Scene Modeling in Game Development Based on Generative Adversarial Networks

Due to booming advance on generative models, people have great interest on designing model structure to produce wonderful pictures, or even 3d shapes. The motivation of this work is that the 3d modeling manufacturing in game development is still challenging and time-consuming, and 3d shape produced from generative models may be powerful tools for the problem. The work tried to build game scenes, such as a city and a cave, the typical scene that requires many random but similar objects. This paper aims to explore a complete workflow for applying the GANs to the game development. The structure of the paper is introducing the background of the game development and the progress of the generative model, giving interpretation about the principle of Generative Adversarial Networks, and propose the process on how to utilize it to improve the productivity in developing game scene. Finally, this work found that it could considerably reduce the repetitive work on making massive and similar objects.