Advances in Computer Graphics. 39th Computer Graphics International Conference, CGI 2022 Virtual Event, September 12–16, 2022 Proceedings

Preface
Organization
Contents
Image Analysis and Processing
Multi-granularity Feature Attention Fusion Network for Image-Text Sentiment Analysis
	1 Introduction
	2 Related Work
		2.1 Text Sentiment Analysis
		2.2 Image Sentiment Analysis
		2.3 Multimodal Sentiment Analysis
	3 Our Proposed Model
		3.1 Model Overview
		3.2 Feature Learning Layer
		3.3 Interactive Information Fusion Layer
		3.4 Interactive Information Fusion Layer
	4 Experimental Results
		4.1 Dataset Description
		4.2 Baselines
		4.3 Experimental Results and Analysis
	5 Conclusions
	References
Toward Efficient Image Denoising: A Lightweight Network with Retargeting Supervision Driven Knowledge Distillation*-12pt
	1 Introduction
	2 Proposed Method
		2.1 Lightweight U-shaped Denoising Network: LUNet
		2.2 Retargeting Supervision Driven Knowledge Distillation
	3 Experimental Results
		3.1 Experimental Settings
		3.2 Quantitative and Qualitative Results
		3.3 Ablation Studies
	4 Conclusions
	References
A Pig Pose Estimation Model for Measuring Pig’s Body Size
	1 Introduction
	2 Related Work
		2.1 Measuring Animal Body Size
		2.2 Key Point Identification
		2.3 Attention Mechanism
	3 Our Approach
		3.1 Our Network Header
		3.2 Down-Sampling Module
		3.3 Selection of Key Points
	4 Experiments
		4.1 Dataset
		4.2 Results on Image and Video
		4.3 Results on Combining Attention Mechanism
		4.4 Results on Using Our Network Header
	5 Conclusion
	References
Topology-Aware Learning for Semi-supervised Cross-domain Retinal Artery/Vein Classification*-12pt
	1 Introduction
	2 Method
		2.1 Overview
		2.2 Spatial Transformation Module
		2.3 Regional Mixing Module
		2.4 Loss Function
	3 Experimental Results
		3.1 Datasets
		3.2 Implementation Details
		3.3 Artery/Vein Classification Performance
	4 Conclusion
	References
Face Super-Resolution with Better Semantics and More Efficient Guidance
	1 Introduction
	2 Method
		2.1 Overview of the Proposed Framework
		2.2 Better Semantic Prior
		2.3 More Efficient Guidance
		2.4 Loss Functions
	3 Experiments
		3.1 Implementation Details
		3.2 Comparisons with the State-of-the-Arts
		3.3 Ablation Study
	4 Conclusion
	References
Graphs and Networks
Layout and Display of Network Graphs on a Sphere*-12pt
	1 Introduction
	2 Related Work
	3 Visualization Pipeline
	4 Graph Visualization Algorithms
		4.1 Layout - Adaptation of Fruchterman-Reingold Algorithm
		4.2 Quantitative Evaluation
		4.3 Visualizing Spheres in Paraview
		4.4 Mathematical Basis for Visualizing Arcs in Paraview
	5 Browser-Based Visualization
	6 Conclusion
	References
Joint Matrix Factorization and Structure Preserving for Domain Adaptation*-12pt
	1 Introduction
	2 Related Work
	3 Proposed Method
		3.1 Notations
		3.2 Problem Formulation
		3.3 Optimization
	4 Experiments
		4.1 Datasets Description
		4.2 Results
		4.3 Parameter Sensitivity and Ablation Study
	5 Conclusion
	References
Graph Adversarial Network with Bottleneck Adapter Tuning for Sign Language Production*-12pt
	1 Introduction
	2 Related Work
		2.1 Sign Language Production
		2.2 Transfer Learning
		2.3 Graph Convolutional Network
	3 Method
		3.1 Adapter-Based Contextual Representation
		3.2 Channel-Wise Target Length Predictor
		3.3 Cross-modal Sign Pose Generator
		3.4 Spatial-Temporal Graph Convolutional Discriminator
