Download
A composable GAN built for developers, researchers, and artists. HyperGAN builds generative adversarial networks in PyTorch and makes them easy to train and share. HyperGAN is currently in pre-release and open beta. Everyone will have different goals when using hypergan. HyperGAN is currently beta. We are still searching for a default cross-data-set configuration. Each of the examples supports search. Automated search can help find good configurations. If you are unsure, you can start with the 2d-distribution.py. Check out random_search.py for possibilities, you'll likely want to modify it. The examples are capable of (sometimes) finding a good trainer, like 2d-distribution. Mixing and matching components seems to work.
Features
- Hyperparameter tuning, or doing something like trying softmax loss with gradient penalty, can all be done in the JSON configuration file
- Custom research
- Use hyperGAN in an app
- Quantize and/or deploy the model
- Build image datasets
- HyperGAN is currently beta
Project Samples
