Computer Vision Systems: 13th International Conference, ICVS 2021, Virtual Event, September 22-24, 2021, Proceedings

Description

This book constitutes the refereed proceedings of the 13th International Conference on Computer Vision Systems, ICVS 2021, held in September 2021. Due to COVID-19 pandemic the conference was held virtually. The 20 papers presented were carefully reviewed and selected from 29 submissions. cover a broad spectrum of issues falling under the wider scope of computer vision in real-world applications, including among others, vision systems for robotics, autonomous vehicles, agriculture and medicine. In this volume, the papers are organized into the sections: attention systems; classification and detection; semantic interpretation; video and motion analysis; computer vision systems in agriculture. Preface Organization Contents Attention Systems Thermal Image Super-Resolution Using Second-Order Channel Attention with Varying Receptive Fields 1 Introduction 2 Second-Order Channel Attention with Varying Receptive Fields 2.1 Second-Order Channel Attention 2.2 Dilated Convolutions 2.3 Compression Through Dilations 2.4 Model Overview 3 Experimental Results 3.1 Datasets 3.2 Implementation 3.3 Evaluation 4 Conclusion References MARL: Multimodal Attentional Representation Learning for Disease Prediction 1 Introduction 2 Related Work 3 Multimodal Attentional Representation Learning 3.1 Preprocessing 3.2 Lung Segmentation with Spatial Fuzzy C-Means 3.3 Patient Biological Data Enrichment 3.4 Multimodal Representation Learning 4 Experimental Work 4.1 IPF Lung Disease Dataset 4.2 Regression of IPF Lung Disease Progression 4.3 IPF Disease Status Classification 5 Conclusion References Object Localization with Attribute Preference Based on Top-Down Attention 1 Introduction 2 Related Work 3 Architecture 3.1 Deep Attribute Network 3.2 Top-Down Attention 4 Experiments 4.1 Datasets 4.2 Pointing Game Evaluation 4.3 Experiment 1: Attribute Localization 4.4 Experiment 2: Joint Attribute-Object Localization 5 Conclusion References See the Silence: Improving Visual-Only Voice Activity Detection by Optical Flow and RGB Fusion 1 Introduction 2 Related Work 3 Our Approach 4 Experimental Evaluation 4.1 Dataset and Training 4.2 Results 5 Conclusion References Classification and Detection Score to Learn: A Comparative Analysis of Scoring Functions for Active Learning in Robotics 1 Introduction 2 Related Work 2.1 Deep Learning Based Object Detection 2.2 Weakly-Supervised Learning of Object Detection 3 Methods 3.1 Overview of the Pipeline 3.2 Scoring Functions 4 Experiments 4.1 Experimental Setup 4.2 Results Analysis on Pascal VOC 4.3 Results Analysis on iCWT 5 Discussion References Enhancing the Performance of Image Classification Through Features Automatically Learned from Depth-Maps 1 Introduction 2 Literature Review on Indoor-Outdoor Image Classification 3 Methodology 3.1 Data Set 3.2 Feature Extraction 3.3 Unsupervised Learning-Based Analysis 3.4 Supervised Learning Based Analysis 4 Results and Discussion 4.1 Experimental Setup 4.2 Results of the Unsupervised Learning Based Analysis 4.3 Results of the Supervised Learning Based Analysis 5 Conclusions and Future Work References Object Detection on TPU Accelerated Embedded Devices 1 Introduction 2 Related Works 2.1 Mobile Object Detectors 2.2 Embedded Deep Learning 3 Methodology 3.1 Framework and Quantization 3.2 Datasets 3.3 Transfer Learning 4 Evaluation 4.1 Inference Time 4.2 Accuracy 4.3 Versatility and Training Speed 4.4 Overall Performance 5 Conclusion References Tackling Inter-class Similarity and Intra-class Variance for Microscopic Image-Based Classification 1 Introduction 2 Related Work 3 Method 3.1 Inter-class Similarity 3.2 Intra-class Variance 3.3 Training Algorithm 4 Experiments 4.1 Datasets 4.2 Baselines 4.3 Evaluation Metrics 4.4 Implementation Details 5 Results 5.1 Diatom Dataset 5.2 WHOI-Plankton Dataset 6 Conclusion References Semantic Interpretation Measuring the Sim2Real Gap in 3D Object Classification for Different 3D Data Representation 1 Introduction 2 Related Work 2.1 View-Based Representation 