Colorização de mapas de jogos 2D utilizando uma rede adversaria generativa condicional

Abstract

Currently, the gaming market has taken a prominent role in the entertainment industry, becoming one of the most popular forms of digital entertainment. With technological advancements and the growth of internet access and mobile devices, electronic games have gained a global and diverse audience. In this context, game development has become a complex and challenging activity, requiring mastery of various areas, including design, programming, and art. The colorization of game maps is an important step in this process, as it contributes to the creation of visually appealing and immersive environments for players. Often, artists use sketches during this stage, which can be in grayscale or even just outlines of the environment that will become the complete map in the future. This work explores methods that automate the colorization of maps in 2D games using Conditional Generative Adversarial Networks (CGAN) through the pix2pix and CycleGAN frameworks. For this purpose, three image databases were created: one with various games and the other two with the same game but in different resolutions. By using similarity measures between the images generated by the frameworks and the real ones, it was observed that maps of the same game with lower resolution were colorized with less error in almost all measures. Consequently, in a mass production context, it is possible to reduce the artistic development time of 2D games, saving artists’ time so they can focus on making the game even more immersive.