GAN-Assisted YUV Pixel Art Generation

Abstract

Procedural Content Generation (PCG) in games has grown in popularity in recent years, with Generative Adversarial Networks (GANs) providing a promising option for applying PCG for game artistic asset generation. In this paper, we introduce a model that uses GANs and the YUV colour encoding system for automatic colouring of game assets. In this model, conditional GANs in Pix2Pix architecture are chosen as the main structure and the YUV colour encoding system is used for data preprocessing and result visualisation. We experimented with parameter settings (number of epochs, activation functions, optimisers) to optimise output. Our experimental results show that the proposed model can generate evenly coloured outputs for both small and larger datasets.