This is a project for experiments with GAN's

• Pytorch implementation

• Gan's are trained on toy datasets: gaussian grid, gaussian circle, gaussian spiral

• There are spectral normalization and gradient penalty options with different loss configurations

• Implemented WGAN-div, LSGAN, original GAN and relativistic GAN

• Also there is code for visualizing gan training

Explore src/toygan.py for more details

Spectral normalization in discriminator with gradient penalty.

Same with layer normalization in generator