WEB OF SCIENCE
SCOPUS
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.advisor | 박상현 | - |
| dc.contributor.author | Soopil Kim | - |
| dc.date.accessioned | 2025-01-20T21:38:52Z | - |
| dc.date.available | 2025-08-31T06:00:30Z | - |
| dc.date.issued | 2024 | - |
| dc.identifier.uri | http://hdl.handle.net/20.500.11750/57579 | - |
| dc.identifier.uri | http://dgist.dcollection.net/common/orgView/200000799013 | - |
| dc.description | Deep Learning, Semantic Segmentation, Few-Shot Learning, Meta-learning, Semi-Supervision, Partially Labeled Datasets, Federated Learning | - |
| dc.description.tableofcontents | Contents i List of Tables v List of Figures viii I. Introduction 1 1. Background and Motivations 1 2. Contributions and Outline 2 3. Publications 4 3.1 Excluded Research 4 II. Meta Learning-Based Semi-Supervised Few-Shot Segmentation 6 1. Introduction 6 2. Related Works 8 2.1 Few Shot Learning for Semantic Segmentation 8 2.2 Pseudo Labels in Semi-Supervised Segmentation 9 2.3 Uncertainty Estimation in Neural Network 9 3. Methods 10 3.1 Problem Setup 10 3.2 Prototype-Based Few Shot Segmentation 11 3.3 Neural Uncertainty Estimation 13 3.4 Semi-Supervised Few Shot Segmentation 14 3.5 Refinement Module 15 3.6 Implementation Details 15 4. Experimental Results 16 4.1 Experimental Setting 16 4.2 Comparison with State-of-the-Art Models 16 4.3 Qualitative Results 18 4.4 T-SNE Visualization 19 4.5 Comparison with PPNet 20 4.6 Effect of Uncertainty Estimation 22 4.7 Effect of the Refinement Module and the number of Unlabeled Images 23 4.8 Effect of the number of Clusters 23 5. Conclusion 23 III. Semi-Supervised Few-Shot Segmentation of Industrial Images with Various Components 26 1. Introduction 26 2. Related Works 28 2.1 Anomaly Detection in Industrial Images 28 2.2 Object Part Segmentation 29 2.3 Few Shot Semantic Segmentation 30 3. Methods 31 3.1 Problem Setting 31 3.2 Overview 31 3.3 Part Segmentation Using Limited Annotations 32 3.4 Handling Multiple Types of Products 33 3.5 Anomaly Detection Using Part Segmentation 33 3.6 Class Histogram Memory Bank 33 3.7 Component Composition Memory Bank 33 3.8 Patch Representation Memory Bank 34 3.9 Aggregating Anomaly Scores of Different Scales 34 3.10 Implementation Details 35 4. Experiments 35 4.1 Experimental Setting 35 4.2 Comparison with State-of-the-art Methods 35 4.3 Qualitative Comparison of FSS Methods 36 4.4 Anomaly Detection Using Different Segmentation Models 38 4.5 Effect of Various Memory Banks And Adaptive Scaling 39 4.6 Logical Anomaly Detection Using Less Training Samples 39 5. Conclusion 40 IV. Few-Shot Learning for 3D Medical Image Segmentation 41 1. Introduction 41 2. Related Works 43 2.1 Few Shot Learning for Segmentation 43 2.2 Recurrent Neural Networks for 3D Medical Image Segmentation 43 2.3 Transfer Learning in Few Shot Learning 44 3. Methods 44 3.1 Problem Setup 44 3.2 Bidirectional RNN-based Few Shot Learning 46 3.3 Transfer Learning-based Adaptation 48 4. Experiments 49 4.1 Dataset 49 4.2 Experimental Settings 49 4.3 Results and Discussion 50 5. Conclusion 53 V. Label-Efficient 3D Medical Image Segmentation with Distributed Partially Labeled Datasets 55 1. Introduction 55 2. Related Works 57 2.1 Learning a Model with Partially Labeled Datasets 57 2.2 Federated Learning 57 3. Methods 58 3.1 Problem Setup 58 3.2 Federated Averaging 59 3.3 Global Knowledge Distillation for Federated Learning 61 3.4 Local Knowledge Distillation for Federated Learning 62 3.5 Baseline Model Architecture 63 3.6 Implementation Details 63 4. Experiments 64 4.1 Experimental Setting 64 4.2 Results 68 5. Conclusion 75 VI. Communication-Efficient 3D Medical Image Segmentation with Distributed Partially Labeled Datasets 77 1. Introduction 77 2. Related Works 79 2.1 One-Shot Federated Learning 79 2.2 Learning a Model with Partially Labeled Datasets 80 2.3 Model Inversion 80 3. Method 81 3.1 Problem Definition 81 3.2 Overview 82 3.3 Global Model Training with Knowledge Distillation 82 3.4 Image Synthesis from Segmentation Model 83 3.5 Implementation Details 84 4. Experiments 85 4.1 Dataset 85 4.2 Comparison Methods 85 4.3 Results 86 5. Conclusion 92 VII. Concluding Remarks 93 1. Conclusion 93 2. Future Work 94 VIII. Acknowledgement 96 References 97 IX. 요약문 119 |
- |
| dc.format.extent | 119 | - |
| dc.language | eng | - |
| dc.publisher | DGIST | - |
| dc.title | Label-Efficient Segmentation Models Using Limited And Distributed Pixel-Level Annotation | - |
| dc.title.alternative | 제한된 정답과 분산 데이터를 효과적으로 활용하는 영역화 모델 | - |
| dc.type | Thesis | - |
| dc.identifier.doi | 10.22677/THESIS.200000799013 | - |
| dc.description.degree | Doctor | - |
| dc.contributor.department | Department of Robotics and Mechatronics Engineering | - |
| dc.identifier.bibliographicCitation | Soopil Kim. (2024). Label-Efficient Segmentation Models Using Limited And Distributed Pixel-Level Annotation. doi: 10.22677/THESIS.200000799013 | - |
| dc.contributor.coadvisor | Kyong Hwan Jin | - |
| dc.date.awarded | 2024-08-01 | - |
| dc.publisher.location | Daegu | - |
| dc.description.database | dCollection | - |
| dc.citation | XT.RD 김56 202408 | - |
| dc.date.accepted | 2024-07-24 | - |
| dc.contributor.alternativeDepartment | 로봇및기계전자공학과 | - |
| dc.subject.keyword | Deep Learning, Semantic Segmentation, Few-Shot Learning, Meta-learning, Semi-Supervision, Partially Labeled Datasets, Federated Learning | - |
| dc.contributor.affiliatedAuthor | Soopil Kim | - |
| dc.contributor.affiliatedAuthor | Sang Hyun Park | - |
| dc.contributor.affiliatedAuthor | Kyong Hwan Jin | - |
| dc.contributor.alternativeName | 김수필 | - |
| dc.contributor.alternativeName | Sang Hyun Park | - |
| dc.contributor.alternativeName | 진경환 | - |