2.2 Grid-Based Representation 2.3 Point-Based Representation 3 Measuring the Sim2Real Gap on ScanObjectNN 3.1 Experimental Setup 3.2 Results and Analysis 4 Disentangling the Impact of Design Choices 4.1 Hierarchical Learning from Object Parts 4.2 Impact of Surface-Based or Euclidean-Based Representation 4.3 Impact of Over- and Under-Segmentation and Scale 4.4 Application-Specific Considerations 5 Conclusion References Spatially-Constrained Semantic Segmentation with Topological Maps and Visual Embeddings 1 Introduction 2 Related Work 3 Methodology 3.1 Topological Representation 3.2 Structuring Vertices 3.3 Spatial Segmentation 3.4 Semantic Segmentation 4 Experimental Results 5 Conclusion and Future Work References Knowledge-Enabled Generation of Semantically Annotated Image Sequences of Manipulation Activities from VR Demonstrations 1 Introduction 2 Overview 2.1 Virtual Environment 2.2 Recorded Episodes 3 Data Collection 3.1 Episode Replay 3.2 3D Sphere Scan 4 Experiments 5 Related Work 6 Conclusion and Future Work References Make It Easier: An Empirical Simplification of a Deep 3D Segmentation Network for Human Body Parts 1 Introduction 2 Related Work 3 Methods 3.1 Annotation Algorithm 3.2 Segmentation Algorithm 4 Experiments 4.1 Model Selection 4.2 Data Augmentation 4.3 Cross-Validation 5 Conclusion References Video and Motion Analysis Video Popularity Prediction Through Fusing Early Viewership with Video Content 1 Introduction 2 Related Work 3 Proposed Video Popularity Prediction Approach 4 Experimental Results 4.1 Datasets 4.2 Evaluation 5 Conclusion References Action Prediction During Human-Object Interaction Based on DTW and Early Fusion of Human and Object Representations 1 Introduction 2 Related Work 3 DTW-Based Action Prediction 3.1 Feature Extraction 3.2 DTW-Based Time Series Alignment 3.3 Early Fusion of Human and Object Representations 4 Experiments 5 Summary References GridTrack: Detection and Tracking of Multiple Objects in Dynamic Occupancy Grids 1 Introduction 2 Related Work 3 Method 3.1 Bayesian Occupancy Grids 3.2 Semantic Grid 3.3 3D Object Detection 3.4 Grid Tracking Network 3.5 Association 3.6 Training 4 Experiments 4.1 Qualitative Results 4.2 Quantitative Results 5 Conclusion References An Efficient Video Desnowing and Deraining Method with a Novel Variant Dataset 1 Introduction 2 Related Works 2.1 Image-Based Snow/Rain Removal 2.2 Video-Based Snow/Rain Removal 3 Dataset Development 3.1 Videos with Synthetic Snow and Synthetic Rain 3.2 Videos with Quasi-snow 3.3 Videos with Real Snow and Rain 4 Proposed Method 5 Experimental Results and Comparison 5.1 Performance Evaluation 5.2 Failure Cases 6 Conclusion References Computer Vision Systems in Agriculture Robust Counting of Soft Fruit Through Occlusions with Re-identification 1 Introduction 2 Related Work 3 Methods 3.1 Tracking 3.2 Re-identification and Class Description Network 3.3 Tracking Sequences 3.4 Evaluation Metrics 4 Results 5 Conclusion References Non-destructive Soft Fruit Mass and Volume Estimation for Phenotyping in Horticulture 1 Introduction 2 Related Work 3 Methods 3.1 Volume Estimation 3.2 Mass Estimation 4 Experiments and Results 4.1 Data Collection 4.2 Phenotypic Trait Predictions 4.3 In-Field Experiments 5 Discussion 6 Conclusions and Future Work References Learning Image-Based Contaminant Detection in Wool Fleece from Noisy Annotations 1 Introduction 2 Related Work 2.1 Contaminant Detection in Wool 2.2 Semantic Segmentation 3 Semantic Segmentation from Noisy Annotations 3.1 Problem Definition 3.2 Learning Probabilistic Output 4 Wool Datasets 5 Experiments 5.1 Implementation Details 5.2 Results and Discussion 6 Conclusion References Active Learning for Crop-Weed Discrimination by Image Classification from Convolutional Neural Network's Feature Pyramid Levels 1 Introduction 2 Literature Review 2.1 Robotic Weeding 2.2 Active Learning 2.3 Convolutional Neural Networks and Features Pyramid Network 3 Proposed Method 3.1 Workflow Description 3.2 Model Training Setup 4 Experimental Setup 5 Experimental Evaluation 6 Conclusions and Future Work References Author Index