CVPR 2023 Papers — Page 17
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
Physical-World Optical Adversarial Attacks on 3D Face Recognition
Yanjie Li (Hong Kong Polytechnic University), Bin Xiao (Hong Kong Polytechnic University)
RecognitionAdversarial AttackPoint CloudMesh
🎯 What it does: Using structured light projection to generate optical noise for physical world adversarial attacks on 3D facial recognition systems.
Physically Adversarial Infrared Patches With Learnable Shapes and Locations
Xingxing Wei (Beihang University), Yao Huang (Beihang University)
Object DetectionAdversarial AttackImage
🎯 What it does: Proposes a method for infrared attacks that can be used in the real world - adversarial infrared patches that can learn shapes and positions;
Physically Realizable Natural-Looking Clothing Textures Evade Person Detectors via 3D Modeling
Zhanhao Hu (Tsinghua University), Xiaolin Hu (Tsinghua University)
Object DetectionAdversarial AttackImage
🎯 What it does: The study designed a 3D modeling-based clothing adversarial texture (AdvCaT) to allow clothing worn by individuals to evade human detectors from multiple perspectives.
Physics-Driven Diffusion Models for Impact Sound Synthesis From Videos
Kun Su (University of Washington), Chuang Gan (University of Massachusetts Amherst)
GenerationData SynthesisTransformerDiffusion modelVideoPhysics RelatedAudio
🎯 What it does: A physics-driven diffusion model is proposed to automatically synthesize high-fidelity impact sounds from silent videos.
Physics-Guided ISO-Dependent Sensor Noise Modeling for Extreme Low-Light Photography
Yue Cao (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
RestorationFlow-based ModelImagePhysics Related
🎯 What it does: This study investigates the formation mechanism of camera noise in extremely low light environments, proposing a physics-guided ISO-dependent noise model implemented within a normalizing flow framework.
Pic2Word: Mapping Pictures to Words for Zero-Shot Composed Image Retrieval
Kuniaki Saito (Boston University), Tomas Pfister (Google Cloud AI Research)
RetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: The Pic2Word model is proposed, which utilizes weakly labeled image-caption pairs and unlabeled images to train under the CLIP framework, mapping images to language word vectors to achieve zero-shot synthesized image retrieval.
Picture That Sketch: Photorealistic Image Generation From Abstract Sketches
Subhadeep Koley, Yi-Zhe Song
Image TranslationGenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a model for generating realistic photos from freehand sketches.
PIDNet: A Real-Time Semantic Segmentation Network Inspired by PID Controllers
Jiacong Xu (Texas A&M University), Shankar P. Bhattacharyya (Texas A&M University)
SegmentationAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: A three-branch network called PIDNet has been developed, which integrates detail, context, and boundary information to achieve real-time semantic segmentation.
PillarNeXt: Rethinking Network Designs for 3D Object Detection in LiDAR Point Clouds
Jinyu Li (QCraft), Xiaodong Yang (QCraft)
Object DetectionAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper re-evaluates local aggregators in LiDAR point cloud 3D object detection and proposes the PillarNeXt network based on pillar encoders, combining 2D detection network architecture and training strategies, ultimately achieving state-of-the-art performance on Waymo and nuScenes.
PiMAE: Point Cloud and Image Interactive Masked Autoencoders for 3D Object Detection
Anthony Chen (Peking University), Shanghang Zhang (Beijing University of Posts and Telecommunications)
Object DetectionSegmentationTransformerAuto EncoderImageMultimodalityPoint Cloud
🎯 What it does: A cross-modal pre-training framework PiMAE based on masked autoencoders is proposed, which jointly learns representations of point clouds and RGB images and fine-tunes them under multiple tasks.
PIP-Net: Patch-Based Intuitive Prototypes for Interpretable Image Classification
Meike Nauta (University of Twente), Christin Seifert (University of Duisburg-Essen)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: An interpretable image classification network PIP-Net is proposed, which utilizes interpretable prototype segments to achieve class decisions.
PIRLNav: Pretraining With Imitation and RL Finetuning for ObjectNav
Ram Ramrakhya (Georgia Institute of Technology), Abhishek Das (Meta AI)
Robotic IntelligenceConvolutional Neural NetworkRecurrent Neural NetworkSupervised Fine-TuningReinforcement LearningImage
🎯 What it does: This paper proposes PIRLNav, a two-stage learning framework that first pre-trains using Behavior Cloning (BC) on a large number of human demonstrations and then fine-tunes using Reinforcement Learning (RL), significantly improving the success rate of ObjectGoal Navigation (OBJECTNAV).
PIVOT: Prompting for Video Continual Learning
Andrés Villa (Pontificia Universidad Católica de Chile), Bernard Ghanem (King Abdullah University of Science and Technology)
ClassificationRecognitionTransformerPrompt EngineeringContrastive LearningVideoMultimodalityBenchmark
🎯 What it does: We propose PIVOT, a method that utilizes a frozen CLIP image-text model and achieves end-to-end training for video class incremental learning (CIL) tasks through learnable spatial and temporal prompts and multimodal contrastive loss.
