Lawrence Du week4 slides

A deep convolutional generative adversarial network (DCGAN) was proposed to generate pixel art to assist in creating art for mobile apps in a cost-effective manner. The DCGAN was trained on a dataset of 19,231 Pokémon animation frames to learn the style of pixel art. It consists of a generator network that produces new images and a discriminator network that evaluates them for realism. Through training the networks play an adversarial game that improves the generator's ability to produce highly realistic generated images resembling the training data. The DCGAN was able to produce new pixel art in the style of Pokémon characters to potentially assist in creating large volumes of sticker art for mobile apps.