Presentation 2022-05-19
Variational Autoencoders Conditioned by Contrastive Features as Style-Feature Extractors
Suguru Yasutomi, Toshihisa Tanaka,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Extracting style features is crucial for investigating the characteristics of data. This paper proposes a variational autoencoder that extracts style features by adding features of contrastive learning as a condition. We can regard the contrastive features as style-independent by assuming that the data augmentation is a perturbation of style. We add the style-independent contrastive features to the input of the decoder, aiming to make the encoder cover the style features for reconstruction. Experiments on MNIST show qualitatively that the proposed method can extract style features. Additional experiments on DAISO-100 evaluate the performance of extracting style quantitatively.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) contrastive learning / variational autoencoders / style extraction
Paper # SIP2022-3,BioX2022-3,IE2022-3,MI2022-3
Date of Issue 2022-05-12 (SIP, BioX, IE, MI)

Conference Information
Committee SIP / BioX / IE / MI / ITE-IST / ITE-ME
Conference Date 2022/5/19(2days)
Place (in Japanese) (See Japanese page)
Place (in English) Kumamoto University Kurokami Campus
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair Yukihiro Bandou(NTT) / Hitoshi Imaoka(NEC) / Kazuya Kodama(NII) / Hidekata Hontani(Nagoya Inst. of Tech.)
Vice Chair Toshihisa Tanaka(Tokyo Univ. Agri.&Tech.) / Takayuki Nakachi(Ryukyu Univ.) / Masatsugu Ichino(Univ. of Electro-Comm.) / Naoyuki Takada(SECOM) / Hiroyuki Bandoh(NTT) / Toshihiko Yamazaki(Univ. of Tokyo) / Hideaki Haneishi(Chiba Univ.) / Takayuki Kitasaka(Aichi Inst. of Tech.)
Secretary Toshihisa Tanaka(Xiaomi) / Takayuki Nakachi(Takushoku Univ.) / Masatsugu Ichino(Tokyo Univ. Agri.&Tech.) / Naoyuki Takada(KDDI Research) / Hiroyuki Bandoh(MitsubishiElectric) / Toshihiko Yamazaki(KDDI Research) / Hideaki Haneishi(Nagoya Inst. of Tech.) / Takayuki Kitasaka(Yamaguchi Univ.) / (Univ. of Hyogo)
Assistant Taichi Yoshida(UEC) / Seisuke Kyochi(Univ. of Kitakyushu) / Hiroyuki Suzuki(Gunma Univ) / Akihiro Hayasaka(NEC) / Shunsuke Iwamura(NHK) / Shinobu Kudo(NTT) / Hotaka Takizawa(Tsukuba Univ.) / Yoshito Otake(NAIST)

Paper Information
Registration To Technical Committee on Signal Processing / Technical Committee on Biometrics / Technical Committee on Image Engineering / Technical Committee on Medical Imaging / Technical Group on Information Sensing Technologies / Technical Group on Media Engineering
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Variational Autoencoders Conditioned by Contrastive Features as Style-Feature Extractors
Sub Title (in English)
Keyword(1) contrastive learning
Keyword(2) variational autoencoders
Keyword(3) style extraction
1st Author's Name Suguru Yasutomi
1st Author's Affiliation Tokyo University of Agriculture and Technology(TUAT)
2nd Author's Name Toshihisa Tanaka
2nd Author's Affiliation Tokyo University of Agriculture and Technology(TUAT)
Date 2022-05-19
Paper # SIP2022-3,BioX2022-3,IE2022-3,MI2022-3
Volume (vol) vol.122
Number (no) SIP-28,BioX-29,IE-30,MI-31
Page pp.pp.13-18(SIP), pp.13-18(BioX), pp.13-18(IE), pp.13-18(MI),
#Pages 6
Date of Issue 2022-05-12 (SIP, BioX, IE, MI)