neerja Thakkar (University of California Berkeley), Jitendra Malik (University of California Berkeley)
Pose EstimationAutonomous DrivingTransformerSequential
🎯 What it does: Proposes the Poly-Autoregressive (PAR) framework, which uses the historical states of multiple agents to predict the future behavior of the target agent;
POMP: Physics-consistent Motion Generative Model through Phase Manifolds
Bin Ji (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
GenerationRobotic IntelligenceMixture of ExpertsDiffusion modelAuto EncoderTime SeriesPhysics Related
🎯 What it does: POMP is proposed, a motion generation model that achieves physically constrained motion synthesis through phase manifolds, enabling real-time physically feasible motion synthesis.
POp-GS: Next Best View in 3D-Gaussian Splatting with P-Optimality
Joey Wilson (University of Michigan), Arnab Sen (Amazon)
OptimizationGaussian SplattingPoint Cloud
🎯 What it does: A method for calculating view information gain in 3D Gaussian spraying (3D-GS) based on optimal experimental design (P-Optimality) is proposed, along with a block diagonal covariance approximation.
POPEN: Preference-Based Optimization and Ensemble for LVLM-Based Reasoning Segmentation
Lanyun Zhu (Singapore University of Technology and Design), Jun Liu (Lancaster University)
SegmentationOptimizationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageText
🎯 What it does: The POPEN framework is proposed to enhance the text quality and pixel-level segmentation accuracy of large audiovisual-language models (LVLM) in reasoning segmentation tasks through preference optimization and integration methods, significantly reducing hallucination phenomena in the text.
Population Normalization for Federated Learning
Zhuoyao Wang (Fudan University), Weizhong Zhang (Fudan University)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes Population Normalization (PN) and its noise version NPN suitable for federated learning, eliminating the limitations of batch normalization in scenarios with data heterogeneity and small batch sizes.
Pos3R: 6D Pose Estimation for Unseen Objects Made Easy
Weijian Deng (Australian National University), Stephen Gould (Australian National University)
Pose EstimationImage
🎯 What it does: Pos3R is proposed, a training-free six-dimensional pose estimation method that requires no training and only uses RGB images.
Pose Priors from Language Models
Sanjay Subramanian (University of California), Trevor Darrell (University of California)
Pose EstimationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Utilize large multimodal models (LMM) to extract body contact information from images and transform this information into constraint losses to optimize 3D human pose, achieving unlabeled contact perception and pose reconstruction.
Pose-Guided Temporal Enhancement for Robust Low-Resolution Hand Reconstruction
Kaixin Fan (Beijing University of Posts and Telecommunications), Jianxin Liao (Beijing University of Posts and Telecommunications)
Pose EstimationTransformerVideo
🎯 What it does: This paper proposes a temporal enhancement framework for low-resolution hand reconstruction: it first captures temporal information using joint features from historical frames, then projects these sparse 3D joint information into dense 2D features through three-plane feature encoding, thereby enhancing the visual features of the current frame, and finally regresses the hand mesh from the enhanced features.
PoseBH: Prototypical Multi-Dataset Training Beyond Human Pose Estimation
Uyoung Jeong (UNIST), Kwang In Kim (POSTECH)
Pose EstimationTransformerImage
🎯 What it does: A multi-dataset training framework named PoseBH is proposed for unified processing of human, animal, and hand keypoint estimation across different skeleton formats.
PoseTraj: Pose-Aware Trajectory Control in Video Diffusion
Longbin Ji (University of Edinburgh), Changjian Li (University of Edinburgh)
GenerationData SynthesisPose EstimationConvolutional Neural NetworkDiffusion modelVideo
🎯 What it does: PoseTraj is proposed, a trajectory control framework that allows video generation models to accurately follow and perceive the 6D rotation pose of objects given a 2D trajectory.
Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising
Tong Li (Beijing Institute of Technology), Hua Huang (Beijing Normal University)
RestorationImage
🎯 What it does: This paper studies a new self-supervised single-image denoising paradigm called Positive2Negative, which constructs multi-scale positive and negative noise images by predicting noise re-noising and trains the network with denoising consistency supervision, successfully breaking through the barrier of information loss.