PivoTAL: Prior-Driven Supervision for Weakly-Supervised Temporal Action Localization
Mamshad Nayeem Rizve (University of Central Florida), Mei Chen (Microsoft)
RecognitionObject DetectionGaussian SplattingVideo
🎯 What it does: This paper proposes a weakly supervised temporal action localization method called PivoTAL, which adopts a training perspective from localization to localization, directly learning action segments, and utilizes action-specific scene priors, action segment generation priors, and learnable Gaussian priors to generate pseudo action segments for self-supervision.
Pix2map: Cross-Modal Retrieval for Inferring Street Maps From Images
Xindi Wu (Princeton University), Deva Ramanan (Carnegie Mellon University)
RetrievalAutonomous DrivingConvolutional Neural NetworkContrastive LearningImageGraph
🎯 What it does: This paper proposes a cross-modal retrieval method called Pix2Map, which directly retrieves corresponding urban street topology maps from onboard camera images, facilitating the update and expansion of high-precision HD maps.
Pixels, Regions, and Objects: Multiple Enhancement for Salient Object Detection
Yi Wang (Dalian University of Technology), Xiangjian He (University of Nottingham)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Achieving salient object detection in complex scenes through a Multi-level Enhancement Network (MENet).
PixHt-Lab: Pixel Height Based Light Effect Generation for Image Compositing
Yichen Sheng (Purdue University), Bedrich Benes (Purdue University)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: For the image synthesis task, we utilize pixel height representation to recover the geometry of the foreground and background, and based on this, achieve controllable rendering of lighting effects such as soft shadows and reflections.
PLA: Language-Driven Open-Vocabulary 3D Scene Understanding
Runyu Ding (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
RecognitionSegmentationTransformerVision Language ModelContrastive LearningImagePoint Cloud
🎯 What it does: This paper proposes a point cloud and language association framework (PLA) based on image bridging, utilizing a pre-trained vision-language model to generate multi-view image captions and performing hierarchical point-text alignment on point clouds, achieving 3D open vocabulary scene understanding through contrastive learning.
PlaneDepth: Self-Supervised Depth Estimation via Orthogonal Planes
Ruoyu Wang (ShanghaiTech University), Shenghua Gao (ShanghaiTech University)
Depth EstimationAutonomous DrivingNeural Radiance FieldImageVideo
🎯 What it does: The PlaneDepth network is proposed to achieve self-supervised monocular depth estimation, using a mixture of Laplace distributions of orthogonal planes (vertical planes and ground planes) to model scene depth.
Planning-Oriented Autonomous Driving
Yihan Hu (OpenDriveLab and OpenGVLab), Hongyang Li
Autonomous DrivingOptimizationTransformerMultimodality
🎯 What it does: This paper proposes a planning-oriented integrated autonomous driving framework called UniAD, which integrates perception, prediction, and planning within a unified Transformer system, using a query interface to connect various modules and achieve end-to-end learning.
Plateau-Reduced Differentiable Path Tracing
Michael Fischer (University College London), Tobias Ritschel (University College London)
OptimizationGaussian SplattingImage
🎯 What it does: Proposes convolution smoothing of the rendering equation in differentiable path tracing to eliminate flat gradient regions, making inverse rendering optimization more stable.
PlenVDB: Memory Efficient VDB-Based Radiance Fields for Fast Training and Rendering
Han Yan (Shanghai Jiao Tong University), Xing Mei (ByteDance Inc)
OptimizationComputational EfficiencyNeural Radiance FieldPoint Cloud
🎯 What it does: Directly train and render NeRF using VDB sparse volume structure, achieving efficient training and inference.
PLIKS: A Pseudo-Linear Inverse Kinematic Solver for 3D Human Body Estimation
Karthik Shetty (Friedrich Alexander University Erlangen-Nuremberg), Bernhard Egger (Friedrich Alexander University Erlangen-Nuremberg)
Pose EstimationConvolutional Neural NetworkGraph Neural NetworkImageMesh
🎯 What it does: This paper proposes a closed-form inverse kinematics solver called PLIKS, which utilizes a linearized SMPL model and pixel-aligned vertex mapping to reconstruct a 3D human mesh from a single 2D image.
Plug-and-Play Diffusion Features for Text-Driven Image-to-Image Translation
Narek Tumanyan (Weizmann Institute of Science), Tali Dekel (Weizmann Institute of Science)
Image TranslationGenerationDiffusion modelImageText
🎯 What it does: A new framework is proposed that extends text-guided image generation to image-to-image translation, utilizing a pre-trained text-to-image diffusion model to generate new images that align with the target text while preserving the semantic layout of the guiding image.
PMatch: Paired Masked Image Modeling for Dense Geometric Matching
Shengjie Zhu (Michigan State University), Xiaoming Liu (Michigan State University)
Object DetectionPose EstimationDepth EstimationTransformerImage
🎯 What it does: This paper proposes a paired mask image modeling (pMIM) pre-training method for the encoder and decoder of dense geometric matching networks, significantly enhancing cross-view feature alignment capabilities.
PMR: Prototypical Modal Rebalance for Multimodal Learning
Yunfeng Fan (Hong Kong Polytechnic University), Song Guo (Hong Kong Polytechnic University)
ClassificationRecognitionVideoMultimodalityAudio
🎯 What it does: A prototype-based modality rebalancing (PMR) strategy is proposed, which accelerates slow-learning modalities using a non-parametric prototype classifier and alleviates the suppression of dominant modalities through prototype entropy regularization, thereby addressing the modality imbalance issue in multimodal learning.
