Title Page
Contents
ABSTRACT 8
Ⅰ. Introduction 10
Ⅱ. Related work 15
2.1. Semi-supervised learning 15
2.1.1. Entropy minimization 15
2.1.2. Consistency regularization 16
2.2. Semi-supervised semantic segmentation 17
Ⅲ. Proposed method 19
3.1. Overview 19
3.2. Teacher model using multi-view integrated ensemble to generated pseudo-labels 24
3.3. Student model that outputs a binary using an ensemble of multi-view teachers 28
Ⅳ. Experiments 30
4.1. Dataset 30
4.2. Implementation details 31
4.3. Evaluation metrics 32
4.4. Comparison to sota on different partition protocols 33
Ⅴ. Conclusion 35
References 37
Abstract (in Korean) 45
〈Table 4-1〉 Binary performance comparison with SOTA for unlabeled data of PASCAL VOC 2012 under different partition protocols. Predictions from... 33
〈Table 4-2〉 Comparison of SOTA and models trained with the help of our model on the PASCAL VOC 2012 val set under different partitioning... 33
〈Figure 1-1〉 (a)-(c) show normalized confusion matrices for unlabeled data in CPS,PS-MT, and U2PL, respectively. (d) shows the intersection over... 12
〈Figure 3-1〉 Example of the transformation of the semantic ground-truth to train all our models when the number of classes is 6. 20
〈Figure 3-2〉 Overview of our proposed MVIE method. MVIE consists of teacher networks with encoder-decoder structures and a student network... 21
〈Figure 3-3〉 An example in which multi-view teacher integrated ensemble is applied to the first teacher when the number of classes is 6. 24