CVPR 2024 Papers — Page 19
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
PAPR in Motion: Seamless Point-level 3D Scene Interpolation
Shichong Peng (Simon Fraser University), Ke Li (Simon Fraser University)
GenerationData SynthesisNeural Radiance FieldPoint Cloud
🎯 What it does: Proposes a point-level 3D scene interpolation task, utilizing unsupervised two-state multi-view training to generate continuous geometric and appearance interpolations, and can render from any new viewpoint.
PARA-Drive: Parallelized Architecture for Real-time Autonomous Driving
Xinshuo Weng (NVIDIA Research), Marco Pavone (Stanford University)
Autonomous DrivingExplainability and InterpretabilityComputational EfficiencyTransformerSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper systematically explores the design space of end-to-end modular autonomous driving architectures, proposing and implementing the fully parallel PARA-Drive architecture, which significantly enhances operational efficiency while maintaining performance and interpretability.
Parameter Efficient Fine-tuning via Cross Block Orchestration for Segment Anything Model
Zelin Peng (Shanghai Jiao Tong University), Wei Shen (Shanghai Jiao Tong University)
SegmentationTransformerSupervised Fine-TuningImageBiomedical Data
🎯 What it does: A cross-block collaborative mechanism called SAM-COBOT is proposed for parameter-efficient fine-tuning of the Segment Anything Model.
Parameter Efficient Self-Supervised Geospatial Domain Adaptation
Linus Scheibenreif (University of St. Gallen), Damian Borth (Stuttgart University of Applied Sciences)
Domain AdaptationTransformerSupervised Fine-TuningImage
🎯 What it does: This study investigates a parameter-efficient self-supervised adaptation method for transferring large visual foundation models to new data modalities and tasks in the remote sensing domain.
ParameterNet: Parameters Are All You Need for Large-scale Visual Pretraining of Mobile Networks
Kai Han (Huawei Noah's Ark Lab), Enhua Wu (State Key Lab of Computer Science, ISCAS & UCAS)
ClassificationComputational EfficiencyConvolutional Neural NetworkMixture of ExpertsImageText
🎯 What it does: This paper proposes ParameterNet, which significantly increases the model parameter count while maintaining low FLOPs by utilizing techniques such as dynamic convolution, thereby addressing the low FLOPs pre-training bottleneck and enhancing the performance of mobile networks after pre-training on ImageNet-22K.
ParamISP: Learned Forward and Inverse ISPs using Camera Parameters
Woohyeok Kim (POSTECH), Sunghyun Cho (POSTECH)
Image TranslationRestorationConvolutional Neural NetworkImage
🎯 What it does: Proposes ParamISP for learning the forward and inverse transformations of camera ISP, utilizing EXIF parameters to achieve high-quality reconstruction of RAW and sRGB images.
PaReNeRF: Toward Fast Large-scale Dynamic NeRF with Patch-based Reference
Xiao Tang (Samsung Research Institute China Xi'an), Hojae Lee (Samsung Research Institute China Xi'an)
Autonomous DrivingComputational EfficiencyNeural Radiance FieldOptical FlowImageVideo
🎯 What it does: A dynamic NeRF model called PaReNeRF is proposed, which is based on a reference decoder. It employs patch sampling, low-resolution rendering with upsampling, and optical flow + structural similarity to search for reference information, achieving faster training and inference in large-scale dynamic scenes.
Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition
Anqi Zhu (University of Melbourne), James Bailey (University of Melbourne)
RecognitionGraph Neural NetworkLarge Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: A zero-shot action recognition method called PURLS is proposed, which enhances the recognition performance of unseen actions by aligning local and global semantics of language and skeleton data.
PartDistill: 3D Shape Part Segmentation by Vision-Language Model Distillation
Ardian Umam (National Yang Ming Chiao Tung University), Yen-Yu Lin (National Yang Ming Chiao Tung University)
SegmentationKnowledge DistillationVision Language ModelPoint Cloud
🎯 What it does: This paper proposes a cross-modal knowledge distillation framework called PartDistill, which utilizes knowledge from 2D visual language models to achieve 3D shape part segmentation.
Partial-to-Partial Shape Matching with Geometric Consistency
Viktoria Ehm (Technical University of Munich), Florian Bernard (University of Bonn)
OptimizationMesh
🎯 What it does: A geometrically consistent part-to-part shape matching framework is proposed, which can simultaneously predict overlapping regions and provide correspondences.
PaSCo: Urban 3D Panoptic Scene Completion with Uncertainty Awareness
Anh-Quan Cao (Inria), Raoul de Charette (Inria)
Object DetectionSegmentationAutonomous DrivingConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: This paper proposes the Panoptic Scene Completion (PSC) task for sparse LiDAR point clouds, which can simultaneously recover complete scene geometry, semantic labels, and instance segmentation, while estimating uncertainty at both voxel and instance levels.
Passive Snapshot Coded Aperture Dual-Pixel RGB-D Imaging
Bhargav Ghanekar (Rice University), Ashok Veeraraghavan (Rice University)
Depth EstimationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: The CADS (Coded Aperture Dual-Pixel Sensing) technology is proposed, achieving RGB-D imaging with a single snapshot, and its feasibility has been validated on prototypes such as DSLR, 3D endoscopes, and dermatoscopes.