Post-pre-training for Modality Alignment in Vision-Language Foundation Models
Shin'ya Yamaguchi (NTT), Daiki Chijiwa (NTT)
ClassificationRetrievalKnowledge DistillationTransformerContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes CLIP-Refine, a lightweight post-pretraining method between CLIP pretraining and fine-tuning, aimed at aligning image and text features and reducing the gap between modalities.
POSTA: A Go-to Framework for Customized Artistic Poster Generation
Haoyu Chen (Hong Kong University of Science and Technology), Xinchao Wang (National University of Singapore)
GenerationLarge Language ModelDiffusion modelImageText
🎯 What it does: The POSTA framework is proposed, enabling customizable generation from text descriptions to complete artistic posters.
PosterMaker: Towards High-Quality Product Poster Generation with Accurate Text Rendering
Yifan Gao (University of Science and Technology of China), Hongtao Xie (University of Science and Technology of China)
GenerationDiffusion modelImageBenchmark
🎯 What it does: This paper presents PosterMaker, an end-to-end product poster generation framework that can automatically generate visually appealing and textually accurate posters given a subject image, background prompts, and text layout.
PosterO: Structuring Layout Trees to Enable Language Models in Generalized Content-Aware Layout Generation
HsiaoYuan Hsu (Wangxuan Institute of Computer Technology Peking University), Yuxin Peng (Wangxuan Institute of Computer Technology Peking University)
GenerationData SynthesisTransformerLarge Language ModelImage
🎯 What it does: A content-aware layout generation framework named PosterO has been developed, which is based on large language models and does not require fine-tuning. It can automatically generate multi-shape elements and versatile layouts for a given image.
POT: Prototypical Optimal Transport for Weakly Supervised Semantic Segmentation
Jian Wang (XJTLU), Jimin Xiao (XJTLU)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: A weakly supervised semantic segmentation framework based on multi-cluster prototypes and similarity-aware optimal transport (POT) is proposed, which can significantly enhance the completeness of class activation maps.
Potential Field Based Deep Metric Learning
Shubhang Bhatnagar (University of Illinois Urbana-Champaign), Narendra Ahuja (University of Illinois Urbana-Champaign)
RetrievalOptimizationContrastive LearningImage
🎯 What it does: A potential field-based deep metric learning (PFML) framework is proposed, using continuous potential fields to model the interactions among all samples.
Pow3R: Empowering Unconstrained 3D Reconstruction with Camera and Scene Priors
Wonbong Jang (University College London), Jerome Revaud (Naver Labs Europe)
Pose EstimationDepth EstimationTransformerImageBenchmark
🎯 What it does: The Pow3R model is proposed, extending DUSt3R to support the injection of multi-modal information such as camera intrinsic parameters, relative poses, dense or sparse depth during testing, achieving more accurate 3D reconstruction, depth prediction, and pose estimation tasks.
PQPP: A Joint Benchmark for Text-to-Image Prompt and Query Performance Prediction
Eduard Poesina (University of Bucharest), Radu Tudor Ionescu (University of Bucharest)
GenerationRetrievalTransformerVision Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: This paper proposes a joint benchmark (PQPP) to evaluate the difficulty of prompts/queries in text-to-image generation and retrieval tasks, and obtains real performance scores through human annotation.
Practical Solutions to the Relative Pose of Three Calibrated Cameras
Charalambos Tzamos (Czech Technical University in Prague), Zuzana Kukelova (Czech Technical University in Prague)
Pose EstimationSimultaneous Localization and MappingImage
🎯 What it does: For the problem of estimating the relative pose of three calibrated cameras using four points and three views, two types of approximate geometric-based minimal solvers (4p3v(A) and 4p3v(M)) are proposed, which are further improved within the RANSAC framework to include approximate corresponding point offsets, early non-minimal re-estimation, three-view filtering, and local optimization to form a complete solver.
