Artist Style Transfer using Cyclic Generative Adversarial Networks

Collaborators: Aakansha Mathur, Sayan Chakraborty

Artistic work is generally recognized through an artist’s brush strokes, color choices, color palette and so on. Each artist’s style is unique and hard to replicate. With advancements in computer vision however, Generative Adversarial Networks are now able to mimic such nuances in a very convincing way. We use GANs (specifically Cycle GAN) to replicate Claude Monet’s artistic style and see if our GAN can create realistic renditions of his work by transforming normal photos, as well as paintings by other famous artists.

Below are a couple of gifs that visualize the learning process of the cyclic-GAN network in carrying out the style transfer while self learning features from the provided artist image datasets.

The GitHub repo can be accessed here : https://github.com/cs1430-GANET/ganet