		3.5 Loss Function
	4 Experiments
		4.1 Datasets
		4.2 Training Settings
		4.3 Evaluation Metrics
		4.4 Quantitative Evaluation
		4.5 Visual Comparison
		4.6 Ablation Study
	5 Conclusion
	References
Estimation and Feature Matching
Facial Landmarks Based Region-Level Data Augmentation for Gaze Estimation*-12pt
	1 Introduction
	2 Related Works
		2.1 Gaze Estimation
		2.2 Data Augmentation
	3 Proposed Method
		3.1 Facial Landmarks Based Region Mask
		3.2 Non-eye Regions Data Augmentation
	4 Experiments
		4.1 Experimental Setup
		4.2 Performance Comparison
		4.3 Ablation Studies
	5 Conclusions
	References
An Efficient Dense Depth Map Estimation Algorithm Using Direct Stereo Matching for Ultra-Wide-Angle Images*-12pt
	1 Introduction
	2 Patch Matching Stereo
		2.1 Camera Projection Model
		2.2 Patch Matching Stereo for Ultra-Wide-Angle Images
	3 Experimental Results
		3.1 The TUM VI Benchmark
		3.2 Comparisons
		3.3 The Oxford RobotCar Dataset
	4 Conclusion
	References
Ad-RMS: Adaptive Regional Motion Statistics for Feature Matching Filtering
	1 Introduction
	2 Related Works
	3 Method
		3.1 Adaptive Regional Division
		3.2 Regional Motion Statistics
	4 Experiment
		4.1 Evaluation
		4.2 Dataset and Input Data
		4.3 Parameters
		4.4 Comparative Analysis of Results
	5 Summary
	References
3D Reconstruction
Visual Indoor Navigation Using Mobile Augmented Reality
	1 Introduction
	2 Related Work
		2.1 Indoor Navigation System
		2.2 Positioning and Navigation Technology
	3 Research Content and Methodology
		3.1 Plane and Point Cloud Detection and Visualization
		3.2 Indoor Map Construction
		3.3 Map Download and Self-positioning
		3.4 Path Planning and Display
	4 Experiment
	5 Analysis of Results
	References
Cost Volume Pyramid Network with Multi-strategies Range Searching for Multi-view Stereo
	1 Introduction
	2 Related Work
		2.1 Coarse-to-Fine MVS Methods
		2.2 Depth Sampling Range
		2.3 Cost Volume
	3 Methods
		3.1 Overview
		3.2 Depth Sampling Range Estimation
		3.3 Supervise on Cost Volume
		3.4 Loss Function
	4 Experiment
		4.1 Dataset
		4.2 Implementation Details
		4.3 Results on DTU Dataset
		4.4 Results on BlendedMVS
		4.5 Effectiveness of Multi-strategies
		4.6 Ablation Study
	5 Conclusion
	References
Reconstructing the Surface Mesh Representation for Single Neuron
	1 Introduction
	2 Related Work
	3 Method
		3.1 Branch Identification
		3.2 Surface Reconstruction of Neurites
		3.3 Surface Reconstruction of the Soma
	4 Experiments and Results
	5 Conclusion
	References
WEmap: Weakness-Enhancement Mapping for 3D Reconstruction with Sparse Image Sequences
	1 Introduction
	2 Related Work
	3 Method
		3.1 Initial Reconstruction
		3.2 Generation of the Enhanced Set
		3.3 Reconstruction with Weighted Samples
	4 Experiments
		4.1 Datasets
		4.2 Experiments on the Sparse DTU Dataset
		4.3 Experiments on the Sparse Tanks and Temples Dataset
	5 Conclusion
	References
Rendering and Animation
Comparing Traditional Rendering Techniques to Deep Learning Based Super-Resolution in Fire and Smoke Animations
	1 Introduction
	2 Related Work
	3 TecoGAN
		3.1 Network Architecture for Video Super-Resolution
		3.2 Usage
	4 Dataset Generation
		4.1 Creating a Smoke Simulation
		4.2 Multiple Simulations
	5 Evaluation
		5.1 Creating Different Renders
		5.2 Comparing Video Quality
		5.3 Cross Simulation Comparison
	6 Conclusion
	References