Patch2Self2: Self-supervised Denoising on Coresets via Matrix Sketching
Shreyas Fadnavis (Johnson and Johnson), Eleftherios Garyfallidis (Indiana University)
RestorationExplainability and InterpretabilityComputational EfficiencyBiomedical DataMagnetic Resonance ImagingDiffusion Tensor Imaging
🎯 What it does: Patch2Self2 is proposed, achieving self-supervised denoising by matrix sketching (co-reset) of the Casorati matrix of Patch2Self.
PatchFusion: An End-to-End Tile-Based Framework for High-Resolution Monocular Metric Depth Estimation
Zhenyu Li (King Abdullah University of Science and Technology), Peter Wonka (King Abdullah University of Science and Technology)
Depth EstimationTransformerImage
🎯 What it does: PatchFusion proposes an end-to-end tile-based high-resolution monocular depth estimation framework.
PBWR: Parametric-Building-Wireframe Reconstruction from Aerial LiDAR Point Clouds
Shangfeng Huang (University of Calgary), Hongxin Yang (University of Calgary)
SegmentationGenerationTransformerPoint Cloud
🎯 What it does: This paper proposes an end-to-end PBWR method that directly regresses parameterized building edges from aerospace LiDAR point clouds and removes redundant edges through E-NMS, ultimately generating a complete building framework model.
PDF: A Probability-Driven Framework for Open World 3D Point Cloud Semantic Segmentation
Jinfeng Xu (Huazhong University of Science and Technology), Min Chen (Huazhong University of Science and Technology)
SegmentationKnowledge DistillationTransformerPoint Cloud
🎯 What it does: This paper proposes a probability-based framework (PDF) to address the open-world semantic segmentation problem of 3D point clouds;
PEEKABOO: Interactive Video Generation via Masked-Diffusion
Yash Jain (Microsoft), Harkirat Behl
GenerationData SynthesisDiffusion modelVideoBenchmark
🎯 What it does: A training-free, zero-latency masked attention module named PEEKABOO is proposed, enabling interactive spatiotemporal control for existing text-to-video diffusion models based on 3D-UNet.
PeerAiD: Improving Adversarial Distillation from a Specialized Peer Tutor
Jaewon Jung (Seoul National University), Jinho Lee (Seoul National University)
Knowledge DistillationAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes PeerAiD, an online adversarial distillation method that utilizes peer networks to specifically generate adversarial samples targeting the student model, thereby enhancing the robustness and natural accuracy of the student network.
PEGASUS: Personalized Generative 3D Avatars with Composable Attributes
Hyunsoo Cha (Seoul National University), Hanbyul Joo (Seoul National University)
GenerationData SynthesisDiffusion modelVideoPoint Cloud
🎯 What it does: This paper proposes the PEGASUS method, which can construct customizable 3D facial avatars using only monocular video, and supports continuous and separable editing of facial attributes such as hairstyle, nose, and hat, while maintaining the target identity.
PELA: Learning Parameter-Efficient Models with Low-Rank Approximation
Yangyang Guo (National University of Singapore), Mohan Kankanhalli (National University of Singapore)
Object DetectionSegmentationCompressionKnowledge DistillationTransformerImage
🎯 What it does: By using low-rank decomposition during the pre-training phase combined with feature distillation and weight perturbation regularization, an efficient Transformer model is obtained with significantly reduced parameters that can be directly used for downstream fine-tuning.
PeLK: Parameter-efficient Large Kernel ConvNets with Peripheral Convolution
Honghao Chen (Institute of Automation, Chinese Academy of Sciences), Kaiqi Huang (Institute of Automation, Chinese Academy of Sciences)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: Proposed peripheral convolution inspired by human eye perception, reducing the parameter count of large convolution kernels and constructing the PeLK network, achieving dense convolutions up to 101×101.
PEM: Prototype-based Efficient MaskFormer for Image Segmentation
Niccolò Cavagnero (Politecnico di Torino), Fabio Cermelli (Focoos AI)
SegmentationTransformerImage
🎯 What it does: An end-to-end efficient Transformer architecture (PEM) is proposed, which supports both semantic segmentation and panoptic segmentation.
PerAda: Parameter-Efficient Federated Learning Personalization with Generalization Guarantees
Chulin Xie (University of Illinois at Urbana-Champaign), Anima Anandkumar (University of Chicago)
Federated LearningKnowledge DistillationImage
🎯 What it does: A parameter-efficient federated learning framework called PERADA is proposed, which utilizes adapters and knowledge distillation to achieve personalized models under data heterogeneity and enhance generalization ability against distribution shifts.
Perception-Oriented Video Frame Interpolation via Asymmetric Blending
Guangyang Wu (Shanghai Jiao Tong University), Qingqing Zheng (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences)
RestorationGenerationOptical FlowVideo
🎯 What it does: This paper proposes a perception-based intermediate frame interpolation method called PerVFI, focusing on addressing the blurriness and ghosting issues produced by traditional methods.
PerceptionGPT: Effectively Fusing Visual Perception into LLM
Renjie Pi (Hong Kong University of Science and Technology), Tong Zhang (Hong Kong University of Science and Technology)
Object DetectionSegmentationRetrievalTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: An end-to-end framework called PerceptionGPT has been developed, which can directly integrate visual perception signals (such as detection boxes and segmentation masks) into large language models.
