| 電子情報通信学会 研究会発表申込システム 講演論文 詳細 |
技報閲覧サービス [ログイン] 技報アーカイブ |
| トップに戻る | 前のページに戻る | [Japanese] / [English] |
| 講演抄録／キーワード | ||
| 講演名 | 2021-10-09 09:15 Cross-modal CycleGAN for Low-Resource Anime Style Face Translation ○Shiping Deng（Hosei Univ./USTC）・Kaoru Uchida（Hosei Univ.）  PRMU2021-18 |
|
| 抄録 | （和） | Human face to anime face translation has attracted the attention of many researchers in recent years, and various works have achieved high-quality style transfer on conventional tasks. However, existing works often have fatal shortcomings when the target domain training data is heavily insufficient, which is named as imbalanced (low-resource) setting. Here the low-resource task, generally means there is no sufficient data on the training dataset compared with the conventional task, e.g. the training data size is fewer than 100. To solve this problem, we propose a multimodal low-resource translation model for a specific style. Based on the cyclic adversarial network and class activation map, we import semantic modality to enhance data information and attention modules, which will help our model focus more on the important areas distinguishing the source domain from the target domain. Unlike the previous unsupervised learning of single modality, our model successfully completes image translation in cross-modal situations by importing pre-trained text-image alignment model. In addition, the use of an asymmetric structure speeds up training and flexibly generates images of the target domain. The experimental results show that our method has superiority in low-resource settings compared with the existing work of the same type. |
| （英） | Human face to anime face translation has attracted the attention of many researchers in recent years, and various works have achieved high-quality style transfer on conventional tasks. However, existing works often have fatal shortcomings when the target domain training data is heavily insufficient, which is named as imbalanced (low-resource) setting. Here the low-resource task, generally means there is no sufficient data on the training dataset compared with the conventional task, e.g. the training data size is fewer than 100. To solve this problem, we propose a multimodal low-resource translation model for a specific style. Based on the cyclic adversarial network and class activation map, we import semantic modality to enhance data information and attention modules, which will help our model focus more on the important areas distinguishing the source domain from the target domain. Unlike the previous unsupervised learning of single modality, our model successfully completes image translation in cross-modal situations by importing pre-trained text-image alignment model. In addition, the use of an asymmetric structure speeds up training and flexibly generates images of the target domain. The experimental results show that our method has superiority in low-resource settings compared with the existing work of the same type. |
|
| キーワード | （和） | Data-imbalanced / style translation / deep neural network / / / / / |
| （英） | Data-imbalanced / style translation / deep neural network / / / / / | |
| 文献情報 | 信学技報, vol. 121, no. 192, PRMU2021-18, pp. 11-16, 2021年10月. | |
| 資料番号 | PRMU2021-18 | |
| 発行日 | 2021-10-01 (PRMU) | |
| ISSN | Online edition: ISSN 2432-6380 | |
| 著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します．(許諾番号：10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) | |
| PDFダウンロード | PRMU2021-18 |
|
| 研究会情報 | |
| 研究会 | PRMU |
| 開催期間 | 2021-10-08 - 2021-10-09 |
| 開催地（和） | オンライン開催 |
| 開催地（英） | Online |
| テーマ（和） | 研究を効率的に進めるためのプロセスや周辺技術 |
| テーマ（英） | Processes and technologies to make research more efficient |
| 講演論文情報の詳細 | |
| 申込み研究会 | PRMU |
| 会議コード | 2021-10-PRMU |
| 本文の言語 | 英語 |
| タイトル（和） | |
| サブタイトル（和） | |
| タイトル（英） | Cross-modal CycleGAN for Low-Resource Anime Style Face Translation |
| サブタイトル（英） | |
| キーワード(1)（和/英） | Data-imbalanced / Data-imbalanced |
| キーワード(2)（和/英） | style translation / style translation |
| キーワード(3)（和/英） | deep neural network / deep neural network |
| キーワード(4)（和/英） | / |
| キーワード(5)（和/英） | / |
| キーワード(6)（和/英） | / |
| キーワード(7)（和/英） | / |
| キーワード(8)（和/英） | / |
| 第1著者 氏名（和/英/ヨミ） | Shiping Deng / Shiping Deng / |
| 第1著者 所属（和/英） | Hosei University/University of Science and Technology of China (略称： 法政大/中国科技大) Hosei University/University of Science and Technology of China (略称： Hosei Univ./USTC) |
| 第2著者 氏名（和/英/ヨミ） | 内田 薫 / Kaoru Uchida / ウチダ カオル |
| 第2著者 所属（和/英） | 法政大学 (略称： 法政大) Hosei University (略称： Hosei Univ.) |
| 講演者 | 第1著者 |
| 発表日時 | 2021-10-09 09:15:00 |
| 発表時間 | 15分 |
| 申込先研究会 | PRMU |
| 資料番号 | PRMU2021-18 |
| 巻番号（vol） | vol.121 |
| 号番号（no） | no.192 |
| ページ範囲 | pp.11-16 |
| ページ数 | 6 |
| 発行日 | 2021-10-01 (PRMU) |
[研究会発表申込システムのトップページに戻る]