| Item type |
SIG Technical Reports(1) |
| 公開日 |
2020-10-29 |
| タイトル |
|
|
タイトル |
Video Face Swapping via Disentangled Representation Learning |
| タイトル |
|
|
言語 |
en |
|
タイトル |
Video Face Swapping via Disentangled Representation Learning |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
セッション2 |
| 資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_18gh |
|
資源タイプ |
technical report |
| 著者所属 |
|
|
|
Japan Advanced Institute of Science and Technology |
| 著者所属 |
|
|
|
University of Fukui |
| 著者所属 |
|
|
|
Japan Advanced Institute of Science and Technology |
| 著者所属 |
|
|
|
University of Fukui |
| 著者所属 |
|
|
|
University of Tokyo |
| 著者所属 |
|
|
|
Japan Advanced Institute of Science and Technology |
| 著者所属(英) |
|
|
|
en |
|
|
Japan Advanced Institute of Science and Technology |
| 著者所属(英) |
|
|
|
en |
|
|
University of Fukui |
| 著者所属(英) |
|
|
|
en |
|
|
Japan Advanced Institute of Science and Technology |
| 著者所属(英) |
|
|
|
en |
|
|
University of Fukui |
| 著者所属(英) |
|
|
|
en |
|
|
University of Tokyo |
| 著者所属(英) |
|
|
|
en |
|
|
Japan Advanced Institute of Science and Technology |
| 著者名 |
Zhengyu, Huang
Chunzhi, Gu
Yi, He
Chao, Zhang
Xi, Yang
Haoran, Xie
|
| 著者名(英) |
Zhengyu, Huang
Chunzhi, Gu
Yi, He
Chao, Zhang
Xi, Yang
Haoran, Xie
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Face swapping is useful and popular technique to manipulate face from images and videos to be indistinguishable from authentic ones. However, it may cause mismatch with audio input in face appearance and shape by the conventional deep learning based approaches with facial masks. To solve this issue, we propose a novel face swapping approach using disentangled representation learning. First, the paired videos are used as training input after face alignment. We adopt GAN-based model to separate the features of the face appearance and mouth shapes explicitly by exchanging latent encodings. To maintain the continuity of video frames, we design the loss function between Inter-frame elaborately. With a given source image, the lip-sync speaking video can be generated by mimicking the target video. Our work is still in progress. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Face swapping is useful and popular technique to manipulate face from images and videos to be indistinguishable from authentic ones. However, it may cause mismatch with audio input in face appearance and shape by the conventional deep learning based approaches with facial masks. To solve this issue, we propose a novel face swapping approach using disentangled representation learning. First, the paired videos are used as training input after face alignment. We adopt GAN-based model to separate the features of the face appearance and mouth shapes explicitly by exchanging latent encodings. To maintain the continuity of video frames, we design the loss function between Inter-frame elaborately. With a given source image, the lip-sync speaking video can be generated by mimicking the target video. Our work is still in progress. |
| 書誌レコードID |
|
|
収録物識別子タイプ |
NCID |
|
収録物識別子 |
AN10100541 |
| 書誌情報 |
研究報告コンピュータグラフィックスとビジュアル情報学（CG）
巻 2020-CG-180,
号 5,
p. 1-2,
発行日 2020-10-29
|
| ISSN |
|
|
収録物識別子タイプ |
ISSN |
|
収録物識別子 |
2188-8949 |
| Notice |
|
|
|
SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc. |
| 出版者 |
|
|
言語 |
ja |
|
出版者 |
情報処理学会 |
IPSJ-CG20180005.pdf (1.4 MB)
