Hands-On Generative Adversarial Networks with Keras

Initializations

The following are the steps:

1. Similar to most GAN training setups, we instantiate iterators to load data, build the discriminator and generator, and then compile the discriminator and generator that will be used during training, including defining their loss functions and optimizers.
2. First, instantiate the logger, as follows:

    logger = SummaryWriter(out_dir)

3. Then, load the data, as follows:

    batch_size = MINIBATCH_OVERWRITES[0]    train_data, val_data = get_data(data)    train_iterator = iterate_minibatches(train_data, batch_size)    val_iterator = iterate_minibatches(val_data, n_checkpoint_images)

4. Next, build the models, as follows:

    D = Discriminator(n_channels, resolution, n_labels) G = Generator(n_channels, resolution, n_labels) ...