Real-Time Light Field Path Tracing
	1 Introduction
	2 Related Work
	3 General Pipeline and Our Implementation
		3.1 General Pipeline
		3.2 Our Implementation
	4 Experiments
	5 Results
		5.1 Runtime
		5.2 Quality
	6 Conclusions
	References
Crowd Simulation with Detailed Body Motion and Interaction
	1 Introduction
	2 Related Work
		2.1 Crowd Simulation
		2.2 Motion Matching
	3 Overview
	4 Hierarchical Control
	5 Agent Model
		5.1 Visual System
		5.2 Blackboard System
		5.3 Decision System
		5.4 Animation System
	6 Co-simulation System
		6.1 Mode Switch
		6.2 Inverse Kinematic
	7 Results
		7.1 Behaviour Simulation
		7.2 Animation Simulation
		7.3 User Study
	8 Conclusion and Future Work
	References
Towards Rendering the Style of 20th Century Cartoon Line Art in 3D Real-Time
	1 Introduction
	2 Related Works
		2.1 Contour Extraction
		2.2 Line Stylisation
		2.3 Offline Rendering
	3 Approaches
		3.1 Edge Marking
		3.2 Geometry Shader Line Rendering
		3.3 Line Weight Map
		3.4 Editing the Line Thickness in Real-Time
		3.5 Recording Line Thickness in Real-Time Animation
	4 Results
	5 Conclusion and Future Works
	References
Detection and Recognition
Face Detection Algorithm in Classroom Scene Based on Deep Learning
	1 Introduction
	2 Related Works
		2.1 Overview of Target Detection
		2.2 Optical Flow Method
	3 Proposed Method
		3.1 Model Overview
		3.2 Network Design
		3.3 Improved Seq NMS
		3.4 Face Count
	4 Experience
		4.1 Dataset
		4.2 Pretreatment
		4.3 Parameter Setting
		4.4 Result
	5 Conclusion
	References
GRVT: Toward Effective Grocery Recognition via Vision Transformer
	1 Introduction
	2 Related Work
	3 Method
		3.1 Vision Transformer
		3.2 GRVT Architecture
		3.3 Multi-scale Patch Embedding
		3.4 Mixed Attention Selection
	4 Experiments
		4.1 Datasets and Implementation Details
		4.2 Comparison with the State of the Art
		4.3 Ablation Study
		4.4 Visualization
	5 Conclusion
	References
A Transformer-Based Cloth-Irrelevant Patches Feature Extracting Method for Long-Term Cloth-Changing Person Re-identification
	1 Introduction
	2 Related Work
		2.1 Cloth-Changing Person Re-identification
		2.2 Vision Transformer
		2.3 Human Parsing
	3 Methodology
		3.1 Transformer Encoder
		3.2 Patch Select Module (PSM)
		3.3 Inverse Cloth-Id Loss
		3.4 Training and Testing
	4 Experiment
		4.1 Datasets and Evaluation Protocals
		4.2 Implementation Details
		4.3 Comparison on the Cloth-Changing Dataset
		4.4 Experiment on Standard Re-ID
		4.5 Ablation Study
		4.6 Limitations
	5 Conclusion
	References
Learning Unified Binary Feature Codes for Cross-Illumination Palmprint Recognition
	1 Introduction
	2 Learning Robust Representation of Cross-Illumination Palmprint Images
		2.1 Multi-illumination Palmprint Recognition
		2.2 Feature Representation and Matching
	3 Experiments
		3.1 PolyU Multi-spectral Palmprint Image Database
		3.2 Cross-Illumination Palmprint Identification
		3.3 Cross-Illumination Palmprint Verification
		3.4 Parameter Analysis
		3.5 Computational Complexity Analysis
	4 Conclusion
	References
Colors, Paintings and Layout
SemiPainter: Learning to Draw Semi-realistic Paintings from the Manga Line Drawings and Flat Shadow
	1 Introduction
	2 Related Work
		2.1 Deep Learning Based Art Paintings Generation
		2.2 Deep Learning Based Shadow Generation
		2.3 Line Art Colorization
		2.4 Photo Colorization
	3 Methods
		3.1 Shadow Generation
		3.2 Line Art Colorization