POEM: Reconstructing Hand in a Point Embedded Multi-View Stereo
Lixin Yang (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
Pose EstimationTransformerPoint CloudMesh
🎯 What it does: A multi-view hand mesh reconstruction method based on point clouds, called POEM, is proposed, which directly performs feature fusion and cross-attention interaction on 3D points.
Point Cloud Forecasting as a Proxy for 4D Occupancy Forecasting
Tarasha Khurana, Deva Ramanan
Autonomous DrivingPoint Cloud
🎯 What it does: Transforming the point cloud prediction task into predicting four-dimensional occupancy (4D occupancy)
Point2Pix: Photo-Realistic Point Cloud Rendering via Neural Radiance Fields
Tao Hu (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
GenerationData SynthesisConvolutional Neural NetworkNeural Radiance FieldPoint Cloud
🎯 What it does: A point cloud rendering framework named Point2Pix is proposed, which can directly render sparse 3D point clouds into high-quality, photorealistic 2D images.
PointAvatar: Deformable Point-Based Head Avatars From Videos
Yufeng Zheng (ETH Zurich), Otmar Hilliges (ETH Zurich)
GenerationData SynthesisVideoPoint Cloud
🎯 What it does: This paper proposes a 3D head animation avatar model based on variable point clouds, capable of learning animatable and relightable head models from single-camera videos.
PointCert: Point Cloud Classification With Deterministic Certified Robustness Guarantees
Jinghuai Zhang (Duke University), Neil Zhenqiang Gong (Duke University)
ClassificationComputational EfficiencyKnowledge DistillationAdversarial AttackPoint Cloud
🎯 What it does: A point cloud classification method called PointCert is proposed, which is based on hashing to partition point clouds and perform majority voting. It provides a deterministic robustness guarantee for any base point cloud classifier, resisting point addition, deletion, modification, and perturbation attacks.
PointClustering: Unsupervised Point Cloud Pre-Training Using Transformation Invariance in Clustering
Fuchen Long (HiDream.ai Inc.), Tao Mei (HiDream.ai Inc.)
ClassificationSegmentationTransformerContrastive LearningPoint Cloud
🎯 What it does: A method for unsupervised point cloud pre-training based on transformation-invariant deep clustering, called PointClustering, is proposed.
PointCMP: Contrastive Mask Prediction for Self-Supervised Learning on Point Cloud Videos
Zhiqiang Shen (Shanghai Jiao Tong University), Xi Zhou (Shanghai Jiao Tong University)
RecognitionRepresentation LearningTransformerContrastive LearningVideoPoint Cloud
🎯 What it does: The PointCMP framework is proposed to achieve self-supervised learning for point cloud videos;
PointConvFormer: Revenge of the Point-Based Convolution
Wenxuan Wu (Oregon State University), Qi Shan (Apple)
SegmentationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A point cloud convolution module named PointConvFormer is proposed, which combines convolution and attention mechanisms to filter neighboring points based on feature differences while maintaining position invariance.
PointDistiller: Structured Knowledge Distillation Towards Efficient and Compact 3D Detection
Linfeng Zhang (Tsinghua University), Kaisheng Ma (DIDI)
Object DetectionAutonomous DrivingComputational EfficiencyKnowledge DistillationPoint Cloud
🎯 What it does: A PointDistiller framework is proposed for knowledge distillation in point cloud 3D detection;
Pointersect: Neural Rendering With Cloud-Ray Intersection
Jen-Hao Rick Chang (Apple), Oncel Tuzel (Apple)
TransformerPoint CloudMesh
🎯 What it does: This paper proposes Pointersect, a method for differentiable rendering that directly performs ray-point cloud intersection using neural networks.
PointListNet: Deep Learning on 3D Point Lists
Hehe Fan (Zhejiang University), Mohan Kankanhalli (National University of Singapore)
Protein Structure PredictionTransformerPoint CloudBiomedical Data
🎯 What it does: A Transformer-style PointListNet network is proposed for molecules with a mixed structure of 1D sequences and 3D coordinates (such as proteins), modeling directly on the original coordinates and sequences.
PointVector: A Vector Representation in Point Cloud Analysis
Xin Deng (University of Science and Technology of China), XinMing Zhang (University of Science and Technology of China)
ClassificationSegmentationConvolutional Neural NetworkPoint Cloud
🎯 What it does: A point cloud local feature aggregation module VPSA based on vector representation is proposed, and it is integrated into the improved PointNeXt network to achieve point cloud classification and segmentation tasks.
Polarimetric iToF: Measuring High-Fidelity Depth Through Scattering Media
Daniel S. Jeon (Korea Advanced Institute of Science and Technology), Min H. Kim (Korea Advanced Institute of Science and Technology)
Depth EstimationImage
🎯 What it does: A polarization iToF imaging method is proposed, which can accurately estimate depth in scattering media.
