-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathFCN_di_new.py
263 lines (236 loc) · 12.2 KB
/
FCN_di_new.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
254
255
256
257
258
259
260
261
262
263
import tensorflow as tf
from keras.layers import Input, Dense, Reshape, Flatten, Dropout, multiply, GaussianNoise
from keras.layers import BatchNormalization, Add, Activation, Embedding, ZeroPadding3D
from keras.layers.advanced_activations import LeakyReLU
from keras.utils import multi_gpu_model
from keras.layers.convolutional import Conv3D
from keras.models import Sequential, Model
from keras.optimizers import Adam
from keras import losses
from keras.utils import to_categorical
import h5py
import atexit
import random
import pandas as pd
from tqdm import trange
import time
import os
import numpy as np
import keras.backend as K
from keras.callbacks import TensorBoard
from keras.utils import multi_gpu_model
K.set_image_data_format("channels_first")
from WMUnet_3d import unet_model_3d
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "2"
tf.set_random_seed(1234)
random.seed(2)
class SR_UnetGAN():
def __init__(self):
self.data_path = '/media/data/fanxuan/data/data_h5data/data_all.h5'
self.test_data_path = '/media/data/fanxuan/data/data_h5data/data_test_drf10.h5'
#data_Siemens_wave_50_bior3_7_3d.h5
self.save_dir = '/media/data/fanxuan/result/FCN_multi_10_drf_10_6E-5'
#self.model_path = '/media/data/fanxuan/result/FCN_multi_4/model/generator_epoch_26.hdf5'
self.img_shape = (8,176,176,16)
#self.common_optimizer = AdamWOptimizer(weight_decay=1e-4, learning_rate=0.001)
self.common_optimizer = Adam(0.00006, 0.5)
self.epochs = 400
self.batch_size = 4
self.generator = self.build_generator()
self.generator.compile(loss="mse", optimizer=self.common_optimizer)
#self.generator_parallel = multi_gpu_model(self.generator,2)
#self.generator_parallel.compile(loss="mse", optimizer=self.common_optimizer)
#self.generator_parallel.summary()
def build_generator(self, path=None):
model = unet_model_3d(self.img_shape)
if path is not None:
model.load_weights(path)
print('Success Loading Model!')
model.summary()
return model
def data_generator_new(self, data, name, index_all, dose_list, length_train, length_1st):
while True:
x_batch, y_batch = [], []
count = 0
for index in index_all:
ni = 0
j = (index-1)%length_train
n = (index-1)//length_train
if j >= length_1st:
ni = 1
j = j - length_1st
# print(name+dose_list[n]+str(j))
x = data.get(name[ni]+dose_list[n])[j]
x = x*10
y = data.get(name[ni]+'_normal')[j]
y = y*10
if len(x.shape)<4:
x = np.expand_dims(x, axis=0)
x_batch.append(x)
if len(y.shape)<4:
y = np.expand_dims(y, axis=0)
y_batch.append(y)
count += 1
if count == self.batch_size:
yield np.array(x_batch), np.array(y_batch)
count = 0
x_batch, y_batch = [], []
def data_generator_test(self, data, name, index):
while True:
x_batch, y_batch = [], []
count = 0
if 1:
for j in index:
x = data.get(name+'_reduce')[j]
x = x*10
y = data.get(name+'_normal')[j]
y = y*10
if len(x.shape)<4:
x = np.expand_dims(x, axis=0)
x_batch.append(x)
if len(y.shape)<4:
y = np.expand_dims(y, axis=0)
y_batch.append(y)
count += 1
if count == 4:
yield np.array(x_batch), np.array(y_batch)
count = 0
x_batch, y_batch = [], []
def data_generator_100(self, data, name, index_all, dose_list, length_train, length_1st):
while True:
x_batch, y_batch = [], []
count = 0
for index in index_all:
ni = 0
j = index
if index >= length_1st:
ni = 1
j = j - length_1st
# print(name+dose_list[n]+str(j))
x = data.get(name[ni]+dose_list)[j]
x = x*10
y = data.get(name[ni]+'_normal')[j]
y = y*10
if len(x.shape)<4:
x = np.expand_dims(x, axis=0)
x_batch.append(x)
if len(y.shape)<4:
y = np.expand_dims(y, axis=0)
y_batch.append(y)
count += 1
if count == self.batch_size:
