[1I5-GS-2] Machine learning: Applied machine learning (2)

[1I5-GS-2-05] Japanese and Western Eclectic Wallpaper Generation with Generative Adversarial Networks

〇Kenta Tanigaki¹, Natsuki Oka¹, Kazuaki Tanaka¹, Takuma Togo¹, Fuki Wada¹ (1. Kyoto Institute of Technology)

Keywords:image generation, Conditional StyleGAN, style mixing, deep learning, Goichi Takeda

Incorporating both Japanese and Western styles is called Japanese-Western eclectic. It was also used by architect Goichi Takeda and is used in various situations. Such creative tasks are important and difficult subjects for AI. Recent developments in GAN have been remarkable and StyleGAN (Karras, Laine, and Aila, 2018) has shown that the features of different generated images can be mixed. In this research, we apply StyleGAN and mix Japanese and Western wallpapers to generate a Japanese-Western eclectic wallpaper typical of Goichi Takeda. Specifically, we generate Japanese and Western wallpapers with Conditional StyleGAN (Oeldorf and Spanakis, 2019) and mix both generated images to achieve Japanese-Western eclectic. In this task, it is important not to simply mix Japanese and Western styles but to harmonize. Therefore, we evaluated the mix balance in Japanese-Western eclectic and showed that a Japanese-Western eclectic wallpaper was generated.