Polarized Color Image Denoising
Zhuoxiao Li, Yinqiang Zheng
RestorationConvolutional Neural NetworkImage
🎯 What it does: The paper explores a specific problem in the field of computer vision and proposes a new solution.
Policy Adaptation From Foundation Model Feedback
Yuying Ge (University of Hong Kong), Xiaolong Wang (Amazon Web Services)
Domain AdaptationRobotic IntelligenceReinforcement Learning from Human FeedbackTransformerReinforcement LearningVision Language ModelContrastive LearningMultimodality
🎯 What it does: The PAFF (Policy Adaptation from Foundation Model Feedback) method is proposed, which allows existing language-conditioned control policies to 'play' in new tasks/environments to generate demonstrations, and then uses a pre-trained vision-language model to automatically relabel instructions, thereby fine-tuning the policy for unsupervised adaptation.
Poly-PC: A Polyhedral Network for Multiple Point Cloud Tasks at Once
Tao Xie (Harbin Institute of Technology), Jian Cheng (University of Electronic Science and Technology of China)
ClassificationObject DetectionSegmentationNeural Architecture SearchPoint Cloud
🎯 What it does: This paper proposes Poly-PC, a unified multi-task point cloud learning framework that can simultaneously perform various tasks such as classification, segmentation, and detection within the same network, and supports training across different dataset domains.
PolyFormer: Referring Image Segmentation As Sequential Polygon Generation
Jiang Liu (Johns Hopkins University), R. Manmatha (Johns Hopkins University)
SegmentationGenerationTransformerImageVideoTextMultimodality
🎯 What it does: This paper proposes PolyFormer, which reformulates the image segmentation problem as a multimodal sequence-to-sequence model that directly generates polygon vertex sequences and converts them into segmentation masks.
Polynomial Implicit Neural Representations for Large Diverse Datasets
Rajhans Singh (Arizona State University), Pavan Turaga (Arizona State University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A generative model based on Polynomial Implicit Neural Representation (Poly-INR) is proposed, capable of generating high-resolution images on large and diverse datasets.
Pose Synchronization Under Multiple Pair-Wise Relative Poses
Yifan Sun (University of Texas at Austin), Qixing Huang (University of Texas at Austin)
Pose EstimationOptimizationImage
🎯 What it does: This paper proposes a three-step algorithm for achieving pose synchronization when multiple relative poses exist for each object pair, most of which are incorrect. First, candidate absolute poses for each object are obtained through diffusion and clustering; then, the Markov random field is solved using the projected power method to jointly select the best pose; finally, the poses are refined using the robust Geman-McClure cost with Gauss-Newton optimization.
Pose-Disentangled Contrastive Learning for Self-Supervised Facial Representation
Yuanyuan Liu (China University of Geosciences), Zhe Chen (University of Sydney)
ClassificationRecognitionPose EstimationRepresentation LearningConvolutional Neural NetworkContrastive LearningImageVideo
🎯 What it does: A pose-disentangled contrastive learning framework (PCL) is proposed, which achieves the separation of pose-related and pose-agnostic facial features through a Pose-Disentangled Decoder and is trained in a self-supervised facial representation.
PoseExaminer: Automated Testing of Out-of-Distribution Robustness in Human Pose and Shape Estimation
Qihao Liu (Johns Hopkins University), Alan L. Yuille (Max Planck Institute for Informatics)
Pose EstimationReinforcement LearningImage
🎯 What it does: An automated testing framework named PoseExaminer has been developed for the systematic evaluation of the robustness of human pose and shape estimation methods in out-of-distribution (OOD) scenarios.
PoseFormerV2: Exploring Frequency Domain for Efficient and Robust 3D Human Pose Estimation
Qitao Zhao (Shandong University), Chen Chen (University of Central Florida)
Pose EstimationTransformerVideoBenchmark
🎯 What it does: By sampling only a few central frames in the time domain for spatial encoding and incorporating the low-frequency DCT coefficients of the entire skeleton sequence into the frequency domain features, this approach integrates time domain and frequency domain information to achieve efficient and robust 3D human pose estimation.
Position-Guided Text Prompt for Vision-Language Pre-Training
Jinpeng Wang (National University of Singapore), Shuicheng Yan (Sea AI Lab)
Object DetectionRetrievalTransformerPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a Position-guided Text Prompt (PTP) to enhance the visual localization and alignment capabilities of Visual-Language Pre-training (VLP) models. The method divides images into N×N blocks, generates object labels for each block using an object detector or the CLIP model, and transforms the visual localization task into a fill-in-the-blank problem by inserting position placeholders in the text (e.g., "The block [P] has a [O]"). This allows the model to learn finer-grained spatial information during the pre-training phase.
Positive-Augmented Contrastive Learning for Image and Video Captioning Evaluation
Sara Sarto (University of Modena and Reggio Emilia), Rita Cucchiara (IIT-CNR)
RetrievalTransformerDiffusion modelContrastive LearningImageVideoText
🎯 What it does: The paper proposes a new image and video caption evaluation metric PAC-S, which is trained using contrastive learning based on forward-enhanced cross-modal embedding space.
Post-Processing Temporal Action Detection
Sauradip Nag (University of Surrey), Tao Xiang (University of Surrey)
RecognitionObject DetectionGaussian SplattingVideo
🎯 What it does: A model-free, training-free post-processing method called Gaussian Approximated Post-processing (GAP) is proposed to correct time quantization errors caused by preprocessing in temporal action detection.