PRaDA: Projective Radial Distortion Averaging
Daniil Sinitsyn (Technical University of Munich), Daniel Cremers (Technical University of Munich)
Pose EstimationOptimizationSimultaneous Localization and MappingImage
🎯 What it does: A projection space-based automatic radial distortion calibration method, PRaDA, is proposed, which can estimate the camera's radial distortion from unordered image collections without the need for 3D reconstruction.
Precise Event Spotting in Sports Videos: Solving Long-Range Dependency and Class Imbalance
Sanchayan Santra (Sony Research India), Vineeth N Balasubramanian (Indian Institute of Technology Hyderabad)
RecognitionObject DetectionConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningVideo
🎯 What it does: An end-to-end precise event localization framework is proposed, utilizing CNN to extract spatio-temporal features, Bi-GRU to model long-range temporal dependencies, and enhancing feature quality through an Adaptive Spatio-Temporal Refinement Module (ASTRM), while addressing the class imbalance issue with Soft-IC loss.
Precise, Fast, and Low-cost Concept Erasure in Value Space: Orthogonal Complement Matters
Yuan Wang (University of Science and Technology of China), Xiangnan He (Institute of Software Chinese Academy of Sciences)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: A training-agnostic concept elimination method called AdaVD has been developed, which can accurately erase target concepts in text-to-image diffusion models while preserving prior knowledge of non-target concepts.
PreciseCam: Precise Camera Control for Text-to-Image Generation
Edurne Bernal-Berdun (Universidad de Zaragoza), Diego Gutierrez (Adobe Research)
GenerationDiffusion modelImageText
🎯 What it does: This paper presents PreciseCam, a method for achieving precise control over four parameters of camera angles (roll, pitch) and lens distortion (vFoV, ξ) in text-to-image generation.
Preconditioners for the Stochastic Training of Neural Fields
Shin-Fang Chng (Australian Institute of Machine Learning, University of Adelaide), Simon Lucey (Australian Institute of Machine Learning, University of Adelaide)
OptimizationNeural Radiance FieldImage
🎯 What it does: A curvature-aware preconditioning framework for the random training of neural fields is proposed, and its acceleration effects are validated under various activation functions.
PrEditor3D: Fast and Precise 3D Shape Editing
Ziya Erkoç, Peiye Zhuang (Technical University of Munich)
Object DetectionSegmentationGenerationComputational EfficiencyPrompt EngineeringDiffusion modelPoint CloudMesh
🎯 What it does: A training-independent 3D editing framework called PrEditor3D is proposed, which utilizes a multi-view diffusion model to generate four-view images. It performs synchronous editing based on Prompt-to-Prompt and user rough masks, then detects the intended areas using Grounding-DINO and SAM-2. Finally, the editing results are projected into 3D space and fused with the original model, allowing for quick and fine local editing of a single shape.
Preserve or Modify? Context-Aware Evaluation for Balancing Preservation and Modification in Text-Guided Image Editing
Yoonjeon Kim (KAIST), Eunho Yang (KAIST)
Image TranslationGenerationTransformerLarge Language ModelContrastive LearningImageText
🎯 What it does: A context-based evaluation metric called AugCLIP is proposed to measure the retention and modification quality of text-guided image editing.
Preserving Clusters in Prompt Learning for Unsupervised Domain Adaptation
Tung-Long Vuong (Monash University), Dinh Phung (Monash University)
Domain AdaptationPrompt EngineeringContrastive LearningImage
🎯 What it does: A CLIP-based prompt learning framework is proposed, which uses weighted pseudo-labels from the source domain prompts to enhance the pseudo-labels in the target domain, and maintains clustering consistency of visual and textual embeddings through optimal transport (Wasserstein) constraints, achieving unsupervised domain adaptation.
Prior Does Matter: Visual Navigation via Denoising Diffusion Bridge Models
Hao Ren (Sun Yat-sen University), Hui Cheng (Sun Yat-sen University)
Autonomous DrivingRobotic IntelligenceConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: This paper proposes a visual navigation framework called NaviBridger based on the diffusion bridge model, which enhances navigation accuracy and efficiency by starting action generation from prior actions rather than Gaussian noise.
