Research on Impression of Playing the 2D Action Game Generated by DC-GAN

Abstract

Generative Adversarial Network (GAN), a type of deep learning, can learn features from a large amount of images and generate new images according to the features. We studied a method to automatically generate the stage of a dot picture based 2D action game using GAN, and resolved the haze and object collapse that occurred in the generation result. In addition, we played the generated stage with AI and compared it with the original stage. In this study, the stage created by GAN was made to play by the person, and the impression about the game was investigated using the questionnaire.