AEGAN-keras
To use, run the script train.py with the following arguments:
-t TYPE
The type of network. Must be one of "GAN", "AE", "AEGAN"
For Generative Adversarial Network, Autoencoder,
or Autoencoding Generative Adversarial Network
-n NAME
The name of the experiment to run (must be unique)
Files will be saved to "../results/<NAME>_<TYPE>/
-p PLOT_EVERY
The number of epochs between saving samples
-l LATENT_DIM
The dimension of the latent space
-b BATCH SIZE
Size of the minibatch
-d DATA_DIR
Directory where images are saved. If not specified,
MNIST is used
-e EPOCHS
Number of epochs to train for
-s STEPS
Number of training steps per epoch
-f PARAMETER_FILE
Parameters for building the network. See
code/params/params_64.json as an example.
Images are resized to match the output size
of the generator.
-x FLIP
Whether (True) or not (False) to flip images before training