Prior-free 3D Object Tracking
Xiuqiang Song (Shandong University), Xueying Qin (Qilu University of Technology)
Object TrackingPose EstimationSimultaneous Localization and MappingImagePoint CloudMesh
🎯 What it does: This paper proposes a 3D object tracking method called BIT, which completely relies on RGB images to generate high-precision mesh models in real-time and accomplish 6DoF pose tracking without any pre-trained models or training data.
ProAPO: Progressively Automatic Prompt Optimization for Visual Classification
Xiangyan Qu (Institute of Information Engineering Chinese Academy of Sciences), Gang Xiong (Institute of Information Engineering Chinese Academy of Sciences)
ClassificationOptimizationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes a Progressive Automatic Prompt Optimization (ProAPO) method, which automatically generates discriminative visual language model prompts for fine-grained classification through an evolutionary search from task templates to category descriptions.
Probabilistic Prompt Distribution Learning for Animal Pose Estimation
Jiyong Rao (Tongji University), Yu Wang (Tongji University)
Pose EstimationTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: This paper proposes an animal pose estimation method based on probabilistic prompt distribution learning, utilizing learnable prompts and cross-modal fusion to enhance cross-species generalization capabilities.
Probability Density Geodesics in Image Diffusion Latent Space
Qingtao Yu (Australian National University), Dylan Campbell (Australian National University)
GenerationData SynthesisOptimizationDiffusion modelScore-based ModelImage
🎯 What it does: This paper proposes a method to calculate the geometric shortest path (geodesic) in the latent space of diffusion models using a probability density inverse proportionality metric, and provides algorithms for solving boundary value problems (BVP) and initial value problems (IVP);
ProbeSDF: Light Field Probes For Neural Surface Reconstruction
Briac Toussaint (Univ. Grenoble Alpes), Jean-Sébastien Franco (Univ. Grenoble Alpes)
GenerationComputational EfficiencyNeural Radiance FieldPoint CloudMesh
🎯 What it does: A new SDF-based neural surface reconstruction method is proposed, utilizing light field probes with separated spatial and angular features, and achieving efficient rendering through a minimal MLP.
Probing the Mid-level Vision Capabilities of Self-Supervised Learning
Xuweiyi Chen (University of Virginia), Zezhou Cheng (University of Virginia)
SegmentationDepth EstimationTransformerContrastive LearningImageBenchmark
🎯 What it does: This paper constructs a mid-level vision benchmark suite to systematically evaluate 22 mainstream self-supervised learning models on 8 mid-level vision tasks (object segmentation, depth/normal estimation, geometric correspondence, image similarity, etc.) and compares the results with ImageNet linear classification performance.
ProbPose: A Probabilistic Approach to 2D Human Pose Estimation
Miroslav Purkrabek (Czech Technical University in Prague), Jiri Matas (Czech Technical University in Prague)
Pose EstimationImage
🎯 What it does: Proposes the ProbPose model, which uses probabilistic graphs and presence probability for 2D human pose estimation.
Prof. Robot: Differentiable Robot Rendering Without Static and Self-Collisions
Quanyuan Ruan (South China University of Technology), Kui Jia (Chinese University of Hong Kong Shenzhen)
OptimizationRobotic IntelligenceGaussian SplattingImage
🎯 What it does: A differentiable robot rendering framework called Prof. Robot is proposed, which utilizes a learned high-dimensional pose classifier and Eikonal regularization to avoid static environments and self-collisions.
Progress-Aware Video Frame Captioning
Zihui Xue (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)
RecognitionGenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoText
🎯 What it does: Proposed and implemented the Progress-Aware Video Frame Captioning task.
Progressive Correspondence Regenerator for Robust 3D Registration
Guiyu Zhao (Xiamen University), Yulan Guo (Sun Yat-sen University)
RecognitionOptimizationPoint Cloud
🎯 What it does: An iterative correspondence regeneration framework named Regor is designed, which utilizes the correspondence points obtained from previous iterations as priors to continuously generate higher quality matches within a local spherical region, thereby achieving robust 3D point cloud registration.