Post-Training Quantization on Diffusion Models
Yuzhang Shang (Illinois Institute of Technology), Yan Yan (Houmo AI)
GenerationCompressionDiffusion modelImage
🎯 What it does: A post-training quantization method PTQ4DM is proposed, allowing diffusion models to be directly quantized to 8 bits while maintaining performance.
PosterLayout: A New Benchmark and Approach for Content-Aware Visual-Textual Presentation Layout
Hsiao Yuan Hsu (Peking University), Qing Zhang (Meituan)
GenerationConvolutional Neural NetworkRecurrent Neural NetworkGenerative Adversarial NetworkImageBenchmark
🎯 What it does: A content-aware visual text layout generation method is proposed, which can automatically arrange elements such as text, logos, and backgrounds on an existing content canvas.
POTTER: Pooling Attention Transformer for Efficient Human Mesh Recovery
Ce Zheng (University of Central Florida), Chen Chen (University of Central Florida)
Pose EstimationComputational EfficiencyTransformerImageMesh
🎯 What it does: A lightweight pure Transformer architecture called POTTER is proposed for human mesh recovery from a single image.
Power Bundle Adjustment for Large-Scale 3D Reconstruction
Simon Weber (Technical University of Munich), Daniel Cremers (University of Oxford)
OptimizationSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A Bundle Adjustment solver PoBA based on the inverse Schur complement power series expansion is proposed, achieving fast and low-memory solutions for large-scale 3D reconstruction problems.
Practical Network Acceleration With Tiny Sets
Guo-Hua Wang (Nanjing University), Jianxin Wu (Nanjing University)
ClassificationCompressionKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a framework called PRACTISE to accelerate CNNs with very few training samples. The core idea is to replace traditional filter-level pruning with block-level pruning and use a 'recoverability' metric to select blocks that are easy to recover, followed by fine-tuning through knowledge distillation with a small number of samples.
Prefix Conditioning Unifies Language and Label Supervision
Kuniaki Saito (Boston University), Tomas Pfister (Google)
ClassificationRecognitionTransformerContrastive LearningImageTextMultimodality
🎯 What it does: This paper unifies image-text contrastive pre-training with image classification supervision, proposing a prefix conditioning method to eliminate biases from the two data sources.
PREIM3D: 3D Consistent Precise Image Attribute Editing From a Single Image
Jianhui Li (Tsinghua University), Jun Zhu (Tsinghua University)
Image TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an efficient, accurate, and 3D-consistent facial attribute editing framework called PREIM3D, which can simultaneously perform image inversion and multi-view attribute editing.
Preserving Linear Separability in Continual Learning by Backward Feature Projection
Qiao Gu (University of Toronto), Florian Shkurti (University of Toronto)
Knowledge DistillationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: A Backward Feature Projection (BFP) method is proposed, which utilizes a learnable linear projection to maintain the linear separability of old classes and promote the plasticity of new class features, thereby reducing catastrophic forgetting in continual learning.
Primitive Generation and Semantic-Related Alignment for Universal Zero-Shot Segmentation
Shuting He (Zhejiang University), Wei Jiang (Nanyang Technological University)
SegmentationGenerationTransformerImage
🎯 What it does: A unified framework of 'primitive generation and collaborative relationship alignment and feature separation' is proposed for zero-shot panoptic, instance, and semantic segmentation.
Principles of Forgetting in Domain-Incremental Semantic Segmentation in Adverse Weather Conditions
Tobias Kalb (Porsche Engineering Group GmbH), Jürgen Beyerer (Fraunhofer IOSB)
SegmentationDomain AdaptationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: This paper studies the catastrophic forgetting mechanism of domain incremental semantic segmentation under adverse weather conditions and proposes enhancing the reusability of low-level features through pre-training, self-supervised augmentation, and continual normalization.
PRISE: Demystifying Deep Lucas-Kanade With Strongly Star-Convex Constraints for Multimodel Image Alignment
Yiqing Zhang (Worcester Polytechnic Institute), Ziming Zhang (Worcester Polytechnic Institute)
OptimizationContrastive LearningImageMultimodality
🎯 What it does: A framework for image registration called PRISE is proposed, which optimizes the deep LK method through strong star convex constraints.
Privacy-Preserving Adversarial Facial Features
Zhibo Wang (Zhejiang University), Kui Ren (Hunan University)
RecognitionSafty and PrivacyAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A privacy-enhancing plugin named AdvFace is proposed, which adds adversarial noise to stored facial features to counteract reconstruction attacks from features to images without changing the existing facial recognition network, while maintaining a low loss in recognition accuracy.
Privacy-Preserving Representations Are Not Enough: Recovering Scene Content From Camera Poses
Kunal Chelani (Chalmers University of Technology), Zuzana Kukelova (Czech Technical University in Prague)
Object DetectionPose EstimationSafty and PrivacySimultaneous Localization and MappingImage
🎯 What it does: By sending a large number of single object images to a cloud-based visual localization service, the returned camera pose information is used to infer the type and location of objects in the scene, thereby demonstrating that even with privacy-preserving representations, pose alone can leak scene content.