		3.3 Composing Shadow and Color
	4 Experiments
		4.1 Qualitative Results
		4.2 Numerical Results
		4.3 User Study
		4.4 Ablative Study
	5 Conclusion and Future Work
	References
Hierarchical Bayesian Network Modeling and Layout of Huizhou Traditional Villages in Geographic Environment
	1 Introduction
	2 Related Work
		2.1 Layout of Modern City
		2.2 Layout of Historic City and Village
	3 Description and Definition of Data
	4 Hierarchical Bayesian Network Modeling
		4.1 Hierarchical Bayesian Network Structure Learning
		4.2 Hierarchical Bayesian Network Parameter Learning
	5 Huizhou Ancient Village Generation
		5.1 Road Skeleton Generation
		5.2 House Placement
	6 Experiment
		6.1 Comparison with Other Methods
		6.2 Bayesian Network Accuracy Verification
		6.3 Example Comparison
	7 Conclusion
	References
AE-GAN: Attention Embedded GAN for Irregular and Large-Area Mask Face Image Inpainting
	1 Introduction
	2 Related Work
		2.1 Face Inpainting
		2.2 Generative Adversarial Network
		2.3 Attention Mechanism
	3 Method
		3.1 Motivation and Framework
		3.2 U-shape Generator
		3.3 Multi-scale Discrimnator
		3.4 Embedded PSE Attention Block
	4 Loss Function
	5 Experiment
		5.1 Datesets and Evaluation Metric
		5.2 Implementation Details
		5.3 Qualitative Comparison
		5.4 Quantitative Comparison
		5.5 Ablation Studies
	6 Conclusion
	References
Synthesis and Generation
Procedural Generation of Landscapes with Water Bodies Using Artificial Drainage Basins
	1 Introduction
	2 Related Work
	3 Overview of Our Approach
	4 Our Terrain Generation Pipeline in Detail
		4.1 Ocean Borders
		4.2 Regions
		4.3 River Networks and Lakes
		4.4 Terrain
		4.5 Visualization
	5 Evaluation
		5.1 Performance Evaluation
		5.2 Qualitative Evaluation
	6 Conclusion
	References
High-Fidelity Dynamic Human Synthesis via UV-Guided NeRF with Sparse Views
	1 Introduction
	2 Related Work
		2.1 Dynamic Human Synthesis
		2.2 Neural Rendering
		2.3 Image-Based Rendering
	3 Method
		3.1 Canonical Space Mapping Module
		3.2 Texture Space Mapping Module
		3.3 UV-Guided Neural Rendering Module
		3.4 Loss
	4 Experiments
		4.1 Dataset
		4.2 Training
		4.3 Comparisons
		4.4 Ablation Study
	5 Conclusion
	References
Rec2Real: Semantics-Guided Photo-Realistic Image Synthesis Using Rough Urban Reconstruction Models
	1 Introduction
	2 Related Work
		2.1 Data Collection Methods
		2.2 Urban Scene Reconstruction
		2.3 Image Style Transfer
	3 Method
		3.1 Rec-Style Image and Annotation Generation
		3.2 Real-to-Rec Style Transfer
		3.3 Rec-to-Real Image Refinement
	4 Experiments and Discussion
		4.1 The Impact of Reconstructed Scenes
		4.2 Comparison with Style Transfer Methods
		4.3 The Validity of Our Method
		4.4 The Impact of Semantic Patches
	5 Conclusions and Limitations
	References
3D Digital City Structure Model Based on Image Modeling Technology
	1 Introduction
	2 Design of 3D Digital City Structure Model
		2.1 Development of a Parametric Camera Model
		2.2 Extraction of Dynamic Texture Matching Points of 3D Digital City Structure Model
		2.3 Selecting Initial Matching Point Pairs for the 3D Structure of Digital Cities
		2.4 Building 3D Digital City Structure Model
	3 Comparative Experiment Analysis
		3.1 Simulation Experiment Platform Setup
		3.2 Experiment Result Analysis
	4 Conclusion
	References
AR and User Interfaces
Augmented Reality-Based Home Interaction Layout and Evaluation
	1 Introduction