Progressive Focused Transformer for Single Image Super-Resolution
Wei Long (University of Electronic Science and Technology of China), Shuhang Gu (University of Electronic Science and Technology of China)
RestorationSuper ResolutionTransformerImage
🎯 What it does: This paper proposes a Progressive Focused Transformer (PFT) that enhances feature aggregation in single image super-resolution tasks through Progressive Focused Attention (PFA).
Progressive Rendering Distillation: Adapting Stable Diffusion for Instant Text-to-Mesh Generation without 3D Data
Zhiyuan Ma (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
GenerationKnowledge DistillationDiffusion modelMesh
🎯 What it does: Proposes Progressive Rendering Distillation (PRD), which directly transforms Stable Diffusion into a local 3D generator without 3D training data, generating high-quality textured meshes in just 1.2 seconds.
ProHOC: Probabilistic Hierarchical Out-of-Distribution Classification via Multi-Depth Networks
Erik Wallin (Saab AB), Lars Hammarstrand (Chalmers University of Technology)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: A probabilistic OOD classification framework based on class hierarchy, ProHOC, is proposed, which can assign unknown samples to appropriate internal nodes in the hierarchy rather than simply labeling them as OOD.
ProjAttacker: A Configurable Physical Adversarial Attack for Face Recognition via Projector
Yuanwei Liu (Wuhan University), Zheng Wang (Wuhan University)
RecognitionAdversarial AttackReinforcement LearningGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes ProjAttacker, which uses a projector to project optical adversarial textures to physically attack faces, capable of deceiving facial recognition and bypassing liveness detection.
Project-Probe-Aggregate: Efficient Fine-Tuning for Group Robustness
Beier Zhu (Nanyang Technological University), Chi Zhang (Westlake University)
ClassificationRecognitionOptimizationComputational EfficiencyTransformerSupervised Fine-TuningVision Language ModelContrastive LearningImageText
🎯 What it does: A three-step framework called Project-Probe-Aggregate (PPA) is proposed to efficiently fine-tune image-text foundational models without the need for training cluster labels, thereby enhancing robustness against group imbalance data.
ProKeR: A Kernel Perspective on Few-Shot Adaptation of Large Vision-Language Models
Yassir Bendou (IMT Atlantique), Adnane Boukhayma (Inria)
ClassificationDomain AdaptationTransformerVision Language ModelImage
🎯 What it does: By reformulating the Tip-Adapter as a kernel regression model and incorporating local linear regression and global RKHS proximal regularization, a training-free few-shot adaptation method called ProKeR is proposed.
Prometheus: 3D-Aware Latent Diffusion Models for Feed-Forward Text-to-3D Scene Generation
Yuanbo Yang (Zhejiang University), Yiyi Liao (Zhejiang University)
GenerationData SynthesisAutonomous DrivingTransformerDiffusion modelGaussian SplattingImagePoint Cloud
🎯 What it does: Prometheus has been developed, a text-to-3D scene generation model based on 3D Gaussian Splatting, capable of synthesizing 3D scenes from single or multiple views in seconds.
Prompt-CAM: Making Vision Transformers Interpretable for Fine-Grained Analysis
Arpita Chowdhury (Ohio State University), Wei-Lun Chao (Ohio State University)
ClassificationExplainability and InterpretabilityTransformerPrompt EngineeringImage
🎯 What it does: A method called PROMPT-CAM is proposed, which utilizes class-specific prompts based on a pre-trained Vision Transformer (ViT) to achieve fine-grained interpretability.
Prompt2Perturb (P2P): Text-Guided Diffusion-Based Adversarial Attack on Breast Ultrasound Images
Yasamin Medghalchi (University of British Columbia), Ilker Hacihaliloglu (University of British Columbia)
GenerationAdversarial AttackConvolutional Neural NetworkPrompt EngineeringDiffusion modelImageBiomedical DataUltrasound
🎯 What it does: This paper proposes Prompt2Perturb (P2P), a method for generating adversarial samples of breast ultrasound images using a text prompt-driven latent diffusion model.
