forked from JoannaBroniarek/DCGAN-for-MRI-images
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train3D.py
253 lines (188 loc) · 8.69 KB
/
train3D.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
import os
import tensorflow as tf
import time
import argparse
from tensorflow.keras.utils import Progbar
from models import *
from create_data import *
from utils import *
def parse_args():
""" Parsing and configuration """
desc = "Tensorflow 2.1 implementation of DCGAN for 3D MRI"
parser = argparse.ArgumentParser(description=desc)
parser.add_argument('--epochs', type=int, default=300, help='The number of epochs to run')
parser.add_argument('--batch_size', type=int, default=16, help='The size of batch')
parser.add_argument('--lr_g', type=float, default=5e-4, help='Generator learning rate')
parser.add_argument('--lr_d', type=float, default=5e-5, help='Discriminator learning rate')
parser.add_argument('--data_path', type=str, default="../train.tfrecords", help='path to train data')
parser.add_argument('--createTFrecords', type=bool, default=False, help='whether to create a TF record file')
parser.add_argument('--rand_seed', type=int, default=42, help='tf random seed')
parser.add_argument('--restore', type=str, default=None, help='path restore model path')
return parser.parse_args()
@tf.function(autograph=True)
def train_step(images, generator, discriminator, generator_optimizer, discriminator_optimizer):
"""
Training step over a batch data.
Parameters:
-----------
images - A batch of images to train on.
generator - An instance of the Generator
discriminator - An instance of the Discriminator
generator_optimizer - An instance of an optimizer for the generator
discriminator_optimizer - An instance of an optimizer for the discriminator
Returns:
--------
gen_loss - The Generator loss
disc_loss - The Discriminator Loss
"""
noise_size = 100
batch_size = int(len(images))
noise = np.random.uniform(-1, 1, size=(batch_size, noise_size))
with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:
generated_images = generator(noise, training=True)
real_output = discriminator(images, training=True)
fake_output = discriminator(generated_images, training=True)
gen_loss = generator_loss(fake_output)
disc_loss = discriminator_loss(real_output, fake_output)
gradients_of_generator = gen_tape.gradient(gen_loss, generator.trainable_variables)
gradients_of_discriminator = disc_tape.gradient(disc_loss, discriminator.trainable_variables)
generator_optimizer.apply_gradients(zip(gradients_of_generator, generator.trainable_variables))
discriminator_optimizer.apply_gradients(zip(gradients_of_discriminator, discriminator.trainable_variables))
return gen_loss, disc_loss
def train(train_type,
gen_lr, disc_lr,
dataset, num_training_samples,
epochs, batch_size, noise_size,
save_dir_path):
"""
The training phase of the DCGAN.
1. Initiate Generator & Discriminator
2. Initiate Optimizers
3. Build the Checkpoint Manager
4. Training Loop
Parameters:
-----------
train_type : {string} - Now only possible '3D' [Future: '2D']
gen_lr : {float} - An initial learning rate for a generator
disc_lr : {float} - An initial learning rate for a discriminator
dataset : <TFRecordDataset> - Dataset for a training
num_training_samples - Number of training examples
epochs : {int} - Number of epochs
batch_size : {int} - A batch size
noise_size : {int} - A noise vector dimension
save_dir_path : {str} - A path to save the intermediate training results
"""
######## 1. Initiate Generator & Discriminator #######
weight_initializer = tf.keras.initializers.TruncatedNormal(stddev=0.02, mean=0, seed=42)
if train_type=='3D':
generator=generator3d(weight_initializer=weight_initializer)
print(generator.summary())
discriminator=discriminator3d(weight_initializer=weight_initializer)