	2 Related Work
		2.1 Indoor Scene Layout Assessment
		2.2 Avatar-Based Indoor Scene Layout
		2.3 Interactive Indoor Scene Layout
	3 Overview
	4 Accessibility Assessment Based on Avatar Roaming
		4.1 Avatar-Based Indoor Scene Layout
		4.2 Accessibility Assessment
	5 Ventilation and Illumination Assessment
		5.1 Indoor Ventilation Assessment
		5.2 Illumination Assessment
	6 Experiment
		6.1 Evaluation Results Under Different Layouts
		6.2 User Study
	7 Conclusion
	References
LiteAR: A Framework to Estimate Lighting for Mixed Reality Sessions for Enhanced Realism
	1 Introduction
	2 Related Work
	3 Illumination Estimation
		3.1 Data Preparation
		3.2 Global Spherical Harmonics Prediction
		3.3 Spatially Varying Environment
		3.4 Spherical Harmonics Rotation
		3.5 Computational Analysis
	4 Results and Discussion
	5 Conclusion and Future Work
	References
Personalized User Interface Elements Recommendation System
	1 Introduction
	2 Related Works
		2.1 Personalized User Interface
		2.2 Metrics
	3 Recommendation System
		3.1 Preliminaries
		3.2 Processing
		3.3 Model
	4 Experiments
		4.1 Data Collection
		4.2 Feature Intersection
		4.3 Metrics
		4.4 Results
	5 Conclusion and Feature Works
	References
Medical Imaging
A Feature Point Extraction Method for Capsule Endoscope Localization
	1 Introduction
	2 Related Work
	3 Method Overview
	4 Method Detail Description
		4.1 Adaptive Thresholds
		4.2 Feature Point Homogenization
		4.3 BEBLID Descriptor
	5 Experimental Results and Analysis
		5.1 Selection of Datasets
		5.2 Evaluation Criteria
		5.3 Experimental Results
		5.4 Comparative Analysis of Methods
	6 Conclusions
	References
Automated Diagnosis of Retinal Neovascularization Pathologies from Color Retinal Fundus Images
	1 Introduction
	2 Related Work
	3 Automated Diagnosis of WAMD and PDR
		3.1 Proposed Method
		3.2 Data Preprocessing and Augmentation
		3.3 Transfer Learning of VGG16 Architecture
	4 Experimental Results
		4.1 Dataset
		4.2 Evaluation Metrics
		4.3 Performance Evaluation of Proposed Method
	5 Conclusion
	References
Segmentation
DDCNet: A Lightweight Network with Variable Receptive Field for Real-Time Portrait Segmentation in Complex Environment
	1 Introduction
	2 Related Work
	3 Our Methods
		3.1 Deformable Depth Separable Convolution Block (DDC)
		3.2 Detail Selection Block (DS)
		3.3 Loss Function
	4 Experimental Results and Analysis
		4.1 Visual Comparison and Index Analysis
		4.2 Ablation Study
		4.3 User Study
	5 Conclusion
	References
A Chromosome Segmentation Method Based on Corner Detection and Watershed Algorithm
	1 Introduction
	2 Method
		2.1 Image Denoising
		2.2 Histogram Equalization Enhancement
		2.3 Target Classification
		2.4 Chromosome Segmentation
	3 Experimental Results and Analysis
	4 Conclusion
	References
Voxel-Based 3D Shape Segmentation Using Deep Volumetric Convolutional Neural Networks
	1 Introduction
	2 Related Work
	3 V-SegNet Framework
		3.1 Generating the 3D Shape Feature Input
		3.2 The Architecture of V-SegNet
		3.3 Network Training
	4 Experimental Results
		4.1 Experimental Setting
		4.2 Comparison
		4.3 Ablation Studies
	5 Limitation and Future Work
	6 Conclusion
	References
Object Detection
Few-Shot Detection Based on an Enhanced Prototype for Outdoor Small Forbidden Objects
	1 Introduction
	2 Related Work
		2.1 Few-Shot Learning
		2.2 Attentional Mechanisms
	3 Methodology
