GAN Theft Auto: Autonomous Texturing of Procedurally Generated Interactive Cities
Oscar Dadfar, Lingdong Huang, Hizal Çelik · 2021
We explore the possibility of producing photo-realistic and stylized videos from semantically segmented image sequences drawn from a procedurally generated interactive 3D environment. We evaluate our environment using the Cityscape and ACDC weather dataset to obtain swappable daytime, nighttime, and various weather texturings from our city. We further use the GTA V image dataset to showcase its feasibility on large interactive scenes for 3D animation and game texturing, demonstrating the ability to repurpose existing video game textures when generating our city. Our algorithm can be used to support video games by autonomizing both the environment generation process, as well as supporting researchers by providing a semantic testing environment for many city style-transfer algorithms. Video documentation at: https://tinyurl.com/vb63k87x