print(discriminator.summary())
########## 2. Initiate Optimizers ##########
gen_init_learning_rate = gen_lr
gen_lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(
gen_init_learning_rate,
decay_steps=600,
decay_rate=0.2,
staircase=True)
disc_init_learning_rate = disc_lr
disc_lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(
disc_init_learning_rate,
decay_steps=1000,
decay_rate=0.3,
staircase=True)
generator_optimizer=tf.keras.optimizers.Adam(gen_lr_schedule, beta_1 = 0.5)
discriminator_optimizer=tf.keras.optimizers.Adam(disc_lr_schedule, beta_1 = 0.5)
######## 3. Build the Checkpoint Manager #######
checkpoint_dir = save_dir_path
checkpoint_prefix = os.path.join("training_checkpoints", "ckpt")
checkpoint = tf.train.Checkpoint(generator_optimizer=generator_optimizer,
discriminator_optimizer=discriminator_optimizer,
generator=generator,
discriminator=discriminator)
manager = tf.train.CheckpointManager(checkpoint,
directory=checkpoint_dir,
max_to_keep = 10,
checkpoint_name=checkpoint_prefix)
try:
status = checkpoint.restore(manager.latest_checkpoint)
restored_epoch_number = int(manager.latest_checkpoint.split("-")[-1])*5
except:
restored_epoch_number = 0
print("restoring checkpoint {}".format(restored_epoch_number))
########## 4. Training Loop ##########
BUFFER_SIZE = 6000
dataset = dataset.shuffle(BUFFER_SIZE).batch(batch_size)
writer = tf.summary.create_file_writer(os.path.join(save_dir_path, 'logs'))
global_step = 0 #TODO: changed it while restoring the epoch // restored_epoch_number*batch_size
for epoch in range(restored_epoch_number, epochs):
print("\nepoch {}/{}".format(epoch+1,epochs))
pb_i = Progbar(target=num_training_samples, verbose=1)
for image_batch in dataset.as_numpy_iterator():
gen_loss, disc_loss = train_step(image_batch,
generator, discriminator,
generator_optimizer,
discriminator_optimizer)
pb_i.add(image_batch.shape[0], values=[('gen_loss', gen_loss), ('disc_loss', disc_loss)]) #TODO: show in ProgBar the average over epoch
with writer.as_default():
tf.summary.scalar('gen_loss', gen_loss, step=global_step)
tf.summary.scalar('disc_loss', disc_loss, step=global_step)
writer.flush()
global_step+=1
# Save the model every 20 epochs
if (epoch + 1) % 20 == 0:
manager.save()
# Produce an image for the GIF every 5 epochs
if (epoch + 1) % 5 == 0:
test_noise = np.random.uniform(-1, 1, size=(1, noise_size))
gen_and_save_images(generator, epoch + 1, test_noise, save_dir_path, gen_loss, disc_loss, False)
def main(args):
tf.random.set_seed(args.rand_seed)
##################################
## Create the TFRecords file ##
##################################
if args.createTFrecords:
create()
#########################
## Load Dataset ##
#########################
train_filename = args.data_path
parsed_dataset = parse_dataset(train_filename)
num_training_examples = sum(1 for _ in tf.data.TFRecordDataset(train_filename))
#########################
## Set up a path ##
#########################
tf.keras.backend.clear_session()
if args.restore:
model_name = args.restore
else:
model_name = time.strftime('%Y-%m-%d_%H:%M:%S')
if not os.path.exists(model_name):
os.mkdir(model_name)
print("saving images in: {}".format(model_name))
#########################
## Set Params ##
#########################
disc_lr = args.lr_d
gen_lr = args.lr_g
noise_dim = 100
EPOCHS = args.epochs
BATCH_SIZE = args.batch_size
#########################
## Train ##
#########################
train(train_type = '3D',
disc_lr = disc_lr,
gen_lr = gen_lr,
dataset=parsed_dataset,
num_training_samples = num_training_examples,
epochs=EPOCHS,
batch_size=BATCH_SIZE,
noise_size = noise_dim
save_dir_path = model_name)
if __name__ == "__main__":
args = parse_args()
if args is None:
exit()
main(args)