International Display Workshops General Incorporated Association

13:40 〜 14:00

[DES1/AIS1-3L(Invited)] Image Generation with a Unified Generative Adversarial Network Training via Self-Labeling and Self-Attention

*Tomoki Watanabe1、Paolo Favaro2 (1.Toshiba Corporation (Japan)、2.University of Bern (Switzerland))

Image generation, Generative adversarial network, Deep learning, Self-supervised learning

https://doi.org/10.36463/idw.2021.0859

Generative Adversarial Network(GAN) is an effective method to obtain an image generation model. We propose a novel GAN training scheme that can handle real images with any level of labeling in a unified manner by introducing a form of artificial labeling. Our scheme consistently improves the quality of generated images.