Private Image Generation With Dual-Purpose Auxiliary Classifier
Chen Chen (University of Sydney), Chang Xu (University of Sydney)
GenerationData SynthesisSafty and PrivacyGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a GAN model under the differential privacy framework, using a dual-purpose auxiliary classifier to enhance the practicality and generalizability of generated images.
PROB: Probabilistic Objectness for Open World Object Detection
Orr Zohar (Stanford University), Serena Yeung (Stanford University)
Object DetectionTransformerImage
🎯 What it does: A probabilistic model-based objectness estimation framework (PROB) is proposed and integrated into Deformable DETR for open-world object detection.
Probabilistic Debiasing of Scene Graphs
Bashirul Azam Biswas (Rensselaer Polytechnic Institute), Qiang Ji (Rensselaer Polytechnic Institute)
OptimizationGraph Neural NetworkTransformerSupervised Fine-TuningGraph
🎯 What it does: In the scene graph generation task, the authors propose a post-processing method that constructs a within-triplet Bayesian network and utilizes virtual evidence for posterior inference on model outputs to correct the long-tail bias in relation predictions.
Probabilistic Knowledge Distillation of Face Ensembles
Jianqing Xu (Tencent), Bryan Hooi (National University of Singapore)
RecognitionKnowledge DistillationImage
🎯 What it does: This paper proposes the Bayesian Ensemble Averaging (BEA) method, which extends traditional mean ensemble through probabilistic modeling, and based on this, designs the BEA-KD knowledge distillation framework that compresses the probabilistic embeddings from multiple teachers into a single student network.
Probabilistic Prompt Learning for Dense Prediction
Hyeongjun Kwon (Yonsei University), Kwanghoon Sohn (Yonsei University)
Object DetectionSegmentationTransformerPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: A probabilistic prompt learning method is proposed to improve dense prediction tasks using visual language models and category-agnostic attribute prompts.
Probability-Based Global Cross-Modal Upsampling for Pansharpening
Zeyu Zhu (Xi'an Jiaotong University), Deyu Meng (Nanyang Technological University)
RestorationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: A probability-based global cross-modal upsampling method (PGCU) for pansharpening is proposed, which fully utilizes the global information of low-resolution multispectral images (LRMS) and the cross-modal information of panchromatic images (PAN), while considering channel specificity.
Probing Neural Representations of Scene Perception in a Hippocampally Dependent Task Using Artificial Neural Networks
Markus Frey (Max Planck Institute for Human Cognitive and Brain Sciences), Caswell Barry (University College London)
SegmentationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Designed and trained an artificial neural network inspired by brain structure, capable of generating scene views from different perspectives and achieving unsupervised object segmentation;
Probing Sentiment-Oriented Pre-Training Inspired by Human Sentiment Perception Mechanism
Tinglei Feng (Nankai University), Jufeng Yang (Nankai University)
ClassificationRecognitionConvolutional Neural NetworkSupervised Fine-TuningContrastive LearningImage
🎯 What it does: This paper proposes an emotion-oriented pre-training method based on the human visual emotional perception mechanism, dividing the pre-training process into three stages (stimulus acquisition, overall organization, and advanced perception), and training independent models at each stage; subsequently, the emotional knowledge from multiple models is integrated into a single target model through feature and logits regularization to enhance visual emotion analysis (VSA) performance.
Procedure-Aware Pretraining for Instructional Video Understanding
Honglu Zhou (Salesforce Research), Juan Carlos Niebles (Salesforce Research)
RecognitionRepresentation LearningGraph Neural NetworkContrastive LearningVideoText
🎯 What it does: This paper proposes the use of Process Knowledge Graphs (PKG) for unsupervised pre-training of instructional videos, learning representations that can encode steps, tasks, and their sequences.
ProD: Prompting-To-Disentangle Domain Knowledge for Cross-Domain Few-Shot Image Classification
Tianyi Ma (University of Technology Sydney), Yi Yang (Zhejiang University)
ClassificationDomain AdaptationTransformerPrompt EngineeringImage
🎯 What it does: A Prompting-to-Disentangle (ProD) method is proposed, which uses dual prompts to separate domain-general and domain-specific knowledge in Transformers, enhancing cross-domain few-shot image classification.
Progressive Backdoor Erasing via Connecting Backdoor and Adversarial Attacks
Bingxu Mu (Xi'an Jiaotong University), Gang Hua (Wormpex AI Research)
Adversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes an advanced backdoor elimination (PBE) method that utilizes the correlation between adversarial attacks and backdoor attacks, capable of gradually removing implanted backdoors through iterative adversarial fine-tuning and sample filtering without clean external data.
Progressive Disentangled Representation Learning for Fine-Grained Controllable Talking Head Synthesis
Duomin Wang (Xiaobing AI), Baoyuan Wang (Xiaobing AI)
GenerationRepresentation LearningConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImageVideo
🎯 What it does: A one-shot speaker head synthesis method is proposed, which can separately control lip movements, head poses, eye gaze/blinking, and emotional expressions.
