def main():
if is_svhn is True:
dataset = load_svhn()
else:
tf.logging.info("Read data from {}".format("MNIST_data"))
dataset = input_data.read_data_sets('MNIST_data', one_hot=True)
with tf.device(device):
tf.logging.info("Build graph on {}".format(device))
opt_g, opt_c, real_data, c_loss, g_loss, fake_logit, true_logit = build_graph(
) # opt_g, opt_c, real_data = build_graph()
merged_all = tf.summary.merge_all()
saver = tf.train.Saver()
config = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)
config.gpu_options.allow_growth = True
config.gpu_options.per_process_gpu_memory_fraction = 1.0
def next_feed_dict():
train_img = dataset.train.next_batch(batch_size)[0]
train_img = 2 * train_img - 1
if is_svhn is not True:
train_img = np.reshape(train_img, (-1, 28, 28))
npad = ((0, 0), (2, 2), (2, 2))
train_img = np.pad(train_img,
pad_width=npad,
mode='constant',
constant_values=-1)
train_img = np.expand_dims(train_img, -1) # [batch, 32, 32, 1]
feed_dict = {real_data: train_img}
return feed_dict
with tf.Session(config=config) as sess:
sess.run(tf.global_variables_initializer())
summary_writer = tf.summary.FileWriter(log_dir, sess.graph)
for i in range(max_iter_step):
if i < 25 or i % 500 == 0:
citers = 100
else:
citers = Citers
run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
for j in range(citers):
feed_dict = next_feed_dict()
if i % 100 == 99 and j == 0:
tf.logging.info("Save critic model at step %i" % i)
_, merged, c_loss_np, fake_logit_np, true_logit_np = sess.run(
[opt_c, merged_all, c_loss, fake_logit, true_logit],
feed_dict=feed_dict,
options=run_options,
run_metadata=run_metadata)
summary_writer.add_summary(merged, i)
summary_writer.add_run_metadata( # save the meta data every 100 steps.
run_metadata, 'critic_metadata {}'.format(i), i)
tf.logging.info(
"Step = {}, c_loss = {:.4f}, f_logit = {}, r_logit = {}"
.format(i, c_loss_np, fake_logit_np[:5],
true_logit_np[:5]))
else:
_, c_loss_np = sess.run([opt_c, c_loss],
feed_dict=feed_dict)
#if i % 100 == 99:
#tf.logging.info("Optimize the critic model step {}/{}, c_loss = {:.4f}".format(j, i, c_loss_np))
feed_dict = next_feed_dict()
if i % 100 == 99:
tf.logging.info("Saving generator model as step %i" % i)
_, merged, g_loss_np, fake_logit_np = sess.run(
[opt_g, merged_all, g_loss, fake_logit],
feed_dict=feed_dict,
options=run_options,
run_metadata=run_metadata)
summary_writer.add_summary(merged, i)
summary_writer.add_run_metadata(
run_metadata, 'generator_metadata {}'.format(i), i)
tf.logging.info(
"Step = {}, g_loss = {:.4f}, f_logit = {}".format(
i, g_loss_np, fake_logit_np[:5]))
else:
# tf.logging.info("Optimize the generator model!")
sess.run(opt_g, feed_dict=feed_dict)
if i % 1000 == 999:
saver.save(sess,
os.path.join(ckpt_dir, "model.ckpt"),
global_step=i)
def main():
if is_svhn is True:
dataset = load_svhn()
else:
dataset = input_data.read_data_sets('MNIST_data', one_hot=True)
with tf.device(device):
opt_g, opt_c, real_data = build_graph()
merged_all = tf.summary.merge_all()
saver = tf.train.Saver()
config = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=True)
config.gpu_options.allow_growth = True
config.gpu_options.per_process_gpu_memory_fraction = 0.8
def next_feed_dict():
train_img = dataset.train.next_batch(batch_size)[0]
train_img = 2 * train_img - 1
if is_svhn is not True:
train_img = np.reshape(train_img, (-1, 28, 28))
npad = ((0, 0), (2, 2), (2, 2))
train_img = np.pad(train_img,
pad_width=npad,
mode='constant',
constant_values=-1)
train_img = np.expand_dims(train_img, -1)
feed_dict = {real_data: train_img}
return feed_dict
with tf.Session(config=config) as sess:
sess.run(tf.global_variables_initializer())
summary_writer = tf.summary.FileWriter(log_dir, sess.graph)
for i in range(max_iter_step):
if i < 25 or i % 500 == 0:
citers = 100
else:
citers = Citers
for j in range(citers):
feed_dict = next_feed_dict()
if i % 100 == 99 and j == 0:
run_options = tf.RunOptions(
trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
_, merged = sess.run([opt_c, merged_all],
feed_dict=feed_dict,
options=run_options,
run_metadata=run_metadata)
summary_writer.add_summary(merged, i)
summary_writer.add_run_metadata(
run_metadata, 'critic_metadata {}'.format(i), i)
else:
sess.run(opt_c, feed_dict=feed_dict)
feed_dict = next_feed_dict()
if i % 100 == 99:
_, merged = sess.run([opt_g, merged_all],
feed_dict=feed_dict,
options=run_options,
run_metadata=run_metadata)
summary_writer.add_summary(merged, i)
summary_writer.add_run_metadata(
run_metadata, 'generator_metadata {}'.format(i), i)
else:
sess.run(opt_g, feed_dict=feed_dict)
if i % 1000 == 999:
saver.save(sess,
os.path.join(ckpt_dir, "model.ckpt"),
global_step=i)
beta2 = 0.9
max_iter_step = 20000
channels = 1 if dataset_name == 'mnist' else 3
log_path = './' + dataset_name + '/log_wgan'
ckpt_path = './' + dataset_name + '/ckpt_wgan'
ckpt_step_path = ckpt_path + '.step'
##################################################################
if dataset_name == 'mnist':
dataset = input_data.read_data_sets('MNIST_data', one_hot=True)
ndf = ngf = 16
elif dataset_name == 'celeba':
download_celeb_a()
dataset = glob(os.path.join('./data/', 'celebA/*.jpg'))
count = len(dataset)
elif dataset_name == 'svhn':
dataset = load_svhn()
##################################################################
def get_image(image_path):
image = Image.open(image_path)
if image.size != (image_width, image_height):
face_width = face_height = 108
j = (image.size[0] - face_width) // 2
i = (image.size[1] - face_height) // 2
image = image.crop([j, i, j + face_width, i + face_height])
image = image.resize([image_width, image_height], Image.BILINEAR)
return np.array(image.convert('RGB'))
def get_batches():