Perceptual Assessment and Optimization of HDR Image Rendering
Peibei Cao (City University of Hong Kong), Kede Ma (City University of Hong Kong)
OptimizationNeural Radiance FieldImage
🎯 What it does: A method is proposed to decompose HDR images into multiple exposure LDR images based on an inverse display model, and to quantitatively evaluate HDR images using existing LDR quality assessment metrics, while using this evaluation as a perceptual loss for optimizing HDR image rendering.
Permutation Equivariance of Transformers and Its Applications
Hengyuan Xu (Shanghai Jiao Tong University), Baochun Li (University of Toronto)
ClassificationSegmentationGenerationSafty and PrivacyTransformerImageText
🎯 What it does: This paper proposes and proves the 'Permutation Equivariance' property of Transformers, which supports both cross-token exchanges and internal dimension permutations during forward and backward propagation, and demonstrates that this property remains invariant in existing models such as ViT, BERT, and GPT. Based on this property, a privacy-enhanced split-learning scheme and model encryption/authorization mechanism are constructed.
Person in Place: Generating Associative Skeleton-Guidance Maps for Human-Object Interaction Image Editing
ChangHee Yang (Sogang University), Suk-Ju Kang (Sogang University)
Image TranslationGenerationGraph Neural NetworkDiffusion modelImage
🎯 What it does: A framework for automatic skeleton generation based on associative attention is proposed for human-object interaction (HOI) image editing, and the generated skeleton is directly projected as a control signal into existing skeleton-guided generation models.
Person-in-WiFi 3D: End-to-End Multi-Person 3D Pose Estimation with Wi-Fi
Kangwei Yan (Xi'an Jiaotong University), Xing Wei (Xi'an Jiaotong University)
Pose EstimationTransformerPoint Cloud
🎯 What it does: We propose Person-in-WiFi 3D, an end-to-end multi-person 3D human pose estimation system that utilizes Wi-Fi CSI signals and Transformers for pose prediction.
Personalized Residuals for Concept-Driven Text-to-Image Generation
Cusuh Ham (Georgia Institute of Technology), Tobias Hinz (Adobe Research)
GenerationTransformerDiffusion modelImageText
🎯 What it does: A concept-driven text-to-image generation method is proposed that utilizes LoRA to learn low-rank residuals and achieves this through Local Attention Guided Sampling (LAG), capable of generating high-quality, identity-preserving images for any concept within 3 minutes on a single GPU.
Perturbing Attention Gives You More Bang for the Buck: Subtle Imaging Perturbations That Efficiently Fool Customized Diffusion Models
Jingyao Xu (Beijing Jiaotong University), Xiang Wei (Beijing Jiaotong University)
GenerationAdversarial AttackTransformerDiffusion modelImage
🎯 What it does: A method for adversarial attacks on customized diffusion models, called CAAT, is proposed. By adding small perturbations to the cross-attention layer, the model fails to generate fake images.
PeVL: Pose-Enhanced Vision-Language Model for Fine-Grained Human Action Recognition
Haosong Zhang (Institute for Infocomm Research), Weisi Lin (Institute for Infocomm Research)
RecognitionTransformerVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes the Pose-Enhanced Vision-Language (PeVL) model, which integrates human pose information into a pre-trained Vision-Language foundation model specifically for fine-grained human action recognition.
PFStorer: Personalized Face Restoration and Super-Resolution
Tuomas Varanka (University of Oulu), Erman Acar (Huawei Finland)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: Utilizing a small number of high-quality reference images, the diffusion model is fine-tuned into a facial restoration model, and personalized recovery is achieved through an adapter.
PH-Net: Semi-Supervised Breast Lesion Segmentation via Patch-wise Hardness
Siyao Jiang (Shenzhen University), Jing Qin (Hong Kong Polytechnic University)
SegmentationContrastive LearningImageUltrasound
🎯 What it does: A semi-supervised learning-based breast ultrasound image lesion segmentation framework, PH-Net, is proposed, which improves segmentation accuracy by utilizing data with a lack of pixel annotations.
Photo-SLAM: Real-time Simultaneous Localization and Photorealistic Mapping for Monocular Stereo and RGB-D Cameras
Huajian Huang (Hong Kong University of Science and Technology), Sai-Kit Yeung (Hong Kong University of Science and Technology)
Pose EstimationOptimizationComputational EfficiencyGaussian SplattingSimultaneous Localization and MappingMultimodality
🎯 What it does: This paper proposes the Photo-SLAM framework, achieving real-time simultaneous localization and optical realistic mapping based on super-primitive.
PhotoMaker: Customizing Realistic Human Photos via Stacked ID Embedding
Zhen Li (Nankai University), Ying Shan (Tencent)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes PhotoMaker, a one-step personalized text-to-image generation framework achieved through stacked ID embedding;
PhyScene: Physically Interactable 3D Scene Synthesis for Embodied AI
Yandan Yang (National Key Laboratory of General Artificial Intelligence, BIGAI), Siyuan Huang (National Key Laboratory of General Artificial Intelligence, BIGAI)
GenerationData SynthesisRobotic IntelligenceDiffusion modelMesh
🎯 What it does: This paper presents PHYSCENE, a method for indoor scene synthesis based on conditional diffusion models, capable of generating 3D scenes with physical interactivity.
