CycleGAN with Segmentation

^{For this project I worked with Andrew Chen and Sameer Manchanda. The idea for this project came from reading the Cycle GAN paper from BAIR. In this paper it shows how Cycle GAN can perform amazing unpaired image to image translation. We wanted to build off of this so we tried to look for some flaws in it. One big flaw was that there was a bleeding effect in some cases, for example when transforming horses to zebras, some of the other objects in the image besides the horse would also get zebrafied. So we decided to use segmentation to help only affect the areas of the image that we were interested in. For segmentation we utilized Mask R-CNN from FAIR.}

^{A report written detailing the process and the results can be found here while the actual code can be found here.}