دانلود کتاب Pattern Recognition and Computer Vision: 4th Chinese Conference, PRCV 2021, Beijing, China, October 29 – November 1, 2021, Proceedings, Part I (Image ... Vision, Pattern Recognition, and Graphics)

فهرست مطالب :

Preface
Organization
Contents – Part I
Object Detection, Tracking and Recognition
High-Performance Discriminative Tracking with Target-Aware Feature Embeddings
1 Introduction
2 Discriminative Tracking with Target-Aware Feature Embeddings
2.1 Target-Unaware Feature Extraction
2.2 Target-Aware Feature Construction
2.3 Ridge Regression with Target-Aware Feature Embeddings
2.4 Offline Training
2.5 Online Tracking
3 Experiments
3.1 Implementation Details
3.2 Feature Comparisons
3.3 State-of-the-Art Comparisons
4 Conclusion
References
3D Multi-object Detection and Tracking with Sparse Stationary LiDAR
1 Introduction
2 Related Work
2.1 3D Object Detection
2.2 3D Multi-Object Tracking
3 Proposed Method
3.1 Tracklet Regression
3.2 Data Association
3.3 Football Game Dataset
4 Experiments
4.1 Settings
4.2 Experimental Results
5 Conclusion
References
CRNet: Centroid Radiation Network for Temporal Action Localization
1 Introduction
2 Related Work
3 Our Approach
3.1 Notation and Preliminaries
3.2 Feature Extractor Network
3.3 Relation Network
3.4 Centroids Prediction
3.5 Instance Generation
3.6 Overall Objective Before Random Walk
3.7 Prediction and Post-processing
4 Experiments
4.1 Datasets
4.2 Implementation Details
4.3 Evaluation of RelNet, CenNet and OffNet
4.4 Performance with Fewer Data
4.5 Comparisons with State-of-the-Art
5 Conclusion
References
Weakly Supervised Temporal Action Localization with Segment-Level Labels
1 Introduction
2 Related Work
3 Our Approach
3.1 Problem Statement and Notation
3.2 Architecture
3.3 Classification Loss
3.4 Partial Segment Loss
3.5 Sphere Loss
3.6 Propagation Loss
3.7 Classification and Localization
4 Experiments
4.1 Experimental Setup
4.2 Exploratory Experiments
4.3 Comparisons with the State-of-the-art
5 Conclusion
References
Locality-Constrained Collaborative Representation with Multi-resolution Dictionary for Face Recognition
1 Introduction
2 Proposed Method
2.1 Notations
2.2 Model of LCCR-MRD
2.3 Optimization
2.4 Classification
3 Experiments
3.1 Experimental Settings
3.2 Results and Discussions
4 Conclusion
References
Fast and Fusion: Real-Time Pedestrian Detector Boosted by Body-Head Fusion
1 Introdution
2 Related Work
3 Fast and Fusion
3.1 Baseline
3.2 Body-Head Fusion
3.3 Auxiliary Training Task
4 Experiment
4.1 Datasets and Evaluation Metric
4.2 Evaluation on Extended CityPersons Dataset
4.3 Evaluation on CrowdHuman Dataset
4.4 Ablation Study
5 Conclusion
References
STA-GCN: Spatio-Temporal AU Graph Convolution Network for Facial Micro-expression Recognition
1 Introduction
2 Related Work
2.1 Micro-Expression Recognition
2.2 Graph Convolution Network
3 Method
3.1 ROI Division
3.2 3D CNN with Non-Local Block
3.3 AU-attention Graph Convolution
3.4 Loss Function
4 Experiment
4.1 Experimental Setting
4.2 Implementation Details
4.3 Experimental Result
5 Conclusion
References
Attentive Contrast Learning Network for Fine-Grained Classification
1 Introduction
2 Method
2.1 Attention Generator
2.2 Contrastive Learning Module
2.3 Synergic Learning Module
2.4 Learning Attentive Contrast Learning Networks
3 Experiments
3.1 Datasets
3.2 Implementation Details
3.3 Ablation Study
3.4 Comparison with Other Methods
3.5 Visualization Results
4 Conclusion
References
Relation-Based Knowledge Distillation for Anomaly Detection
1 Introduction
2 Related Work
2.1 CAE-Based Methods
2.2 GAN-Based Methods
2.3 KD-Based Methods
3 Method
3.1 Gram Matrix and the “FSP Matrix”
3.2 The Proposed Approach
4 Experiments
4.1 Implementation Details
4.2 Datasets
4.3 Results
5 Conclusion
References
High Power-Efficient and Performance-Density FPGA Accelerator for CNN-Based Object Detection