PhysGaussian: Physics-Integrated 3D Gaussians for Generative Dynamics
Tianyi Xie (University of California Los Angeles), Chenfanfu Jiang (University of California Los Angeles)
GenerationData SynthesisComputational EfficiencyNeural Radiance FieldGaussian SplattingMeshPhysics Related
🎯 What it does: A unified simulation rendering framework called PhysGaussian based on 3D Gaussian splatting is proposed, which allows physical dynamics and rendering to share the same Gaussian kernel, directly generating high-quality motion synthesis.
Physical 3D Adversarial Attacks against Monocular Depth Estimation in Autonomous Driving
Junhao Zheng (Xi'an Jiaotong University), Chao Shen (Xi'an Jiaotong University)
Depth EstimationAutonomous DrivingAdversarial AttackImage
🎯 What it does: This paper proposes 3D Depth Fool, which utilizes fully covered 3D textures to physically attack monocular depth estimation models.
Physical Backdoor: Towards Temperature-based Backdoor Attacks in the Physical World
Wen Yin (Huazhong University of Science and Technology), Lichao Sun (Lehigh University)
Object DetectionAdversarial AttackImageVideo
🎯 What it does: The study investigates physical backdoor attacks in thermal infrared object detection (TIOD), proposing two attack methods (OAA, RAA) and validating their effectiveness in both digital and real-world scenarios.
Physical Property Understanding from Language-Embedded Feature Fields
Albert J. Zhai (University of Illinois at Urbana-Champaign), Shenlong Wang (University of Illinois at Urbana-Champaign)
Large Language ModelVision Language ModelNeural Radiance FieldContrastive LearningPoint CloudPhysics Related
🎯 What it does: Using NeRF point clouds constructed from multi-view images and integrating CLIP visual-language features, a candidate material dictionary is generated with the help of LLM, and zero-shot CLIP kernel regression is employed to predict the physical properties of each point, achieving dense physical property predictions for any object.
Physics-Aware Hand-Object Interaction Denoising
Haowen Luo (Tsinghua University), Li Yi (Tsinghua University)
RestorationRobotic IntelligenceRecurrent Neural NetworkAuto EncoderPoint CloudPhysics Related
🎯 What it does: A physics-aware hand-object interaction denoising framework is proposed, which performs denoising through physical constraints on hand pose sequences.
Physics-guided Shape-from-Template: Monocular Video Perception through Neural Surrogate Models
David Stotko (University of Bonn), Reinhard Klein (University of Bonn)
OptimizationVideoMeshPhysics Related
🎯 What it does: A physics-guided shape recovery method is proposed, utilizing monocular RGB videos and template meshes, combined with differentiable physical simulation and rendering, to achieve joint reconstruction of fabric geometry and physical parameters.
PhysPT: Physics-aware Pretrained Transformer for Estimating Human Dynamics from Monocular Videos
Yufei Zhang (Rensselaer Polytechnic Institute), Qiang Ji (Rensselaer Polytechnic Institute)
RecognitionPose EstimationTransformerVideoPhysics Related
🎯 What it does: Utilizing a pre-trained Transformer framework in monocular video, combined with physical knowledge, to achieve refined estimation of human 3D motion trajectories and infer joint driving forces and contact forces.
PI3D: Efficient Text-to-3D Generation with Pseudo-Image Diffusion
Ying-Tian Liu (Tsinghua University), Song-Hai Zhang (Qinghai University)
GenerationData SynthesisDiffusion modelScore-based ModelMesh
🎯 What it does: Proposes the PI3D framework, which utilizes a pre-trained text-image diffusion model to generate 3D shapes through pseudo-image generation and refines them quickly using Score Distillation.
PIA: Your Personalized Image Animator via Plug-and-Play Modules in Text-to-Image Models
Yiming Zhang (Shanghai Artificial Intelligence Laboratory), Kai Chen (Shanghai Artificial Intelligence Laboratory)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: PIA is proposed, a plugin-based personalized image animation framework that converts any personalized text-to-image model into video animations.
Pick-or-Mix: Dynamic Channel Sampling for ConvNets
Ashish Kumar (Indian Institute of Technology Kanpur), Laxmidhar Behera (Indian Institute of Technology Kanpur)
ClassificationSegmentationCompressionComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: A dynamic channel sampling module named PiX is proposed, which can dynamically select channels on a per-pixel basis, replacing traditional 1×1 convolutions, and achieving multifunctional capabilities such as channel compression, network scaling, and dynamic pruning.
PICTURE: PhotorealistIC virtual Try-on from UnconstRained dEsigns
Shuliang Ning (Chinese University of Hong Kong), Xiaoguang Han (Chinese University of Hong Kong)
Image TranslationGenerationTransformerSupervised Fine-TuningDiffusion modelImageTextMultimodality
🎯 What it does: The ucVTON task is proposed, allowing users to specify clothing styles through images or text, using full clothing images, cropped parts, or image patches to specify textures, achieving free personalized synthesis for virtual try-ons.
PIE-NeRF: Physics-based Interactive Elastodynamics with NeRF
Yutao Feng (Zhejiang University), Yin Yang (University of Utah)
GenerationData SynthesisComputational EfficiencyNeural Radiance FieldImagePhysics Related
🎯 What it does: A physics-driven elastic dynamics framework based on NeRF, called PIE-NeRF, is proposed, which reduces computational load through meshless sampling and quadratic GMLS, achieving real-time interactive deformation synthesis.
