|
Hokkaido University Collection of Scholarly and Academic Papers >
Graduate School of Information Science and Technology / Faculty of Information Science and Technology >
Peer-reviewed Journal Articles, etc >
Learning intra-domain style-invariant representation for unsupervised domain adaptation of semantic segmentation
This item is licensed under:Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
| Title: | Learning intra-domain style-invariant representation for unsupervised domain adaptation of semantic segmentation |
| Authors: | Li, Zongyao Browse this author | | Togo, Ren Browse this author | | Ogawa, Takahiro Browse this author →KAKEN DB | | Haseyama, Miki Browse this author →KAKEN DB |
| Keywords: | Style -invariant representation | | Self-ensembling | | Domain adaptation |
| Issue Date: | 20-Jul-2022 |
| Publisher: | Elsevier |
| Journal Title: | Pattern recognition |
| Volume: | 132 |
| Start Page: | 108911 |
| Publisher DOI: | 10.1016/j.patcog.2022.108911 |
| Abstract: | A B S T R A C T In this paper, we aim to tackle the problem of unsupervised domain adaptation (UDA) of semantic seg-mentation and improve the UDA performance with a novel conception of learning intra-domain style -invariant representation. Previous UDA methods focused on reducing the inter-domain inconsistency between the source domain and the target domain. However, due to the different data distributions of the two domains, reducing the inter-domain inconsistency cannot ensure the generalization abil-ity of the trained model in the target domain. Therefore, to improve the UDA performance, we take into consideration the intra-domain diversity of the target domain for the first time in studies on UDA and aim to train the model to generalize well to the diverse intra-domain styles. To achieve this, we propose a self-ensembling method to learn the intra-domain style-invariant representation and we in-troduce a semantic-aware multimodal image-to-image translation model to obtain images with diver-sified intra-domain styles. Our method achieves state-of-the-art performance on two synthetic-to-real adaptation benchmarks, and we demonstrate the effectiveness of our method by conducting extensive experiments. (c) 2022 Elsevier Ltd. All rights reserved. |
| Rights: | © 2022. This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/ | | http://creativecommons.org/licenses/by-nc-nd/4.0/ |
| Type: | article (author version) |
| URI: | http://hdl.handle.net/2115/92818 |
| Appears in Collections: | 情報科学院・情報科学研究院 (Graduate School of Information Science and Technology / Faculty of Information Science and Technology) > 雑誌発表論文等 (Peer-reviewed Journal Articles, etc)
|
Submitter: 李 宗曜
|
