This repo provides several implementations of various forms of Generative Advesarial Networks (GAN)
- Original GAN (https://arxiv.org/abs/1406.2661)
- Wasserstein GAN (https://arxiv.org/abs/1701.07875)
- Deep Convolutional GAN (https://arxiv.org/pdf/1511.06434.pdf)
At the moment, all the models are tested on MNIST dataset.
To run the experiments, execute train_gan.py
To specify which model to train, use flag --model. Say, to train GAN, use --model GAN.
Additional parameters (minibatch size, learning rate) could be specified as well. See train_gan.py for details.
To plot sample, execute plot.py with proper model flag.
Default parameters (e.g. learning rate, batch size, etc.) should work fine, but it's might be a good idea to try different settings.