PIGEON: Predicting Image Geolocations
Lukas Haas (Stanford University), Chelsea Finn (Stanford University)
RetrievalOptimizationTransformerContrastive LearningImage
🎯 What it does: Trained and evaluated two geographic localization models, PIGEON and PIGEOTTO, which achieved global image localization by combining semantic geographic units, haversine smooth labels, multi-task contrastive pre-training of CLIP, and cross-unit retrieval refinement.
PikeLPN: Mitigating Overlooked Inefficiencies of Low-Precision Neural Networks
Marina Neseem (Brown University), Daniele Moro (Google)
Computational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: The ACE v2 evaluation metric is proposed, and a fully quantized PikeLPN model is designed, significantly reducing the arithmetic energy consumption of low-precision networks.
PIN: Positional Insert Unlocks Object Localisation Abilities in VLMs
Michael Dorkenwald (University of Amsterdam), Yuki M. Asano (University of Amsterdam)
Object DetectionSegmentationTransformerVision Language ModelDiffusion modelImage
🎯 What it does: This work proposes a lightweight spatial prompt called Positional Insert (PIN) that enables zero-shot object localization without modifying the weights of the frozen caption-based Vision-Language Model (VLM) and without using any annotated detection data.
Pink: Unveiling the Power of Referential Comprehension for Multi-modal LLMs
Shiyu Xuan (Peking University), Shiliang Zhang (Ant Group)
RecognitionObject DetectionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: A high-quality instruction tuning dataset was constructed by designing various reference reasoning (RC) tasks and utilizing a self-consistent bootstrapping method. Based on this, parameter-efficient tuning of the visual encoder of a multimodal large language model was performed, significantly improving the model's performance on fine-grained image understanding tasks.
pix2gestalt: Amodal Segmentation by Synthesizing Wholes
Ege Ozguroglu (Columbia University), Carl Vondrick (Columbia University)
Object DetectionSegmentationGenerationDiffusion modelImage
🎯 What it does: This paper proposes the pix2gestalt framework, which utilizes a pre-trained diffusion model for zero-shot complete object reconstruction and segmentation of invisible parts.
Pixel-Aligned Language Model
Jiarui Xu (Google Research), Cordelia Schmid (Google Research)
SegmentationGenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This paper presents PixelLLM, a model that combines large language models with visual encoders, capable of generating image descriptions while outputting the corresponding pixel positions for each word.
Pixel-level Semantic Correspondence through Layout-aware Representation Learning and Multi-scale Matching Integration
Yixuan Sun (Fudan University), Wenqiang Zhang (Fudan University)
SegmentationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: A high-resolution semantic correspondence framework LPMFlow is designed based on layout-aware representation learning, progressive feature super-resolution, and multi-scale matching flow fusion for pixel-level semantic correspondence.
PixelLM: Pixel Reasoning with Large Multimodal Model
Zhongwei Ren (Beijing Jiaotong University), Xiaojie Jin (ByteDance Inc.)
Object DetectionSegmentationComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: We propose PixelLM, a multimodal model that supports multi-target pixel-level inference and can generate multiple target masks in a single inference.
PixelRNN: In-pixel Recurrent Neural Networks for End-to-end-optimized Perception with Neural Sensors
Haley M. So (Stanford University), Gordon Wetzstein (Stanford University)
RecognitionCompressionComputational EfficiencyRecurrent Neural NetworkImage
🎯 What it does: Designed and implemented a pixel-level recurrent neural network PixelRNN for extracting spatiotemporal features on sensor processors, significantly reducing transmission bandwidth.
pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction
David Charatan (Massachusetts Institute of Technology), Vincent Sitzmann (Simon Fraser University)
GenerationDepth EstimationOptimizationComputational EfficiencyTransformerGaussian SplattingImage
🎯 What it does: We propose PixelSplat, a forward model that utilizes two images to predict an editable 3D radiance field based on 3D Gaussian primitives in a single forward pass, achieving real-time renderable general novel view synthesis;
PKU-DyMVHumans: A Multi-View Video Benchmark for High-Fidelity Dynamic Human Modeling
Xiaoyun Zheng (Peking University), Ronggang Wang (Peking University)
SegmentationGenerationPose EstimationNeural Radiance FieldVideoBenchmark
🎯 What it does: The PKU-DyMVHumans dynamic multi-view human dataset (8.2M frames, 32 actors, 45 scenes) has been proposed and made publicly available, along with a unified benchmark experimental pipeline for high-fidelity human reconstruction and rendering.
PLACE: Adaptive Layout-Semantic Fusion for Semantic Image Synthesis
Zhengyao Lv (Harbin Institute of Technology), Kwan-Yee K. Wong (University of Hong Kong)
SegmentationGenerationData SynthesisDiffusion modelImage
🎯 What it does: The PLACE module is proposed, achieving adaptive layout-semantic fusion based on pre-trained Stable Diffusion for high-quality semantic image synthesis.
PlatoNeRF: 3D Reconstruction in Plato's Cave via Single-View Two-Bounce Lidar
Tzofi Klinghoffer (Massachusetts Institute of Technology), Rakesh Ranjan (Meta)
Data SynthesisDepth EstimationNeural Radiance FieldPoint Cloud
🎯 What it does: A method called PlatoNeRF is proposed, which utilizes single-view double-bounce single-photon radar time-domain measurements combined with NeRF for 3D geometric reconstruction, capable of simultaneously recovering the absolute depth of visible and occluded areas.
