I researched the viability and usability of artificial intelligence in the production of assets for video games. By utilizing a DCGAN, I wished to explore the possibility of creating 2D image assets that could be used to speed up the pipeline of asset generation.
The work can be found at this Github Gist.
Before tackling creating a DCGAN, I decided to first implement its "ancestor," an ordinary GAN. The key difference here is that a GAN does not have a convolutional layer. This convolutional layer essentially enhances the images it is given, resulting in better generated images.