Image and Vision. Computing 37th International Conference, IVCNZ 2022 Auckland, New Zealand, November 24–25, 2022 Revised Selected Papers

Preface
Organization
Contents
StencilTorch: An Iterative and User-Guided Framework for Anime Lineart Colorization
	1 Introduction
	2 Background and Related Work
	3 Proposed Method
		3.1 Dataset Curation
		3.2 Input Generation
		3.3 Model Architecture
		3.4 Mask Inpainting
		3.5 Curriculum Learning
	4 Implementation
	5 Evaluation Setup
	6 Results
	7 Conclusion
	References
UnseenNet: Fast Training Detector for Unseen Concepts with No Bounding Boxes
	1 Introduction
	2 Task Definition
	3 Related Work
	4 UnseenNet Framework
		4.1 Training Baseline Detector Offline for Seen Concept (with Bounded Vocabulary)
		4.2 Training Online Detector for Unseen Concept (for Unbounded Vocabulary)
	5 Experiments
		5.1 Implementation Details
		5.2 Quantitative Evaluation on Unseen Categories
		5.3 Qualitative Evaluation on Unseen Categories
	6 Conclusions and Future Work
	References
Person Detection Using an Ultra Low-Resolution Thermal Imager on a Low-Cost MCU
	1 Introduction
	2 Related Work
		2.1 Person Detection
		2.2 Model Compression
	3 Approach
		3.1 Dataset
		3.2 Model
		3.3 Compression
	4 Experiments
		4.1 Comparison
		4.2 Background Subtraction
		4.3 Model Compression
		4.4 Deployment
	5 Conclusion
	References
A Real-Time Kiwifruit Detection Based on Improved YOLOv7
	1 Introduction
	2 Related Work
		2.1 YOLOv7
		2.2 Attention Mechanism
	3 Our Methods
		3.1 Dataset
		3.2 Modelling
	4 Our Results
		4.1 Evaluation Metrics for Kiwifruit Detection
		4.2 Experimental Environment and Training Parameters
		4.3 Experimental Results and Comparisons
	5 Conclusion
	References
A Novel Explainable Deep Learning Model with Class Specific Features
	1 Introduction
	2 Proposed Method
	3 Experiments
	4 Results and Analysis
	5 Conclusion
	References
Correcting Charge Sharing Distortions in Photon Counting Detectors Utilising a Spatial-Temporal CNN
	1 Introduction
	2 Background
		2.1 Charge Sharing
		2.2 Medipix3RX
	3 Proposed Framework
		3.1 Network
	4 Experimental Setup
		4.1 Training and Testing Data
		4.2 Network Training and Testing
	5 Results
		5.1 Metrics and Evaluation
		5.2 Results for Experiment One
		5.3 Results for Experiment Two
	6 Discussion
	7 Conclusion
	8 Future Work
	References
Vehicle-Related Distance Estimation Using Customized YOLOv7
	1 Introduction
	2 Literature Review
	3 Our Methods
	4 Experimental Results
	5 Conclusion
	References
Extending Temporal Data Augmentation for Video Action Recognition
	1 Introduction
	2 Related Work
		2.1 Spatial Augmentation
		2.2 Video Recognition
		2.3 Temporal Augmentation
	3 Methods
		3.1 Single Video Augmentation
		3.2 MagAugment
		3.3 Temporal Deleting, Cut-and-Pasting, and Blending
	4 Experiments
		4.1 Single Video Augmentation
		4.2 Temporal Deleting, Cut-and-Pasting, and Blending
	5 Conclusions
	References
Pre-text Representation Transfer for Deep Learning with Limited and Imbalanced Data: Application to CT-Based COVID-19 Detection
	1 Introduction
	2 Proposed Method
		2.1 Traditional Model Transfer
		2.2 Pre-text Representation Transfer (Training-I)
		2.3 Subsequent Transfer Learning (Training-II)
		2.4 Feature Extraction and Dictionary Classifier (Training-III)
		2.5 Output Predictions (Testing)
	3 Experiments
	4 Conclusion
	References
FNR-GAN: Face Normalization and Recognition with Generative Adversarial Networks
	1 Introduction
	2 Related Works
		2.1 Face Augmentation: One-to-Many Face Generation
		2.2 Face Normalization: Many-to-One Face Generation
		2.3 Identity Preservation
		2.4 Face Recognition
	3 Proposed Method
		3.1 Face Normalization
		3.2 Face Recognition
	4 Experiments and Results
		4.1 Expression Neutralization
		4.2 Head Pose Frontalization
		4.3 Face Verification
