Executive Summary

Generative Adversarial Networks (GANs) have been one of the breakthroughs in the field of Neural Networks since their introduction in 2014, with Image-to-Image translation one of their widely known uses — turning a horse into a zebra, or a summer landscape into a winter setting. CycleGANs, a variant of GANs, solve one of the difficulties of image-to-image translation by eliminating the need for paired samples.

This project shows how CycleGAN works and how it can be applied to turn random landscapes into Monet-like paintings. Although the Monet-fied results are satisfying according to the groups of people shown the images, validation remains the difficult part of the GAN creation process, especially with an art-related application, due to its subjectivity.

Keywords: generative adversarial networks, CycleGAN, art generation

Full text article and source codes can be provided upon request.