PLGSLAM: Progressive Neural Scene Represenation with Local to Global Bundle Adjustment
Tianchen Deng (Shanghai Jiao Tong University), Weidong Chen (Shanghai Jiao Tong University)
OptimizationRobotic IntelligenceNeural Radiance FieldSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper presents PLGSLAM, a neural visual SLAM system that combines progressive scene representation and local-to-global bundle adjustment for high-fidelity 3D reconstruction and real-time camera tracking in large-scale indoor environments.
Plug and Play Active Learning for Object Detection
Chenhongyi Yang (University of Edinburgh), Elliot J. Crowley (University of Edinburgh)
Object DetectionImage
🎯 What it does: A two-stage plugin-based active learning framework called PPAL is proposed, which first selects candidate images based on difficulty-corrected uncertainty sampling, and then performs diversity screening of multi-instance images through Category Conditional Matching Similarity (CCMS);
Plug-and-Play Diffusion Distillation
Yi-Ting Hsiao (University of Michigan), Ratheesh Kalarot (Adobe Inc.)
GenerationKnowledge DistillationDiffusion modelImage
🎯 What it does: A lightweight guide model is proposed, which trains this model while keeping the original diffusion model unchanged, achieving guidance for the base model and thus distilling classifier-free guidance (CFG), balancing speed and quality.
PNeRV: Enhancing Spatial Consistency via Pyramidal Neural Representation for Videos
Qi Zhao (Nanjing University), Zhan Ma (Nanjing University)
RestorationConvolutional Neural NetworkVideo
🎯 What it does: Proposes PNeRV, which enhances the spatial consistency of videos through a pyramid structure.
POCE: Primal Policy Optimization with Conservative Estimation for Multi-constraint Offline Reinforcement Learning
Jiayi Guan (Tongji University), Changjun Jiang (Tongji University)
OptimizationReinforcement LearningTabular
🎯 What it does: A multi-constraint offline reinforcement learning algorithm POCE is proposed, which can simultaneously satisfy cumulative cost and state step cost constraints.
Point Cloud Pre-training with Diffusion Models
Xiao Zheng (Shandong University), Yongshun Gong (Shandong University)
ClassificationObject DetectionSegmentationTransformerDiffusion modelPoint Cloud
🎯 What it does: This paper proposes a point cloud pre-training framework called PointDif based on diffusion models, which learns the hierarchical geometric priors of point clouds by progressively denoising noisy point clouds at different noise levels.
Point Segment and Count: A Generalized Framework for Object Counting
Zhizhong Huang (Fudan University), Hongming Shan (Fudan University)
Object DetectionKnowledge DistillationPrompt EngineeringImage
🎯 What it does: A general framework called PseCo based on point-segmentation-counting is proposed for few-shot and zero-shot object counting/detection.
Point Transformer V3: Simpler Faster Stronger
Xiaoyang Wu (Hong Kong University), Hengshuang Zhao (Hong Kong University)
Object DetectionSegmentationAutonomous DrivingComputational EfficiencyTransformerPoint Cloud
🎯 What it does: Proposed and implemented Point Transformer V3, a more concise, efficient, and scalable point cloud Transformer backbone network.
Point-VOS: Pointing Up Video Object Segmentation
Sabarinath Mahadevan (RWTH Aachen University), Bastian Leibe (RWTH Aachen University)
Object DetectionSegmentationVideoMultimodalityBenchmark
🎯 What it does: This paper proposes the Point-VOS task for Video Object Segmentation based on sparse point annotations, and constructs two large-scale multimodal point-annotated video datasets (PV-Oops, PV-Kinetics) along with corresponding benchmarks.
Point2CAD: Reverse Engineering CAD Models from 3D Point Clouds
Yujia Liu (ETH Zurich), Konrad Schindler (ETH Zurich)
SegmentationGenerationPoint CloudMesh
🎯 What it does: A complete pipeline for reverse generating CAD models from 3D point clouds is proposed.
Point2RBox: Combine Knowledge from Synthetic Visual Patterns for End-to-end Oriented Object Detection with Single Point Supervision
Yi Yu (Harbin Institute of Technology), Junchi Yan (Shanghai Jiao Tong University)
Object DetectionImage
🎯 What it does: Proposes an end-to-end single-point supervised slanted object detection method Point2RBox, utilizing synthetic visual patterns and transformation self-supervision to achieve RBox detection.
PointBeV: A Sparse Approach for BeV Predictions
Loick Chambon (Valeo), Matthieu Cord (Valeo)
SegmentationAutonomous DrivingPoint Cloud
🎯 What it does: A sparse bird's-eye view (BEV) segmentation model called PointBeV is proposed, which utilizes sparse point sets instead of full grid maps for BEV feature extraction and prediction.
PointInfinity: Resolution-Invariant Point Diffusion Models
Zixuan Huang (Meta), Chao-Yuan Wu (Meta)
GenerationData SynthesisTransformerDiffusion modelPoint Cloud
🎯 What it does: PointInfinity is proposed, a resolution-invariant point cloud diffusion model that can be trained with low resolution (e.g., 1024 points) and generate high-quality point clouds of up to 131k points during inference.