Progressive Neighbor Consistency Mining for Correspondence Pruning
Xin Liu (Nankai University), Jufeng Yang (Nankai University)
Pose EstimationGraph Neural NetworkImage
🎯 What it does: A Neighborhood Consistency Mining Network (NCMNet) is proposed for correspondence pruning, which can efficiently identify inliers in feature matching and estimate camera pose.
Progressive Open Space Expansion for Open-Set Model Attribution
Tianyun Yang (Institute of Computing Technology, Chinese Academy of Sciences), Sheng Tang (Research Institute of Intelligent Computing, Zhejiang Lab)
ClassificationRecognitionGenerationConvolutional Neural NetworkGenerative Adversarial NetworkImageBenchmark
🎯 What it does: A framework for Open Set Model Attribution (OSMA) is proposed, utilizing a progressively expanded lightweight gain model to simulate the fingerprints of unknown models within a known model space, achieving simultaneous identification of both known and unknown generative models.
Progressive Random Convolutions for Single Domain Generalization
Seokeon Choi (Qualcomm), Sungrack Yun (Qualcomm)
SegmentationDomain AdaptationConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: Proposed an advanced random convolution (Pro-RandConv) enhancement method for single-source domain generalization.
Progressive Semantic-Visual Mutual Adaption for Generalized Zero-Shot Learning
Man Liu (Beijing Jiaotong University), Yao Zhao (Beijing Jiaotong University)
ClassificationRecognitionDomain AdaptationTransformerImage
🎯 What it does: In the task of generalized zero-shot learning, this paper proposes a Progressive Semantic-Visual Mutual Adaption (PSVMA) network, which utilizes a Dual Semantic-Visual Transformer Module (DSVTM) to progressively learn instance-centered attribute prototypes and generate unambiguous visual representations to enhance recognition of unseen categories.
Progressive Spatio-Temporal Alignment for Efficient Event-Based Motion Estimation
Xueyan Huang (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
Pose EstimationOptimizationComputational EfficiencyOptical FlowImageVideo
🎯 What it does: A progressive spatiotemporal alignment framework based on event cameras is proposed, which can efficiently estimate motion parameters for rotation, homography, and 6-DOF motion models.
Progressive Transformation Learning for Leveraging Virtual Images in Training
Yi-Ting Shen (University of Maryland), Shuvra S. Bhattacharyya (University of Maryland)
Object DetectionDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: Proposes the Progressive Transformation Learning (PTL) method, which gradually transforms virtual images into realistic styles and incorporates them into the training set to enhance UAV human detection performance.
Progressively Optimized Local Radiance Fields for Robust View Synthesis
Andréas Meuleman, Johannes Kopf (Meta)
Data SynthesisPose EstimationOptimizationNeural Radiance FieldOptical FlowVideo
🎯 What it does: Utilizing long-duration handheld videos to simultaneously estimate camera trajectory and local radiance fields, achieving high-quality novel view synthesis.
Promoting Semantic Connectivity: Dual Nearest Neighbors Contrastive Learning for Unsupervised Domain Generalization
Yuchen Liu (Shanghai Jiao Tong University), Hongkai Xiong (Shanghai Jiao Tong University)
Domain AdaptationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a dual-nearest neighbor-based contrastive learning framework (DN A), which suppresses intra-domain connectivity through strong data augmentation and constructs cross-domain positive samples of the same class using cross-domain dual-lock nearest neighbors and intra-domain circular nearest neighbors, thereby achieving unsupervised domain generalization.
Prompt-Guided Zero-Shot Anomaly Action Recognition Using Pretrained Deep Skeleton Features
Fumiaki Sato (Konica Minolta), Taiki Sekii (Konica Minolta)
RecognitionAnomaly DetectionGraph Neural NetworkPrompt EngineeringContrastive LearningVideoPoint Cloud
🎯 What it does: This paper proposes an unsupervised zero-shot abnormal action recognition framework based on a pre-trained skeleton feature extractor and user text prompts, capable of recognizing video-level abnormal behaviors without training a target domain model.
Prompt, Generate, Then Cache: Cascade of Foundation Models Makes Strong Few-Shot Learners
Renrui Zhang (Shenzhen Institute of Advanced Technology Chinese Academy of Science), Hongsheng Li (Chinese University of Hong Kong)
ClassificationGenerationTransformerLarge Language ModelPrompt EngineeringContrastive LearningImage
🎯 What it does: This paper proposes CaFo Cascaded Foundation Models, which integrate the pre-trained knowledge of CLIP, DINO, DALL-E, and GPT-3 through a three-step process of Prompt-Generate-Cache to achieve few-shot image classification.
PromptCAL: Contrastive Affinity Learning via Auxiliary Prompts for Generalized Novel Category Discovery
Sheng Zhang (Hong Kong University of Science and Technology), Fahad Shahbaz Khan (Linköping University)
ClassificationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: A two-stage framework called PromptCAL is proposed, which combines visual prompts with contrastive affinity learning to address the problem of Generalized Novel Category Discovery (GNCD).
Prompting Large Language Models With Answer Heuristics for Knowledge-Based Visual Question Answering
Zhenwei Shao (Hangzhou Dianzi University), Jun Yu (Hangzhou Dianzi University)
TransformerLarge Language ModelPrompt EngineeringImageTextMultimodality
🎯 What it does: By generating answer candidates and answer-related examples as two types of heuristics and embedding them into the prompt, we utilize GPT-3 to perform reasoning on knowledge-based VQA tasks.