1 Introduction
2 Related Work
3 Method
3.1 System Framework
3.2 Neural Network Accelerator
4 Experiments
5 Conclusion
References
Relation-Guided Actor Attention for Group Activity Recognition
1 Introduction
2 Related Works
3 Method
3.1 Location-Aware Relation Module
3.2 Relation-Guided Actor Attention Module
3.3 Classification Layer
4 Experiments
4.1 Datasets and Implementation Details
4.2 Ablation Study
4.3 Comparison with the State-of-the-Arts
5 Conclusion
References
MVAD-Net: Learning View-Aware and Domain-Invariant Representation for Baggage Re-identification
1 Introduction
2 Related Works
2.1 Representation Learning and Metric Learning in ReID
2.2 View-Based Methods for ReID
2.3 Domain Adaptation
3 The Proposed Method
3.1 Baggage ReID Baseline
3.2 Multi-view Attention Model
3.3 Domain-Invariant Learning
4 Experiments
4.1 Dataset and Protocols
4.2 Implementation Details
4.3 Effectiveness of Multi-view Attention
4.4 Effectiveness of Domain-Invariant Learning
4.5 Comparison with Other Methods
5 Conclusion
References
Joint Attention Mechanism for Unsupervised Video Object Segmentation
1 Introduction
2 Related Work
3 Method
3.1 Joint Attention Mechanism
3.2 Network Architecture
4 Experiments
4.1 Datasets and Evaluation
4.2 Ablation Study
4.3 Comparison to the State-Of-The-Arts
5 Conclusion
References
Foreground Feature Selection and Alignment for Adaptive Object Detection
1 Introduction
2 Related Work
2.1 Object Detection
2.2 Adaptive Object Detection
3 Method
3.1 Framework Overview
3.2 Foreground Selection Module
3.3 Multi-level Domain Adaptation
3.4 Overall Objective
4 Experiments
4.1 Implementation Details
4.2 Adaptation Results
4.3 Visualization and Discussion
5 Conclusions
References
Exploring Category-Shared and Category-Specific Features for Fine-Grained Image Classification
1 Introduction
2 Proposed Method
2.1 Category-Shared Feature Extraction Module
2.2 Category-Specific Feature Extraction Module
3 Experiment
3.1 Implementation Details
3.2 Experimental Results
3.3 Ablation Studies
3.4 Visualizations
4 Conclusions
References
Deep Mixture of Adversarial Autoencoders Clustering Network
1 Introduction
2 Mixture of Adversarial Autoencoders
2.1 Adversarial Block
2.2 Target Distribution
2.3 Loss Function
2.4 Training Procedure
3 Experiment
3.1 Clustering Results
3.2 Reconstruct and Generate
4 Conclusion
References
SA-InterNet: Scale-Aware Interaction Network for Joint Crowd Counting and Localization
1 Introduction
2 Related Work
3 Proposed Method
3.1 Scale-Aware Feature Extractor
3.2 Density-Localization Interaction Module
3.3 Loss Function
4 Experiments
4.1 Datasets
4.2 Evaluation Metrics
4.3 Implementation Details
4.4 Comparison with State-of-the-Arts
4.5 Ablation Study
5 Conclusion
References
Conditioners for Adaptive Regression Tracking
1 Introduction
2 Related Work
2.1 One-Stage Visual Tracking
2.2 Conditional Instance Learning
3 The Proposed Conditional Regression Tracking
3.1 Conditional Batch Normalization
3.2 Visual Context and Trajectory Formulating
3.3 Visual Context Network
3.4 Trajectory Network
3.5 Implementation, Training and Inference
4 Experiments
4.1 Ablation Study
5 Conclusions
References
Attention Template Update Model for Siamese Tracker
1 Introduction
2 Related Work
3 Proposed Method
3.1 Traditional Update
3.2 Network Architecture
3.3 Adjustment and Update Blocks
3.4 Channel Attention Block
3.5 Training Model
4 Experiment
4.1 Implementation Details
4.2 Checkpoint Selection
4.3 Performance in Several Benchmarks
4.4 Ablation Studies
5 Conclusion
References
Insight on Attention Modules for Skeleton-Based Action Recognition
1 Introduction
2 Related Work
2.1 Skeleton-Based Action Recognition
2.2 Attention Mechanisms
3 Multi-category Attention Modules
3.1 Spatial-Wise Attention Module
3.2 Temporal-Wise Attention Module
3.3 Spatiotemporal Attention Module
4 Experiments
4.1 Datasets
4.2 Ablation Studies