		4.4 Face Recognition
	5 Conclusions
	References
Object Tracking with Multiple Dynamic Templates Updating
	1 Introduction
	2 Related Work
		2.1 Siamese Trackers
		2.2 Temporal Exploitation for Tracking
	3 Method
		3.1 Dynamic Template Pool Updating Siamese Tracker
		3.2 Dissimilarity Based Hard Example Sampling
	4 Experiments
		4.1 Implementation Details
		4.2 Comparison with Baselines
		4.3 Ablation Experiments
	5 Conclusion
	References
Detection and Tracking of Pinus Radiata Catkins
	1 Introduction
	2 Previous Research
	3 Methodology
		3.1 Catkin Detection
		3.2 Depth Estimation
		3.3 Catkin Tracking
	4 Results
	5 Limitations
	6 Conclusion
		6.1 Future Research
	References
Assessing the Condition of Copper Conductors Using Deep Learning
	1 Introduction
	2 Related Work
	3 Dataset and Data Preparation
		3.1 Laboratory Dataset
		3.2 Data Preparation
	4 Experimental Setup
	5 Results and Discussion
		5.1 Assessing the Assessors
		5.2 Inspecting the Network
		5.3 Visualization Using t-SNE
	6 Conclusion and Future Work
	References
Evolving U-Nets Using Genetic Programming for Tree Crown Segmentation
	1 Introduction
	2 Background
		2.1 U-Net for Image Segmentation
		2.2 Evolutionary Automated Design of Deep Neural Networks
	3 The Proposed Approaches
		3.1 Layer U-Net
		3.2 GP U-Net
	4 Design of the Experiment
	5 Results and Discussions
	6 Conclusions
	References
Probability Mapping of Spectral CT Material Decomposition to Aid in Determining Material Identification and Quantification Likelihood
	1 Introduction
	2 Materials and Methods
		2.1 Datasets
		2.2 Material Decomposition (MD)
		2.3 Probability Mapping
	3 Results and Discussion
	4 Conclusion
	References
Explainable Network Pruning for Model Acceleration Based on Filter Similarity and Importance
	1 Introduction
	2 Related Work
		2.1 Filter Level Network Pruning
		2.2 Approaches for Filter Pruning
		2.3 Existing State of Explainability in Network Pruning
	3 Method
		3.1 Similarity-Based Filter Pruning
		3.2 Importance-Based Filter Pruning
		3.3 Hybrid Filter Pruning
	4 Experiment Setup
	5 Experiment Results
		5.1 Comparisons of Similarity-Based Filter Pruning and Importance-Based Filter Pruning
		5.2 Comparisons of Our Hybrid Filter Pruning with the State-of-the-Art Filter Pruning Methods
	6 Conclusions
	References
Outlier Detection for Visual Odometry in Vegetated Scenes Using Local Flow Consistency
	1 Introduction
	2 Related Work
	3 Method
		3.1 Neighbourhood Formation
		3.2 Outlier Identification
	4 Experiments
		4.1 Methods
		4.2 Test Procedure
		4.3 Dataset
		4.4 Synthetic Data
		4.5 Real Data
	5 Conclusion
	References
M3T: Multi-class Multi-instance Multi-view Object Tracking for Embodied AI Tasks
	1 Introduction
	2 Related Work
		2.1 Object Detection
		2.2 Object Re-identification
		2.3 Multiple Object Tracking
	3 Multi-class Multi-instance Multi-view Object Tracking
		3.1 Task Definition
		3.2 Policy Definition
	4 M3T Model
		4.1 Stages of M3T Model
		4.2 M3T-Round Model
	5 Experiments
		5.1 Implementation Details
		5.2 Dataset
		5.3 Results
	6 Discussion and Limitations
		6.1 Limitations
	7 Conclusions
	References
A VR Tool for Labelling 3D Data Sets
	1 Introduction
	2 Point Cloud Labelling Tools
		2.1 Point Cloud Labelling with 2D Interfaces
		2.2 Point Cloud Labelling in VR
	3 A VR Point Cloud Labelling Tool
	4 Experimental Evaluation
		4.1 Labelling Tasks
		4.2 Method
		4.3 Participants
	5 Results and Discussion
		5.1 Task Completion Time
		5.2 Task Accuracy
		5.3 Survey Results
		5.4 Other Observations
	6 Conclusions and Future Work
	References
Determining Realism of Procedurally Generated City Road Networks
	1 Introduction
	2 Related Work
	3 Design
		3.1 System Overview
		3.2 Road Network
		3.3 City Blocks
	4 Implementation
	5 Evaluation
		5.1 Methodology
		5.2 Results
	6 Conclusion and Future Work
