Modular GAN
A DCGAN implemented with keras model subclassing.
Implemented because I wanted to replicate François Chollet's tweet.
Uses MNIST dataset.
Usage
To run training :
python -m gan.main --mode=train --save '{"batch_size": 32, "epochs": 50, "dataset": "mnist", "latent_dim": 100, "buffer_size": 60000}'Once model is trained and saved, modify the following for running evaluation :
python -m gan.main --mode=evaluate --save '{"generator_path": "<model_save_path>", "latent_dim": 100, "num_images": 20}'