This is a demo implementation of using mixup in GAN training of two 2-d toy examples as shown in the paper.

Simply run

python example_gan.py

You will need PyTorch and the tqdm package to run this script.

It may take a few hours (about 5 hours on a Nvidia GTX 1070) to run all the settings in this script. After the experiments finish, you should see a set of images similar to the ones shown here in the images folder, which are visualizations of the target distribution and generated samples during the training process. The samples folder contains the (x, y) values of corresponding samples.