Progressive Growing GAN for Terrain Generation

Abstract

Terrain generations can be done using Generative Adversarial Networks as recent works have shown that GAN can generate various kinds of objects, including terrains. Terrain generations would require high resolution for terrain's topological data. As NVidia's Progressive Growing GAN has proven to generate object's images in high resolutions, we chose to generate terrains using Progressive Growing GAN. The GAN doesn't only generate colors of surface of the earth, but it also generates the heights together. The training process provides how to utilize a Convolutional Neural Network to achieve high relevance between the colors and heights.