	References
Video Quality Assessment Considering the Features of the Human Visual System
	1 Introduction
	2 Related Work
	3 Video Quality Assessment Considering the Features of the Human Visual System: Adaptation Aspects, Spatial and Temporal Frequency, Eccentricity, Luminance
		3.1 Video Quality Assessment Metric
	4 Methods
	5 Results and Discussion
	6 Conclusion
	References
Small Visual Object Detection in Smart Waste Classification Using Transformers with Deep Learning
	1 Introduction
	2 Related Work
		2.1 Small Object Detection
		2.2 Visual Object Detection
	3 Our Method
	4 Result Analysis
		4.1 Our Dataset
		4.2 Evaluation Methods
		4.3 Result Analysis
		4.4 Ablation Experiments
	5 Conclusion
	References
A Hybrid Human-Machine System for Image-Based Multi-weather Detection
	1 Introduction
	2 Related Work
		2.1 Weather Classification
	3 Overall Weather Features
		3.1 Illumination
		3.2 Colorfulness
		3.3 Presence of Haze
		3.4 Sky Segmentation
		3.5 Sky Features
		3.6 Contrast Energy
		3.7 Snow Feature
		3.8 CNN Model
	4 Weather Dataset
		4.1 Sky Dataset and SKYNET
		4.2 Multi-class Latent Joint Support Vector Machine (MLJSVM)
		4.3 Experimental Results
	5 Concluding Remarks and Scope for Future Work
	References
A Novel CNN-Based Approach for Distinguishing Between COVID and Common Pneumonia
	1 Introduction
	2 Related Work
	3 Proposed Approach
	4 Comparison of Performance
		4.1 Classification Using Radiomic Features
		4.2 Classification Using Convolutional Features
		4.3 Hybrid of Convolutional and Radiomic Features
	5 Results
		5.1 Evaluation Metrics
		5.2 Average Performance
	6 Conclusions and Scope for Future Work
	References
Improving Masked Face Recognition Using Dense Residual Unit Aided with Quadruplet Loss
	1 Introduction
	2 Literature Review
	3 Methodology
		3.1 Dense Residual Unit
		3.2 Triplet Loss
		3.3 Quadruplet Loss
	4 Materials and Implementation
		4.1 Pre Trained Face Recognition Model Backbone
		4.2 Simulated Mask Generation Technique
		4.3 Datasets
		4.4 Experimental Settings
		4.5 Model Training Setup
		4.6 Evaluation Metrics
	5 Experimental Evaluations
		5.1 Degradation of Existing Face Recognition Model Performance Due to Mask
		5.2 Improvement in Verification Performance with the Addition of DRU Trained with Quadruplet Loss
		5.3 Comparison of Triplet and Quadruplet Loss Performance
		5.4 Comparison of Resnet-101 and MobileFaceNet Backbone
		5.5 Comparison of Each Backbones Performance on LFW and MFR2
		5.6 Correct vs Incorrect Results
		5.7 Comparison with the State-of-the-Art (SoTA)
	6 Discussion
	7 Conclusion
	References
MobileACNet: ACNet-Based Lightweight Model for Image Classification
	1 Introduction
	2 Related Work
	3 MobileACNet
		3.1 Adaptively Connected Neural Network (ACNet)
		3.2 Our Proposed Variant of ACNet
		3.3 Architecture of the Proposed MobileACNet
	4 Experiments
		4.1 Datasets
		4.2 Experiments of MobileACNet
		4.3 Analysis and Discussion
	5 Conclusions
	References
GRETINA: A Large-Scale High-Quality Generated Retinal Image Dataset for Security and Privacy Assessment
	1 Introduction
	2 Publicly Available Retinal Datasets
	3 Related Work
	4 Synthetic Retinal Data Generation Approach
		4.1 Training GANs with Limited Data
		4.2 Evaluating the Generated Images
		4.3 Experimental Setup
		4.4 Experimental Results
	5 Conclusion
	References
Texture Generation Using a Graph Generative Adversarial Network and Differentiable Rendering
	1 Introduction
	2 Related Work
		2.1 Generative Adversarial Networks
		2.2 Differentiable Rendering
		2.3 Texturing
		2.4 Deformable Models
		2.5 Graph Neural Networks
	3 Models
		3.1 Model-Baseline
		3.2 Model-UV
		3.3 Model-Deformable
		3.4 Model-Graph
	4 Experiments
	5 Results
	6 Conclusion
	References
Medical VQA: MixUp Helps Keeping it Simple
	1 Introduction
	2 Related Work
	3 Datasets
