def main():
train_images, test_images, train_labels, test_labels = dataset.train_test_data(
'./PetImages_resize', batch_size)
train_size = len(train_labels)
test_size = len(test_labels)
train_dataset = dataset.create_dataset(train_images, train_labels)
train_dataset = train_dataset.cache().shuffle(
buffer_size=train_size).batch(batch_size).repeat(
num_epoch).make_one_shot_iterator().get_next()
test_dataset = dataset.create_dataset(test_images, test_labels)
test_dataset = test_dataset.cache().shuffle(
buffer_size=10).batch(test_size).make_one_shot_iterator().get_next()
with tf.Session() as sess:
model = DCGAN(sess,
train_dataset=train_dataset,
test_dataset=test_dataset,
train_size=train_size,
test_size=test_size,
batch_size=batch_size,
num_epoch=num_epoch)
model.build_model()
model.intialize_variables()
#model.create_image_from_generator()
model.train()
def main(_):
# gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.333)
# run_config = tf.ConfigProto(gpu_options=gpu_options)
# run_config.gpu_options.allow_growth = True
# with tf.Session(config=run_config) as sess:
with tf.Session() as sess:
dcgan = DCGAN(sess, FLAGS.input_height, FLAGS.input_width)
dcgan.build_model()
dcgan.train(FLAGS)
def train(args):
if args.input_width is None:
args.input_width = args.input_height
if args.output_width is None:
args.output_width = args.output_height
args.save_dir = args.save_dir + '_' + args.model
args.temp_samples_dir = args.temp_samples_dir + '_' + args.model
if not os.path.exists(args.save_dir):
os.makedirs(args.save_dir)
if not os.path.exists(args.temp_samples_dir):
os.makedirs(args.temp_samples_dir)
if args.init_from is not None:
# check if all necessary files exist
assert os.path.isfile(os.path.join(args.init_from,"config.pkl")), "config.pkl file does not exist in path %s" % args.init_from
# get ckpt
ckpt = tf.train.get_checkpoint_state(args.init_from)
# open old config and check if models are compatible
with open(os.path.join(args.init_from, 'config.pkl'), 'rb') as f:
saved_args = cPickle.load(f)
need_be_same=['y_dim','t_dim','z_dim','input_height','input_width','output_height','output_width']
for checkme in need_be_same:
assert vars(saved_args)[checkme]==vars(args)[checkme],"Command line argument and saved model disagree on '%s' "%checkme
else:
with open(os.path.join(args.save_dir, 'config.pkl'), 'wb') as f:
cPickle.dump(args, f)
run_config = tf.ConfigProto()
run_config.gpu_options.allow_growth=False
with tf.Session(config=run_config) as sess:
if args.init_from is not None:
init_from = args.init_from
args = saved_args
args.init_from = init_from
if args.model == 'DCGAN':
gan = DCGAN(
sess,
args.model,
input_width=args.input_width,
input_height=args.input_height,
output_width=args.output_width,
output_height=args.output_height,
batch_size=args.batch_size,
sample_num=args.batch_size,
y_dim=args.y_dim,
t_dim=args.t_dim,
z_dim=args.z_dim,
dataset_name=args.dataset,
input_fname_pattern=args.input_fname_pattern,
crop=args.crop,
save_dir=args.save_dir,
temp_samples_dir=args.temp_samples_dir,
tag_filename=args.tag_filename,
tag_filename_sp=args.tag_filename_sp)
elif args.model == 'WGAN':
gan = WGAN(
sess,
args.model,
input_width=args.input_width,
input_height=args.input_height,
output_width=args.output_width,
output_height=args.output_height,
batch_size=args.batch_size,
sample_num=args.batch_size,
y_dim=args.y_dim,
t_dim=args.t_dim,
z_dim=args.z_dim,
dataset_name=args.dataset,
input_fname_pattern=args.input_fname_pattern,
crop=args.crop,
save_dir=args.save_dir,
temp_samples_dir=args.temp_samples_dir,
tag_filename=args.tag_filename,
tag_filename_sp=args.tag_filename_sp,
clipping_value = args.clipping_value)
elif args.model == 'WGAN_v2':
gan = WGAN_v2(
sess,
args.model,
input_width=args.input_width,
input_height=args.input_height,
output_width=args.output_width,
output_height=args.output_height,
batch_size=args.batch_size,
sample_num=args.batch_size,
y_dim=args.y_dim,
t_dim=args.t_dim,
z_dim=args.z_dim,
dataset_name=args.dataset,
input_fname_pattern=args.input_fname_pattern,
crop=args.crop,
save_dir=args.save_dir,
temp_samples_dir=args.temp_samples_dir,
tag_filename=args.tag_filename,
tag_filename_sp=args.tag_filename_sp,
scale = args.scale)
gan.build_model()
if args.init_from is not None:
gan.load(args.init_from)
show_all_variables()
gan.train(args)
# Below is codes for visualization
OPTION = 1
def generate(args):
assert os.path.isfile(
os.path.join(args.init_from, "config.pkl")
), "config.pkl file does not exist in path %s" % args.init_from
# open old config and check if models are compatible
with open(os.path.join(args.init_from, 'config.pkl'), 'rb') as f:
saved_args = cPickle.load(f)
