JSAI2018

Presentation information

Oral presentation

Organized Session » [Organized Session] OS-2

[1K2-OS-2b] [Organized Session] OS-2

Tue. Jun 5, 2018 3:20 PM - 5:00 PM Room K (3F Azisai Mokuren)

4:40 PM - 5:00 PM

[1K2-OS-2b-04] Character Representation Adaption in Individual Manga

〇Koki Tsubota1, Toru Ogawa1, Toshihiko Yamasaki1, Kiyoharu Aizawa1 (1. The University of Tokyo)

Keywords:Manga, Deep Learning, Adaptation

Our goal is character clustering of individual manga with bounding boxes of characters’ faces and frames. We
propose a method to adapt character representations to individual manga for clustering. We use deep metric
learning by making positive and negative pairs considering manga specific nature that the same character tends to
appear in the near page and the characters in the same frame almost always differ. We showed that adapting to
individual manga improves the accuracy of clustering.