	4 Methodology
		4.1 MixUp (MX)
		4.2 VQAMixUp (VMX)
		4.3 MixPool (MP)
		4.4 Question Generation (QG)
		4.5 SuperLoss (SL)
		4.6 Label Smoothing (LS)
		4.7 Stacked Attention Network (SAN)
	5 Experimental Results
		5.1 Model Training
		5.2 Model Comparison
		5.3 Results
		5.4 VQAMixUp Discussion
	6 Conclusion
	References
TTF-ST: Diversified Text to Face Image Generation Using Best-Match Search and Latent Vector Transformation
	1 Introduction
	2 Related Work
		2.1 Text Classification and Caption Generation
		2.2 Image Generation
		2.3 Text-To-Face Image Synthesis
	3 Proposed Methodology
		3.1 Overview
		3.2 Text Encoder
		3.3 Image Generator
		3.4 Image Encoder
		3.5 Dataset
		3.6 Best-Match Search
		3.7 Latent Vector Transformation
	4 Results
		4.1 Pipeline Execution Example
		4.2 Quantitative Evaluation
	5 Conclusion and Future Work
	References
Lensless Image Reconstruction with an Untrained Neural Network
	1 Introduction
	2 Theoretical Basis
	3 Experimental Setup
	4 Methodology
		4.1 Network Architecture
		4.2 Pipeline
	5 Results and Analysis
		5.1 Visual Comparison
		5.2 Ablation Study
		5.3 Discussion
	6 Conclusion
	References
A Lexicon and Depth-Wise Separable Convolution Based Handwritten Text Recognition System
	1 Introduction
	2 Related Work
	3 System Design
		3.1 Convolutional Layer
		3.2 Recurrent Layer
		3.3 Decoder
	4 Experimental Setup and Results
		4.1 Datasets
		4.2 Preprocessing
		4.3 Evaluation Metric
		4.4 Training Details
		4.5 Results and Comparison
	5 Discussion
	6 Conclusion
	References
Face Recognition System Using Multicolor Image Analysis and Template Protection with BioCryptosystem
	1 Introduction
	2 Proposed Methodology
		2.1 Face Recognition
		2.2 BioCryptosystem
	3 Experimental Results
		3.1 Databases Used
		3.2 Results and Discussion
		3.3 Security Analysis of the Proposed Template Protection Scheme
		3.4 Complexity Analysis of the Proposed Template Protection Scheme
		3.5 Novelty of the Proposed System
	4 Conclusion and Future Directions
	References
Conformer-Based Lip-Reading for Japanese Sentence
	1 Introduction
	2 Datasets
		2.1 Existing Public Datasets
		2.2 Our Datasets
	3 Proposed Method
		3.1 Scene Division and Construction
		3.2 Conformer-Based Lip-Reading
	4 Evaluation Experiment
		4.1 Experimental Condition
		4.2 Experimental Result
	5 Conclusion
	References
Immuno-Inspired Augmentation of Siamese Neural Network for Multi-class Classification
	1 Introduction
	2 Background and Related Work
		2.1 Siamese Neural Network
		2.2 Immune Network
	3 Methodology
		3.1 Affinity Function
		3.2 Churning Gold Standards
		3.3 Creating a Network of Gold Standards
		3.4 Classification of a Test Sample
	4 Experimental Analyses and Results
		4.1 Comparison of GS Selection Methods
		4.2 Impact of in During Inferencing
		4.3 Comparison of STIN with KNN Based Approach for Multi-class Classification
		4.4 Comparison of STIN and Its Variants with Random Sampling Based Approaches for Multi-class Classification
		4.5 Comparison over Multiple Metrics for Multi-class Classification
	5 Conclusions and Future Work
	References
PCMask: A Dual-Branch Self-supervised Medical Image Segmentation Method Using Pixel-Level Contrastive Learning and Masked Image Modeling
	1 Introduction
	2 Related Work
		2.1 Self-supervised Learning in Natural Images
		2.2 Self-supervised Learning in Medical Images
	3 Method
		3.1 Overview
		3.2 Masked Image Modeling Branch
		3.3 Pixel-Level Contrastive Learning Branch
		3.4 Loss Function
	4 Experiments and Results
		4.1 Experimental Setup
		4.2 Results
	5 Conclusion
	References
Vision Transformer-Based Bark Image Recognition for Tree Identification
	1 Introduction
	2 Proposed Method
		2.1 ROI Extraction
		2.2 Recognition
	3 Experiment
		3.1 Datasets
		3.2 Recognition Experiment
	4 Conclusion
	References
Author Index