4.3 Comparison with the State-of-the-Art
5 Conclusions
References
AO-AutoTrack: Anti-occlusion Real-Time UAV Tracking Based on Spatio-temporal Context
1 Introduction
2 Related Work
2.1 Discriminative Correlation Filter Tracking Algorithm
2.2 Anti-occlusion Object Tracking
2.3 DCF Onboard UAV
3 Proposed Tracking Approach
3.1 Review AutoTrack
3.2 Temporal Regularization Analysis and Improvement
3.3 Re-detection Mechanism
4 Experiment
4.1 Implementation Details
4.2 Comparison with Hand-Crafted Based Trackers
4.3 Re-detection Evaluation
5 Conclusions
References
Two-Stage Recognition Algorithm for Untrimmed Converter Steelmaking Flame Video
1 Introduction
2 Related Works
3 Method
3.1 Feature Extraction
3.2 Recognition for Untrimmed Flame Videos
4 Experiments
4.1 Datasets
4.2 Implemented Details
4.3 Data Analysis
5 Conclusion
References
Scale-Aware Multi-branch Decoder for Salient Object Detection
1 Introduction
2 Related Work
3 Method
3.1 Encoder
3.2 Multi-branch Decoder
3.3 Supervision
4 Experiments
4.1 Experimental Setup
4.2 Comparison with State-of-the-Arts
4.3 Ablation Study
5 Conclusion
References
Densely End Face Detection Network for Counting Bundled Steel Bars Based on YoloV5
1 Introduction
2 Densely End Face Detection Network
2.1 Cross Stage Partial Backbone with Attention Module
2.2 Lightweight Module of Adaptively Spatial Feature Fusion
2.3 Cluster-Weighted NMS in Post-processing
3 Experiments and Results
3.1 Experiment Configuration
3.2 Backbone Training and Evaluation
3.3 Performance Comparison Experiment for Different Backbones
3.4 Training and Evaluation of the Network
3.5 Model Comparison
4 Discussion
5 Conclusions
References
POT: A Dataset of Panoramic Object Tracking
1 Introduction
2 Related Work
3 POT Dataset
3.1 Database Characteristics
3.2 Attributes
3.3 Annotation Methodology
3.4 Evaluation Methodology
4 The Local FOV Based Framework
4.1 Framework Overview
4.2 Tangent Plane Projection
4.3 Target Description with Irregular Bounding Box
5 Experiment
5.1 Setup
5.2 Overall Performance
5.3 Performance Analysis by Attributes
6 Conclusion
References
DP-YOLOv5: Computer Vision-Based Risk Behavior Detection in Power Grids
1 Introduction
1.1 Computer Vision
1.2 Automatic Risk Behavior Detection
2 Method
2.1 Revisit YOLOv5s
2.2 Selection of Enhancement
3 Experiments
3.1 Dataset
3.2 Ablation Studies
3.3 Comparison with Other State-of-art Detectors
4 Conclusion
References
Distillation-Based Multi-exit Fully Convolutional Network for Visual Tracking
1 Introduction
2 Related Work
2.1 Visual Tracking
2.2 Multi-exit Architecture
2.3 Knowledge Distillation
3 Proposed Tracking Method
3.1 Identify the Number of Exits
3.2 Network Architecture
3.3 Distillation Training for Multi-exit
4 Online Tracking Algorithm
4.1 Main Process of Tracking
4.2 Optimization and Settings
5 Experimental Results and Analysis
5.1 Ablation Study
5.2 Comparison with the State-of-the-Art
5.3 Discussion
6 Conclusions
References
Handwriting Trajectory Reconstruction Using Spatial-Temporal Encoder-Decoder Network
1 Introduction
2 Spatial-Temporal Encoder-Decoder Network
2.1 Key Point Detector
2.2 Spatial Encoder Network
2.3 Temporal Decoder Network
3 Handwriting Reconstruction Constraints
4 Loss Function
5 Experiment
5.1 Dataset Processing
5.2 Implementation Details
5.3 Evaluation Metrics
5.4 Experiment and Result Analysis
6 Conclusion-0.4em
References
Scene Semantic Guidance for Object Detection
1 Introduction
2 Related Work
2.1 General Object Detection
2.2 Scene Context for Object Detection
3 Approach
3.1 Overview of Scene Semantic Guidance
3.2 Scene Sematic Embedding
3.3 Semantic Consistency Guidance
4 Experiments
4.1 Dataset and Evaluation Metrics
4.2 Implementation Details
4.3 Comparisons with Baselines
4.4 Individual Module Effect
5 Conclusions
References
Training Person Re-identification Networks with Transferred Images
1 Introduction
