Analysis of Procedural Generated Textures for Video Games Using a CycleGAN

Abstract:

The creation of content for video games is a costly and time consuming task if done manually. Especially in modern times, the amount of content required to create a video game is expected to be much greater than in the past. Procedural Content Generation techniques can be used to alleviate this issue and allow the creation of more content with less input. The proposal in this work is to explore the use of Generative Adversarial Networks for video game textures to be developed by a computer, more specifically style-transfer with a CycleGAN. The generation experiments in this work use a variety of textures commonly used in video games such as sky-boxes and tiled images. In addition, the analysis performed has a usability perspective and an emotional one. The usability analysis detects any problems in the implementation. The affec-tive/emotional analysis measures the players’ evoked feelings and then cross-references the traditionally-generated and the computer-generated emotional result vectors e = (V, A) (composed of valence and arousal). The results obtained show an emotional transfer between the vectors to be at minimum 26.53% and at best 91.23%. These results demonstrate that the performed experiments always had a successful emotional transfer.

Año de publicación:

2022

Keywords:

• Textures
• PCG
• Procedural
• GaN
• CycleGAN
• Adversarial learning
• video games

Fuente:

scopus