PointOBB: Learning Oriented Object Detection via Single Point Supervision
Junwei Luo (Wuhan University), Yansheng Li (Wuhan University)
Object DetectionImage
🎯 What it does: A direction target detection framework under single-point supervision, PointOBB, is proposed, which transforms single point annotations into oriented bounding boxes.
Polarization Wavefront Lidar: Learning Large Scene Reconstruction from Polarized Wavefronts
Dominik Scheuble (Mercedes-Benz AG), Felix Heide (Princeton University)
Depth EstimationAutonomous DrivingTransformerSupervised Fine-TuningPoint Cloud
🎯 What it does: Develop a long-distance polarized wavefront lidar (PolLidar) and train a neural network based on raw wavefront data to achieve dense distance and normal reconstruction.
PolarMatte: Fully Computational Ground-Truth-Quality Alpha Matte Extraction for Images and Video using Polarized Screen Matting
Kenji Enomoto (Adobe Research), Gavin Miller (Adobe Research)
RestorationSegmentationImageVideo
🎯 What it does: This paper proposes a method for extracting polarization matrices called PolarMatte, which can automatically generate high-quality alpha masks for images and videos without human intervention.
PolarRec: Improving Radio Interferometric Data Reconstruction Using Polar Coordinates
Ruoqi Wang (Hong Kong University of Science and Technology), Feng Wang (Guangzhou University)
RestorationGenerationComputational EfficiencyTransformerNeural Radiance FieldImage
🎯 What it does: This paper proposes a polar coordinate-based visibility data reconstruction method called PolarRec, which can interpolate sparse radio interferometric visibility data to achieve full frequency coverage and generate high-quality astronomical images.
Polos: Multimodal Metric Learning from Human Feedback for Image Captioning
Yuiga Wada (Keio University), Komei Sugiura (Keio University)
GenerationRetrievalTransformerContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes Polos, a supervised image caption evaluation metric based on multimodal human feedback learning;
Poly Kernel Inception Network for Remote Sensing Detection
Xinhao Cai (Nanjing University of Science and Technology), Yazhou Yao (Nanjing University of Science and Technology)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a lightweight feature extraction backbone network named Poly Kernel Inception Network (PKINet) to enhance the performance of object detection in remote sensing images.
PoNQ: a Neural QEM-based Mesh Representation
Nissim Maruani (Inria), Mathieu Desbrun (Inria)
GenerationOptimizationMesh
🎯 What it does: A new learnable 3D shape representation called PoNQ is proposed, which uses points, normal vectors, and the Quadratic Error Metric (QEM) to jointly describe local geometry, and directly generates non-intersecting, watertight meshes through Delaunay tetrahedralization and graph cut algorithms.
POPDG: Popular 3D Dance Generation with PopDanceSet
Zhenye Luo (Beijing Normal University), Li Yao (Beijing Normal University)
GenerationTransformerDiffusion modelVideo
🎯 What it does: This paper proposes a new popular dance dataset called PopDanceSet and constructs the POPDG model based on iDDPM, utilizing a spatial enhancement algorithm and alignment module to generate 3D dances that are highly synchronized and diverse with music.
Portrait4D: Learning One-Shot 4D Head Avatar Synthesis using Synthetic Data
Yu Deng, Baoyuan Wang
GenerationData SynthesisTransformerNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a one-step four-dimensional (4D) head avatar generation method trained using large-scale synthetic data. It first learns a 4D generative model, GenHead, to produce multi-view, full-motion synthetic avatars, and then utilizes a Transformer-based animatable three-plane reconstructor, Portrait4D, to achieve complete animation of the face, eyes, mouth, and neck from a single image, along with foreground-background separation.
PortraitBooth: A Versatile Portrait Model for Fast Identity-preserved Personalization
Xu Peng (Xiamen University), Rongrong Ji (Xiamen University)
GenerationData SynthesisDiffusion modelImageVideoText
🎯 What it does: A one-shot text-to-portrait generation framework called PortraitBooth is proposed, which can quickly generate personalized, expression-editable, and identity-preserving portraits without model fine-tuning.
Pose Adapted Shape Learning for Large-Pose Face Reenactment
Gee-Sern Jison Hsu (National Taiwan University of Science and Technology), Wei-Jie Hong (National Taiwan University of Science and Technology)
Image TranslationGenerationPose EstimationTransformerGenerative Adversarial NetworkImage
🎯 What it does: A three-module framework (Pose-Adapted Encoder, Cycle-consistent Shape Generator, and Attention-Embedded Generator) is designed to achieve large-angle face reenactment.
Pose-Guided Self-Training with Two-Stage Clustering for Unsupervised Landmark Discovery
Siddharth Tourani, Muhammad Haris Khan
Object DetectionPose EstimationDiffusion modelAuto EncoderContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised landmark detection framework based on a pre-trained diffusion model (Stable Diffusion). It first presents a zero-shot clustering baseline, which is then further improved through self-training and two-stage clustering, ultimately resulting in the D-ULD++ algorithm.
Pose-Transformed Equivariant Network for 3D Point Trajectory Prediction
Ruixuan Yu (Shandong University), Jian Sun (Xi'an Jiaotong University)
Pose EstimationAutonomous DrivingGraph Neural NetworkPoint Cloud
🎯 What it does: A pose transformation-based equivariant network PT-EvNet is proposed for predicting 3D point trajectories.