		3.1 Learning and Extraction of Common Generic Prototypes
		3.2 Regional Many-to-Many Attention Reinforcement Prototype Module
		3.3 Two-Stage Fine-Tuning Method: Regionally Mixed Consistent Loss Function
	4 Experiment
		4.1 Data Set Introduction
		4.2 Ablation Studies in KS Dataset
		4.3 Comparison Experiments in KS Dataset
		4.4 Generalizability Experiments
	5 Conclusion
	References
Research on Real-Time Forestry Pest Detection Based on Improved YOLOv5
	1 Introduction
	2 Related Work
		2.1 Data Augmentation
		2.2 Object Detection Models
		2.3 Overview of YOLOv5
	3 Method
		3.1 BoTR
		3.2 M-BiFPN
		3.3 D-CBAM
	4 Experiments
		4.1 Copy-Pasting Strategy to Augment Dataset
		4.2 Implementation Details
		4.3 Performance Comparison
		4.4 Training Results
		4.5 Ablation Studies
		4.6 Visualization
	5 Conclusions
	References
Power Line Detection Based on Feature Fusion Deep Learning Network
	1 Introduction
	2 Related Work
		2.1 Related Deep Learning Models
		2.2 Existing PLD Methods
	3 The Proposed Method
		3.1 Overall Network Architecture Design
		3.2 Rough Feature Extraction Module Design
		3.3 Residual Module
		3.4 Inception Module
		3.5 Output Fusion Module Design
	4 Experimental Results and Analysis
		4.1 Dataset, Experimental Configuration and Evaluation Metric
		4.2 Experimental Comparison Results and Analysis
		4.3 Robustness and Generalization Tests
	5 Conclusion
	References
Image Attention and Perception
Wider and Higher: Intensive Integration and Global Foreground Perception for Image Matting
	1 Introduction
	2 Related Work
	3 Methodology
		3.1 Network Architecture
		3.2 Intensive Connections
		3.3 Global Foreground Perception
		3.4 Loss Functions and Implementation Details
	4 Experiments
		4.1 Results on Adobe Composition-1k
		4.2 Ablation Study
		4.3 Results on Distinctions-646
		4.4 Results on the Alpha Matting Benchmark
		4.5 Results on Natural Images
	5 Conclusions and Future Works
	References
Authenticity Identification of Qi Baishi\'s Shrimp Painting with Dynamic Token Enhanced Visual Transformer
	1 Introduction
	2 Method
		2.1 Network Framework
		2.2 Dynamic Patch Creation and Embedding Strategy
		2.3 Dynamic Tokens Enhancement Module
		2.4 Loss Function
	3 Experiments
		3.1 Experiments Settings and Evaluation Metrics
		3.2 Validation Datasets and Data Preprocessing
		3.3 Hyper-parameters Selection
		3.4 Ablation Studies
		3.5 Comparison with State-of-the-Art Methods
		3.6 Discussions and Limitations
	4 Conclusion
	References
Modeling and Simulation
An Optimized Material Point Method for Soil-Water Coupled Simulation
	1 Introduction
	2 Related Works
		2.1 Computational Geomechanics
		2.2 Material Point Method
		2.3 CG Simulation of Porous Flow
	3 Physical Model
		3.1 Governing Equations
		3.2 Soil Constitutive Model
		3.3 Water Consititutive Model
	4 Implementation
		4.1 Particle-Grid Mapping Scheme
		4.2 Update Stress
		4.3 Update Nodal Velocity
		4.4 Boundary Conditions
	5 Results
		5.1 Performance
		5.2 Soil Elastoplasticity
		5.3 Soil-Water Coupling
	6 Discussion
	References
SlimFliud-Net: Fast Fluid Simulation Using Admm Pruning
	1 Introduction
	2 Related Work
		2.1 Fluid Simulation
		2.2 Data-Driven Fluid Simulation
		2.3 Fluid Acceleration
	3 Method
		3.1 ADMM Pruning Method
		3.2 Original Model
		3.3 Implementation
	4 Experiments
		4.1 Datasets
		4.2 Quantitative Result
		4.3 Qualitative Result
	5 Discussion and Conclusions
	References
Author Index