Identity-conditioned Face Transformations Using Generative Adversarial Networks

• Item type: SIG Technical Reports(1)
• 公開日: 2018-07-23
• タイトル: Identity-conditioned Face Transformations Using Generative Adversarial Networks
• タイトル / 言語: en
• タイトル: Identity-conditioned Face Transformations Using Generative Adversarial Networks
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 機械学習・ニューラルネットワーク
• 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: Fujitsu Laboratories LTD
• 著者所属: Fujitsu Laboratories LTD
• 著者所属: Fujitsu Laboratories LTD
• 著者所属: Fujitsu Laboratories LTD
• 著者所属: Fujitsu Laboratories LTD
• 著者所属(英): en
• 著者所属(英): Fujitsu Laboratories LTD
• 著者所属(英): en
• 著者所属(英): Fujitsu Laboratories LTD
• 著者所属(英): en
• 著者所属(英): Fujitsu Laboratories LTD
• 著者所属(英): en
• 著者所属(英): Fujitsu Laboratories LTD
• 著者所属(英): en
• 著者所属(英): Fujitsu Laboratories LTD
• 著者名: Gregorio, Nuevo Castro; Akihiko, Kasagi; Masafumi, Yamazaki; Tsuguchika, Tabaru; Atsushi, Ike
• 著者名(英): Gregorio, Nuevo Castro; Akihiko, Kasagi; Masafumi, Yamazaki; Tsuguchika, Tabaru; Atsushi, Ike
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: We research the possibility of combining Siamese networks with GANs to perform face transformations conditioned on the identity of an individual. The discriminator learns a face embedding, a vector that is dependent on the identity of the person. The generator receives an input image and a target ID embedding. The input image is transformed so that its ID vector lies close to the conditioning embedding. Our aim is to explore the space in between face attribute transformations and face swap transformations. Our transformations do not fully change the face of an individual, as face swap techniques do, but they present a more general approach when compared to binary attribute transformations. Results demonstrate the system can learn interesting and unsupervised transformations that change the face of a person in a realistic and sensible way.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: We research the possibility of combining Siamese networks with GANs to perform face transformations conditioned on the identity of an individual. The discriminator learns a face embedding, a vector that is dependent on the identity of the person. The generator receives an input image and a target ID embedding. The input image is transformed so that its ID vector lies close to the conditioning embedding. Our aim is to explore the space in between face attribute transformations and face swap transformations. Our transformations do not fully change the face of an individual, as face swap techniques do, but they present a more general approach when compared to binary attribute transformations. Results demonstrate the system can learn interesting and unsupervised transformations that change the face of a person in a realistic and sensible way.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10096105
• 書誌情報: 研究報告システム・アーキテクチャ（ARC）; 巻 2018-ARC-232, 号 12, p. 1-6, 発行日 2018-07-23
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8574
• 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