Propagate and Calibrate: Real-Time Passive Non-Line-of-Sight Tracking
Yihao Wang (Shanghai AI Laboratory), Xuelong Li (Northwestern Polytechnical University)
Object TrackingConvolutional Neural NetworkRecurrent Neural NetworkVideo
🎯 What it does: This study investigates a purely passive non-line-of-sight tracking method that utilizes only RGB cameras to achieve real-time tracking of hidden individuals through walls.
ProphNet: Efficient Agent-Centric Motion Forecasting With Anchor-Informed Proposals
Xishun Wang (QCraft), Xiaodong Yang (QCraft)
Autonomous DrivingTransformerAgentic AIMultimodality
🎯 What it does: An efficient agent-centric motion prediction framework ProphNet is proposed, utilizing anchor-informed proposals to achieve multimodal predictions while maintaining low inference latency.
Proposal-Based Multiple Instance Learning for Weakly-Supervised Temporal Action Localization
Huan Ren (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
RecognitionObject DetectionContrastive LearningOptical FlowVideo
🎯 What it does: This paper proposes a Proposal-based Multiple Instance Learning (P-MIL) framework for weakly supervised temporal action localization, which directly classifies candidate proposals to eliminate the inconsistency between training and testing objectives in traditional S-MIL.
ProTeGe: Untrimmed Pretraining for Video Temporal Grounding by Video Temporal Grounding
Lan Wang (Microsoft), Mei Chen (Microsoft)
RetrievalTransformerContrastive LearningVideoText
🎯 What it does: ProT'eG'e is proposed, utilizing untrimmed videos and aggregated subtitles from HowTo100M for video temporal grounding (VTG) pre-training, and based on this, the VT-SGM module and multi-task loss are constructed;
ProtoCon: Pseudo-Label Refinement via Online Clustering and Prototypical Consistency for Efficient Semi-Supervised Learning
Islam Nassar (Monash University), Gholamreza Haffari (Monash University)
ClassificationContrastive LearningImage
🎯 What it does: This paper proposes a new semi-supervised learning method called PROTOCON, which enhances model performance through pseudo-label refinement in low-label scenarios.
Prototype-Based Embedding Network for Scene Graph Generation
Chaofan Zheng (University of Electronic Science and Technology of China), Jingkuan Song (University of Electronic Science and Technology of China)
RecognitionObject DetectionGenerationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes the Prototype-based Embedding Network (PE-Net), which maps entities and predicates into a semantic space based on word vectors and performs gated fusion of instance features to generate compact and distinguishable representations, achieving entity-predicate matching to enhance the relationship recognition performance in Scene Graph Generation (SGG).
Prototypical Residual Networks for Anomaly Detection and Localization
Hui Zhang (Fudan University), Yu-Gang Jiang (Fudan University)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: A Prototypical Residual Network (PRN) framework is proposed for anomaly detection and localization, which learns residual representations through multi-scale prototypes and multi-size self-attention mechanisms, and alleviates data imbalance using anomaly generation strategies.
Proximal Splitting Adversarial Attack for Semantic Segmentation
Jérôme Rony (École de technologie supérieure), Ismail Ben Ayed (École de technologie supérieure)
SegmentationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A white-box adversarial attack method for semantic segmentation models, ALMA prox, is proposed, which can generate extremely small ∞-norm perturbations.
ProxyFormer: Proxy Alignment Assisted Point Cloud Completion With Missing Part Sensitive Transformer
Shanshan Li (Nanjing University of Aeronautics and Astronautics), Mingqiang Wei (Nanjing University of Aeronautics and Astronautics)
RestorationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: Proposes ProxyFormer, a point cloud completion framework based on proxies and a missing part-sensitive Transformer, which directly generates missing point proxies on partial point clouds and completes fine-grained completion;
Pruning Parameterization With Bi-Level Optimization for Efficient Semantic Segmentation on the Edge
Changdi Yang (Northeastern University), Yanzhi Wang (Northeastern University)
SegmentationOptimizationComputational EfficiencyTransformerImage
🎯 What it does: This paper proposes a soft mask-based pruning parameterization method and combines it with a dual-layer optimization to search for the optimal width in the Token Pyramid module of TopFormer, achieving efficient semantic segmentation on mobile devices.
Pseudo-Label Guided Contrastive Learning for Semi-Supervised Medical Image Segmentation
Hritam Basak (Stony Brook University), Zhaozheng Yin (Stony Brook University)
SegmentationConvolutional Neural NetworkContrastive LearningImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A pseudo-label guided contrastive learning framework is proposed for semi-supervised medical image segmentation.
PSVT: End-to-End Multi-Person 3D Pose and Shape Estimation With Progressive Video Transformers
Zhongwei Qiu (University of Science and Technology Beijing), Jingdong Wang (Baidu)
Pose EstimationTransformerVideo
🎯 What it does: An end-to-end multi-person 3D pose and shape estimation framework PSVT is proposed, utilizing video Transformers to achieve complete video spatiotemporal modeling.