2 Related Works
2.1 Person Re-identification
2.2 Image Generation and Person Re-identification
2.3 Loss Functions
3 A Unified Generation Framework and Pipeline
3.1 The Framework and Pipeline
3.2 Market1501-EX
3.3 Person Re-identification and Labeling
4 Experiment
4.1 Generated Images
4.2 Re-ID Results
5 Conclusion
References
ACFIM: Adaptively Cyclic Feature Information-Interaction Model for Object Detection
1 Introduction
2 Related Works
2.1 Anchor-Based Object Detectors
2.2 Anchor-Free Object Detectors
2.3 Visual Attention Models
3 The Proposed Method
3.1 Problems and Motivations
3.2 Feature Information-Interaction Model
3.3 Adaptively Cyclic Feature Information-interaction Model
3.4 Loss Function
4 Experimental Results
4.1 Datasets
4.2 Experimental Setup
4.3 Pascal Voc2007
4.4 Ablation Study on PASCAL VOC2007
4.5 Ms Coco
5 Conclusion
References
Research of Robust Video Object Tracking Algorithm Based on Jetson Nano Embedded Platform
1 Introduction
2 Integration of KCF Algorithm with Color Features
3 Adaptive Cross-Linking of KCF Algorithm with Electronic Control Platform
4 Experiments
4.1 Comparison with KCF Tracker
4.2 The Selection of Thresholds
4.3 Pedestrian Tracking by Adding Object Detection
5 Conclusions
References
Classification-IoU Joint Label Assignment for End-to-End Object Detection
1 Introduction
2 Related Work
2.1 Prediction-Aware Label Assignment
2.2 Classification-IoU Joint Representation
3 Methodology
3.1 Classification-IoU Joint Label Assignment
3.2 Classification-IoU Joint Representation
4 Experiments
4.1 Implementation Details
4.2 Ablation Study
4.3 Comparison with State-of-the-art
5 Conclusion
References
Joint Learning Appearance and Motion Models for Visual Tracking
1 Introduction
2 Related Work
3 Proposed Method
3.1 Classification Task
3.2 Estimation Task
4 Experiments
4.1 Implementation Details
4.2 Comparison Results
4.3 Ablation Study
5 Conclusion
References
ReFlowNet: Revisiting Coarse-to-fine Learning of Optical Flow
1 Introduction
2 Related Work
3 Proposed Method: ReFlowNet
3.1 Network Structure
3.2 Symmetrical Factorization Convolution Block
3.3 Confidence Map Module
3.4 Normalized Correlation Layer
3.5 Training Loss
4 Experiments
4.1 Implementation Details
4.2 Results and Analyses
4.3 MPI-Sintel Test Results
5 Conclusions
References
Local Mutual Metric Network for Few-Shot Image Classification
1 Introduction
2 Related Works
2.1 Optimization-Based Approaches
2.2 Metric-Based Approaches
3 Methodology
3.1 Problem Formulation
3.2 Overview
3.3 Local Representation Fusion Layer
3.4 Attention Module
3.5 Local Mutual Metric Module (LM3)
4 Experiments
4.1 Datasets
4.2 Network Architecture
4.3 Experimental Settings
4.4 Comparisons with Other Methods
5 Discussion
5.1 Ablation Study
5.2 Cross-Domain FSIC Analysis
5.3 Influence of Different Local Compare Functions
6 Conclusion
References
SimplePose V2: Greedy Offset-Guided Keypoint Grouping for Human Pose Estimation
1 Introduction
2 METHOD
2.1 Preliminary
2.2 Definition of Gaussian Responses
2.3 Definition of Guiding Offsets
2.4 Post-processing
2.5 Network Structure and Losses
2.6 Peak Regularization for Heatmap Regression
3 Experiments
3.1 Implementation Details
3.2 Error Analyses
3.3 Ablation Studies
3.4 Results
4 Conclusions and Future Work
References
Control Variates for Similarity Search
1 Introduction
2 Our Contributions
3 Review of Preliminary Concepts
3.1 Binary Hashes
3.2 Discrete Hashes
4 Our Control Variate Estimator: Binary Hashes
5 Our Estimator: Discrete Hashes
6 Application to Sign Random Projections
7 Experimental Results
8 Discussion on Control Variates
9 Conclusion
References
Pyramid Self-attention for Semantic Segmentation
1 Introduction
2 Related Work
3 Approach
3.1 Self-attention Module Revisited
3.2 Pyramid Self-attention Mechanism
4 Experiments
4.1 Experimental Setup
4.2 Ablation Study
4.3 Main Results
5 Conclusion