# parse testing texts and encode them
# with open(args.skipthoughts_model, 'rb') as f:
# skipthoughts_model = cPickle.load(f)
skipthoughts_model = skipthoughts.load_model()
test_dict = {}
with open(args.testing_text, 'r') as f:
lines = f.readlines()
for line in lines:
idx, desc = re.split(',', line)
if saved_args.y_dim == 9600:
hair = re.findall('[a-zA-Z]+ hair', desc, flags=0)
hair = [re.sub(' hair', '', hair[0])]
vec_hair = skipthoughts.encode(skipthoughts_model,
hair,
verbose=False)
eyes = re.findall('[a-zA-Z]+ eyes', desc, flags=0)
eyes = [re.sub(' eyes', '', eyes[0])]
vec_eyes = skipthoughts.encode(skipthoughts_model,
eyes,
verbose=False)
test_dict[idx] = np.concatenate([vec_hair, vec_eyes], 1)
else:
vec = skipthoughts.encode(skipthoughts_model, [desc.strip()],
verbose=False)
test_dict[idx] = vec
run_config = tf.ConfigProto()
run_config.gpu_options.allow_growth = False
with tf.Session(config=run_config) as sess:
if saved_args.model == 'DCGAN':
gan = DCGAN(sess,
saved_args.model,
input_width=saved_args.input_width,
input_height=saved_args.input_height,
output_width=saved_args.output_width,
output_height=saved_args.output_height,
batch_size=saved_args.batch_size,
sample_num=saved_args.batch_size,
y_dim=saved_args.y_dim,
t_dim=saved_args.t_dim,
z_dim=saved_args.z_dim,
dataset_name=saved_args.dataset,
input_fname_pattern=saved_args.input_fname_pattern,
crop=saved_args.crop,
save_dir=saved_args.save_dir,
temp_samples_dir=saved_args.temp_samples_dir,
tag_filename=saved_args.tag_filename,
tag_filename_sp=saved_args.tag_filename_sp,
infer=True)
elif saved_args.model == 'WGAN':
gan = WGAN(sess,
saved_args.model,
input_width=saved_args.input_width,
input_height=saved_args.input_height,
output_width=saved_args.output_width,
output_height=saved_args.output_height,
batch_size=saved_args.batch_size,
sample_num=saved_args.batch_size,
y_dim=saved_args.y_dim,
t_dim=saved_args.t_dim,
z_dim=saved_args.z_dim,
dataset_name=saved_args.dataset,
input_fname_pattern=saved_args.input_fname_pattern,
crop=saved_args.crop,
save_dir=saved_args.save_dir,
temp_samples_dir=saved_args.temp_samples_dir,
tag_filename=saved_args.tag_filename,
tag_filename_sp=saved_args.tag_filename_sp,
clipping_value=saved_args.clipping_value,
infer=True)
elif saved_args.model == 'WGAN_v2':
gan = WGAN_v2(sess,
saved_args.model,
input_width=saved_args.input_width,
input_height=saved_args.input_height,
output_width=saved_args.output_width,
output_height=saved_args.output_height,
batch_size=saved_args.batch_size,
sample_num=saved_args.batch_size,
y_dim=saved_args.y_dim,
t_dim=saved_args.t_dim,
z_dim=saved_args.z_dim,
dataset_name=saved_args.dataset,
input_fname_pattern=saved_args.input_fname_pattern,
crop=saved_args.crop,
save_dir=saved_args.save_dir,
temp_samples_dir=saved_args.temp_samples_dir,
tag_filename=saved_args.tag_filename,
tag_filename_sp=saved_args.tag_filename_sp,
scale=saved_args.scale,
infer=True)
gan.build_model()
try:
tf.global_variables_initializer().run()
except:
tf.initialize_all_variables().run()
# show_all_variables()
could_load, checkpoint_counter = gan.load(args.init_from)
if could_load:
# args.sample_dir = args.sample_dir + str(checkpoint_counter)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
else:
print('load fail!!!')
for idx, vec in test_dict.items():
sample_z = np.random.uniform(-1,
1,
size=(gan.batch_size,
saved_args.z_dim))
sample_y = np.repeat(vec, gan.batch_size, axis=0)
samples = sess.run(gan.sampler,
feed_dict={
gan.z: sample_z,
gan.y: sample_y
})
print(samples.shape)
## Randomly take 5 images for each tag
for sid in range(5):
# manifold_h = int(np.ceil(np.sqrt(samples.shape[0])))
# manifold_w = int(np.floor(np.sqrt(samples.shape[0])))
# save_images(samples[sid], [1, 1],
# os.path.join(args.sample_dir, 'sample_{}_{}.jpg'.format(str(idx), str(sid+1))))
scipy.misc.imsave(
os.path.join(
args.sample_dir,
'sample_{}_{}.jpg'.format(str(idx), str(sid + 1))),
samples[sid])
from model import DCGAN
# from cocob_optimizer import COCOB
train_dir = 'train/dcgan/face_gen/'
k = 1
num_epoch = 30
save_epoch = 5
print_step = 500
summary_step = 2500
learning_rate_D = 0.0002
learning_rate_G = 0.001
if __name__ == '__main__':
model = DCGAN(batch_size=64)
model.build_model()
# set up optimization
opt_D = tf.train.AdamOptimizer(learning_rate=learning_rate_D, beta1=0.5)
opt_G = tf.train.AdamOptimizer(learning_rate=learning_rate_G, beta1=0.5)
collection_D = tf.get_collection(
tf.GraphKeys.TRAINABLE_VARIABLES, scope='Discriminator'
) # collections of variables in variable_scope('discriminator')
collection_G = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,
scope='Generator')
with tf.control_dependencies(collection_D):
train_D = opt_D.minimize(loss=model.loss_D, var_list=model.var_D)
with tf.control_dependencies(collection_G):
train_G = opt_G.minimize(loss=model.loss_G,