In this repo we present the results of our work around Cycle GANs as part of the Deep Learning course taught at ENSAE by Marco Cuturi (2021 class).
The repo consists of :
-
one detailed report (pdf) structured as follows :
- (1) a short bibliographical review (DC-GAN and Wasserstein GAN)
- (2) technical details on the Cycle GAN architecture
- (3) analysis of our implementation on the MNIST/USPS datasets
-
one Python notebook (designed to be run on Google Collab) where we implement a CycleGAN architecture in order to translate from the MNIST dataset to the USPS dataset. We follow Roger Grosse's detailed assignment (handout and base code can be found here and here).