Philosophy

Nowadays there are a lot of repositories for training Generative Adversarial Networks in Pytorch, however, there are some challenges which still remain:

• Most of these are not documented
• Majority of them are not maintained
• They are built without considering the ease of usage in mind
• These are not properly tested and often are not supported by the newer releases of Pytorch
• There is no proper unified API among these repositories

The idea of this framework is to provide an elegant design to solve issues regarding training and visualizing GANs. The design principles of this framework are the following:

• A common unified API for designing GANs
• Well documented code and API
• Proper examples to facilitate ease of use
• Easy to integrate with your applications
• Provide a easy API for fast prototyping and research
• Provide advanced features without taking away the ability to customize from users
• Presence of popular loss functions, metrics and modules from cutting edge research