PromptHash:Affinity-Prompted Collaborative Cross-Modal Learning for Adaptive Hashing Retrieval
Qiang Zou (Xinjiang University), Jiayi Chen (Xinjiang University)
RetrievalTransformerPrompt EngineeringContrastive LearningImageTextMultimodality
🎯 What it does: This work proposes a novel cross-modal hashing framework called PromptHash, which combines prompt learning and state space modeling to enhance retrieval performance.
PromptHMR: Promptable Human Mesh Recovery
Yufu Wang (Meshcapade), Muhammed Kocabas (Max Planck Institute for Intelligent Systems)
Pose EstimationTransformerPrompt EngineeringVision Language ModelMesh
🎯 What it does: A method called PromptHMR is proposed for human mesh recovery, which improves the accuracy and robustness of 3D human pose and shape estimation through spatial and semantic prompts.
Prompting Depth Anything for 4K Resolution Accurate Metric Depth Estimation
Haotong Lin (Zhejiang University), Bingyi Kang (ByteDance)
Depth EstimationTransformerPrompt EngineeringPoint Cloud
🎯 What it does: Introducing Prompt Depth Anything, which utilizes low-cost LiDAR as a quantization prompt to activate depth-based models, directly outputting 4K level precise metric depth.
ProReflow: Progressive Reflow with Decomposed Velocity
Lei Ke (Tsinghua University), Linfeng Zhang (Shanghai Jiao Tong University)
GenerationData SynthesisDiffusion modelFlow-based ModelRectified FlowImage
🎯 What it does: The ProReflow method is proposed, which significantly improves the quality of few-step diffusion generation by gradually approaching a linear trajectory in stages and emphasizing speed direction matching.
Prosody-Enhanced Acoustic Pre-training and Acoustic-Disentangled Prosody Adapting for Movie Dubbing
Zhedong Zhang (Hangzhou Dianzi University), Yuankai Qi (Macquarie University)
GenerationGenerative Adversarial NetworkAudio
🎯 What it does: This paper proposes a two-stage movie dubbing generation framework, which first performs acoustic pre-training with voiceprint enhancement, and then freezes the acoustic system to achieve acoustic decoupling of voiceprint and emotional alignment in a non-acoustic mixing manner, thereby generating high-quality and emotionally consistent movie dubbing.
Protecting Your Video Content: Disrupting Automated Video-based LLM Annotations
Haitong Liu (Tsinghua University), Shu-Tao Xia (Tsinghua University)
Safty and PrivacyAdversarial AttackLarge Language ModelVideo
🎯 What it does: This paper proposes two covert watermarking methods to prevent large language models from automatically annotating personal videos without authorization.
ProtoDepth: Unsupervised Continual Depth Completion with Prototypes
Patrick Rim (Yale Vision Lab), Alex Wong (Yale Vision Lab)
RestorationDepth EstimationConvolutional Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: This paper studies a prototype learning method called ProtoDepth, which enables continual learning in unsupervised depth completion tasks, allowing adaptation to new non-stationary data distributions without forgetting knowledge from old domains.
Prototype-Based Image Prompting for Weakly Supervised Histopathological Image Segmentation
Qingchen Tang (University of New South Wales), Yang Song (University of New South Wales)
SegmentationTransformerContrastive LearningImageBiomedical Data
🎯 What it does: A weakly supervised histopathological image segmentation framework based on Prototype-Based Image Prompting (PBIP) is proposed, which constructs a multi-prototype image library using image-level labels and achieves feature matching through contrastive learning, thereby generating more accurate Class Activation Maps (CAM).
Provoking Multi-modal Few-Shot LVLM via Exploration-Exploitation In-Context Learning
Cheng Chen (Alibaba Group), Jia Li (Alibaba Group)
OptimizationTransformerReinforcement LearningVision Language ModelImageMultimodality
🎯 What it does: This paper proposes a context learning method for a multimodal few-shot large visual language model (LVLM) based on an exploration-exploitation reinforcement learning framework, aimed at automatically selecting the optimal demonstration combinations to enhance the few-shot reasoning performance of LVLM.