Advanced Deep Learning with Keras

Cross-Domain GANs
models (Models): Source/Target Discriminator/Generator,
Adversarial Model
data (tuple): source and target training data
params (tuple): network parameters
test_params (tuple): test parameters
test_generator (function): used for generating predicted target
and source images
"""
# the models
g_source, g_target, d_source, d_target, adv = models
# network parameters
batch_size, train_steps, patch, model_name = params
# train dataset
source_data, target_data, test_source_data, test_target_data =
data
titles, dirs = test_params
# the generator image is saved every 2000 steps
save_interval = 2000
target_size = target_data.shape[0]
source_size = source_data.shape[0]
# whether to use patchgan or not
if patch > 1:
d_patch = (patch, patch, 1)
valid = np.ones((batch_size,) + d_patch)
fake = np.zeros((batch_size,) + d_patch)
else:
valid = np.ones([batch_size, 1])
fake = np.zeros([batch_size, 1])
valid_fake = np.concatenate((valid, fake))
start_time = datetime.datetime.now()
for step in range(train_steps):
# sample a batch of real target data
rand_indexes = np.random.randint(0, target_size,
size=batch_size)
real_target = target_data[rand_indexes]
# sample a batch of real source data
rand_indexes = np.random.randint(0, source_size,
[ 222 ]