Reconstitution of Simple Game Stage Generated by DC-GAN for Game Playing

Abstract

There have been various studies on image generation using the Generative Adversarial Network (GAN), and we have considered a game stage generation of 2D action games. In my previous study, we encoded to a simple image to decrease noises on the game stage, and the system was able to decrease noises on the generated game stage. In this study, we describe a method to restore encoded stages and generate playable game stages as a game. Specifically, we calculated the pixel values of the image generated by GAN an d the values defined by the conversion rules, converted them into objects according to the calculation results, and restored the playable game stage.