[4Pin1] インタラクティブ(2)

[4Pin1-10] Conditional Cycle GAN based Transformation of Food Image Category

〇Daichi Horita¹, Shu Naritomi¹, Ryosuke Tanno¹, Wataru Shimoda¹, Keiji Yanai¹ (1. The University of Electro-Communications, Tokyo)

Keywords:Deep Learning, Image Transformation

This paper proposes ``Food Image Transformation'' based on the Conditional
Cycle GAN (cCycle GAN) with a large-scale food image data collected from the
Twitter Stream for more than five years. By the experiments, we showed that
two hundred and thirty thousand food images with cCycle GAN enabled very
natural food photo transform among ten kinds of typical Japanese foods: ramen
noodle, curry rice, fried rice, gyodon, cold Chinese noodle, spaghetti with
meat source, white rice, eel bown, and yakisoba.