This repository contains the report and code of a project on Generative Adversarial Networks carried out with a friend during my Master's year.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


This repository is the result of a project I did on Generative Adversarial Networks for my Master's Computer Vision & Object Recognition class . The aim of the project was to study in depth Cycle consistent Generative Adversarial Networks.

This work was mainly based on the original article on Cycle consistent GANs : First, we reproduced some of the results presented in the article, transfering photos to Monet paintings and vice-versa, also trying out the algorithm on some holiday pictures. An experiment was also carried out on a new dataset, trying to perform style transfer from day to night using photos of roads taken from a car.

Finally, using the paper on Wasserstein loss which can be found here, we modified the loss function used in the Cycle GAN paper to implement the Wasserstein loss and compare results with the previous loss.

We used the Pytorch implementation of Cycle-GANS which can be found here :, and only modified the file in order to add a WGAN model which we implemented in the file. Those two contributions can be found in this repository.