PoseIRM: Enhance 3D Human Pose Estimation on Unseen Camera Settings via Invariant Risk Minimization
Yanlu Cai (Fudan University), Cheng Jin (Fudan University)
Pose EstimationTransformerImage
🎯 What it does: This paper proposes a multi-view 3D human pose estimation framework called PoseIRM based on Invariant Risk Minimization (IRM). By synthesizing 2D pose data from various camera setups and using IRM constraints during training, the model can maintain good performance even under unseen camera configurations.
Positive-Unlabeled Learning by Latent Group-Aware Meta Disambiguation
Lin Long (Zhejiang University), Junbo Zhao (Zhejiang University)
Representation LearningMeta LearningContrastive LearningImageBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: This paper proposes a new positive/unlabeled learning framework called LaGAM, which combines hierarchical contrastive learning and meta-learning to achieve label disambiguation and high-quality representation learning.
Posterior Distillation Sampling
Juil Koo (KAIST), Minhyuk Sung (KAIST)
Image TranslationGenerationOptimizationDiffusion modelNeural Radiance FieldImage
🎯 What it does: A parameterized image editing method based on Posterior Distillation Sampling (PDS) is proposed, which enables text-guided editing while maintaining the identity of the original image.
PostureHMR: Posture Transformation for 3D Human Mesh Recovery
Yu-Pei Song (Southwest Jiaotong University), Qiang Peng (Southwest Jiaotong University)
Pose EstimationTransformerDiffusion modelMesh
🎯 What it does: This paper proposes PostureHMR, a 3D human mesh recovery framework that gradually transforms from the SMPL T-pose to the target pose.
Practical Measurements of Translucent Materials with Inter-Pixel Translucency Prior
Zhenyu Chen (Nanjing University), Yanwen Guo (OPPO)
RestorationData SynthesisConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a low-cost method for measuring the scattering properties of translucent materials that relies solely on ordinary photographs.
PracticalDG: Perturbation Distillation on Vision-Language Models for Hybrid Domain Generalization
Zining Chen (Beijing University of Posts and Telecommunications), Hongying Meng (Brunel University)
Domain AdaptationKnowledge DistillationTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Proposed SCI-PD, which transfers knowledge from VLM to lightweight visual models through perturbation distillation from three perspectives: scoring, categories, and instances, achieving open set domain generalization.
PRDP: Proximal Reward Difference Prediction for Large-Scale Reward Finetuning of Diffusion Models
Fei Deng (Google), Matthias Grundmann (Google)
GenerationOptimizationReinforcement Learning from Human FeedbackTransformerReinforcement LearningDiffusion modelText
🎯 What it does: A stable black-box reward fine-tuning method named PRDP is proposed, which can perform reward maximization training on diffusion models with large-scale (over 100,000) text prompts;
Pre-trained Model Guided Fine-Tuning for Zero-Shot Adversarial Robustness
Sibo Wang (Institute of Computing Technology Chinese Academy of Sciences), Shiguang Shan (Institute of Computing Technology Chinese Academy of Sciences)
ClassificationAdversarial AttackTransformerSupervised Fine-TuningContrastive LearningImage
🎯 What it does: Conduct adversarial fine-tuning on the CLIP pre-trained model and add an auxiliary branch based on the original model features to enhance adversarial robustness under zero-shot conditions.
Pre-trained Vision and Language Transformers Are Few-Shot Incremental Learners
Keon-Hee Park (Kyung Hee University), Gyeong-Moon Park (Kyung Hee University)
ClassificationKnowledge DistillationRepresentation LearningTransformerPrompt EngineeringImage
🎯 What it does: A few-shot class incremental learning framework called PriViLege is proposed, based on a pre-trained visual and language Transformer, which significantly improves the model's incremental learning performance through pre-trained knowledge tuning, entropy-based divergence loss, and semantic knowledge distillation.
Pre-training Vision Models with Mandelbulb Variations
Benjamin Naoto Chiche (Rist Inc.), Ryo Fujita (Rist Inc.)
ClassificationAnomaly DetectionConvolutional Neural NetworkTransformerImage
🎯 What it does: A formula-driven unsupervised pre-training dataset based on 3D Mandelbulb fractal variations is proposed, and pre-training is conducted on CNN and ViT models to validate their effectiveness in classification and anomaly detection tasks.
Predicated Diffusion: Predicate Logic-Based Attention Guidance for Text-to-Image Diffusion Models
Kota Sueyoshi (Osaka University), Takashi Matsubara (Osaka University)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: The Predicated Diffusion framework is proposed, which converts text prompts into predicate logic propositions and uses attention maps as fuzzy predicates to provide differentiable guidance to the diffusion model, thereby generating more faithful images.
PredToken: Predicting Unknown Tokens and Beyond with Coarse-to-Fine Iterative Decoding
Xuesong Nie (Zhejiang University), Donglian Qi (Zhejiang University)
GenerationData SynthesisAnomaly DetectionTransformerGenerative Adversarial NetworkVideoTime Series
🎯 What it does: The PredToken framework is proposed, transforming spatiotemporal prediction from frame-level to token-level, and improving token representation and coarse-fine iterative decoding through the DQR structure to enhance prediction quality.