References
Re-Identify Deformable Targets for Visual Tracking
1 Introduction
2 Related Work
2.1 Tracking with the Update Module
2.2 Long-Term Tracking
3 Methods
3.1 Overview
3.2 Similarity Estimation Module
4 Experiments
4.1 Long-Term Tracking Datasets
4.2 Normal Tracking Datasets
4.3 Ablation Study
5 Conclusion
References
End-to-End Detection and Recognition of Arithmetic Expressions
1 Introduction
2 Related Works
2.1 Mathematical Expression Detection
2.2 Mathematical Expression Recognition
2.3 Arithmetic Expression Spotting
3 Methodology
3.1 Framework Overview
3.2 Feature Extractor
3.3 Detection Network
3.4 RoI Feature Extraction
3.5 Recognition Network
4 HAED Dataset and Annotation
5 Experiments
5.1 Datasets
5.2 Implementation Details
5.3 Experimental Results
5.4 Ablation Studies
6 Conclusion
References
FD-Net: A Fully Dilated Convolutional Network for Historical Document Image Binarization
1 Introduction
2 Related Work
3 Proposed Network Architecture: FD-Net
3.1 Hybrid Dilation Rate Settings
3.2 Implementation Details
4 Experiments
4.1 Ablation Study
4.2 More Segmentation Experiments
5 Conclusion
References
Appearance-Motion Fusion Network for Video Anomaly Detection
1 Introduction
2 Related Work
3 Method
3.1 Overview
3.2 Encoder and Decoder
3.3 Feature Fusion Module
3.4 Loss Function
3.5 Abnormality Score
4 Experiments
4.1 Datasets
4.2 Implementation Detail
4.3 Results
4.4 Ablation Study
5 Conclusions
References
Can DNN Detectors Compete Against Human Vision in Object Detection Task?
1 Introduction
2 Related Work
2.1 Object Detection
2.2 Deep Neural Networks vs. Human
3 Methods
3.1 Dataset
3.2 Human Experiments
3.3 DNN Detectors
3.4 Evaluation Metrics
4 Results
5 Conclusion
References
Group Re-Identification Based on Single Feature Attention Learning Network (SFALN)
1 Introduction
2 Related Work
2.1 Person Re-Identification
2.2 Group Re-Identification
2.3 Style Transfer
2.4 Attention Model
3 Proposed Method
3.1 Style Transfer
3.2 Single Feature Attention Learning Network
3.3 Attention Network Architecture Overview
3.4 Online Prediction
4 Experiment
4.1 Datasets
4.2 Implementation Details
4.3 Ablation Experiment
4.4 Comparisons with the State-of-the-art Methods
5 Conclusion
References
Contrastive Cycle Consistency Learning for Unsupervised Visual Tracking
1 Introduction
2 Related Work
3 Proposed Method
3.1 Skipping Frame Strategy
3.2 Step-by-Step Cycle Tracking Strategy
3.3 Contrastive Learning Tracking Framework
3.4 Individual Data Augmentation Operators
4 Experimental Results
4.1 Dataset and Implementation Details
4.2 Comparisons with the State-of-the-Art
4.3 Ablation Studies and Analysis
5 Conclusion
References
Group-Aware Disentangle Learning for Head Pose Estimation
1 Introduction
2 Related Work
3 Proposed Method
3.1 Problem Formulation
3.2 Reconstruction Loss
3.3 KL Divergence
3.4 Training Objective
4 Experiments
4.1 Evaluation Datasets and Metric
4.2 Implementation Details
4.3 Comparisons with State of the Arts
4.4 Ablation Study
5 Conclusion
References
Facilitating 3D Object Tracking in Point Clouds with Image Semantics and Geometry
1 Introduction
2 Related Work
2.1 3D Object Tracking
2.2 Camera-LiDAR Fusion
2.3 Tracking by Detection Scheme
3 Method
3.1 Two-Branch Feature Extraction
3.2 Pseudo 3D Offset Generation
3.3 Target Clue Embedment and Hough Voting
3.4 Loss Functions
4 Experiments
4.1 Datasets and Evaluation Metric
4.2 Implementation Details
4.3 Quantitative and Qualitative Comparisons
4.4 Ablation Study
5 Conclusion
References
Multi-criteria Confidence Evaluation for Robust Visual Tracking
1 Introduction
2 Related Work
2.1 ASRCF
2.2 Confidence Evaluation for Tracking Result
3 Method
3.1 Multi-criteria Confidence Evaluation
3.2 Evaluation Functions
3.3 Mechanisms of Management and Updating
4 Experiment
4.1 Dataset and Setting
4.2 Effectiveness
4.3 Comparison
5 Conclusions
References
Author Index