yield np.array(x_batch), np.array(y_batch)
count = 0
x_batch, y_batch = [], []
def write_log(self, callback, name, value, batch_no):
summary = tf.Summary()
summary_value = summary.value.add()
summary_value.simple_value = value
summary_value.tag = name
callback.writer.add_summary(summary, batch_no)
callback.writer.flush()
def train(self):
data = h5py.File(self.data_path, mode='r')
test_data = h5py.File(self.test_data_path, mode='r')
train_filenames = np.array(data['train_filenames'])
valid_filenames = np.array(data['valid_filenames'])
length_train = len(train_filenames)+len(valid_filenames)
index_all = np.arange(0,(len(train_filenames)+len(valid_filenames))*1)
random.shuffle(index_all)
#dose_list = ['_2_dose', '_4_dose', '_10_dose', '_20_dose', '_50_dose', '_100_dose']
#data_generator = self.data_generator_new(data, ['train','valid'], index_all, dose_list, length_train, len(train_filenames))
data_generator = self.data_generator_100(data, ['train','valid'], index_all, '_10_dose', length_train, len(train_filenames))
test_filenames = np.array(test_data['test_filenames'])
index_all2 = np.arange(0,len(test_filenames))
random.shuffle(index_all2)
valid_generator = self.data_generator_test(test_data, 'test', index_all2)
log_dir = os.path.join(self.save_dir, 'log')
if not os.path.exists(log_dir):
os.makedirs(log_dir)
tensorboard = TensorBoard(log_dir="{}/{}".format(log_dir, time.asctime()))
tensorboard.set_model(self.generator)
batch_df = pd.DataFrame()
epoch_df = pd.DataFrame()
batch_val = pd.DataFrame()
epoch_val = pd.DataFrame()
epoch_ori = pd.DataFrame()
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.8)
config = tf.ConfigProto(gpu_options=gpu_options)
config.gpu_options.allow_growth = True
with tf.Session(graph=tf.get_default_graph(), config=config) as sess:
sess.run(tf.global_variables_initializer())
K.set_session(sess)
def save_model():
print('Saving model.')
self.generator.save_weights(os.path.join(self.save_dir, 'generator_latest.hdf5'))
atexit.register(save_model)
self.generator.load_weights('/media/data/fanxuan/result/FCN_multi_10_drf_10_2/model/generator_epoch_7.hdf5')
# Add a loop, which will run for a specified number of epochs:
for epoch in range(1, self.epochs+1):
gen_losses = []
number_of_batches = int(len(index_all) / self.batch_size)
for index in trange(number_of_batches):
# number_of_batches
print('Epoch: {}/{}\n Batch: {}/{}'.format(epoch, self.epochs, index+1, number_of_batches))
x_batch, y_batch = next(data_generator)
#ori_loss = np.mean(np.square(np.array(y_batch) - np.array(x_batch)))
#print(" ori_loss: {}\n".format(ori_loss))
#g_loss = self.generator_parallel.train_on_batch(x_batch, y_batch)
g_loss = self.generator.train_on_batch(x_batch, y_batch)
gen_losses.append(g_loss)
print(" G_loss: {}\n".format(g_loss))
# Create two lists to store losses
val_losses = []
ori_losses = []
for ind in range(int(len(test_filenames) / 4)):
# int(len(valid_filenames) / 4)
x_val, y_val = next(valid_generator)
pred_val = self.generator.predict(x_val)
val_loss = np.mean(np.square(np.array(y_val) - np.array(pred_val)))
ori_loss = np.mean(np.square(np.array(y_val) - np.array(x_val)))
print(" val_loss: {}\n".format(val_loss))
print(" ori_loss: {}\n".format(ori_loss))
val_losses.append(val_loss)
ori_losses.append(ori_loss)
#te = Test()
#te.generate_result_3d(self.generator,epoch)
batch_df = batch_df.append(pd.DataFrame({'epoch': [epoch] * len(gen_losses),
'batch': np.arange(1, len(gen_losses)+1),
'generator_loss': gen_losses}))
epoch_df = epoch_df.append(pd.DataFrame({'epoch': [epoch],
'generator_loss': [np.mean(gen_losses)]}))
batch_df = batch_df[['epoch', 'batch', 'generator_loss']]
epoch_df = epoch_df[['epoch', 'generator_loss']]
batch_df.to_csv(os.path.join(log_dir, 'batch_loss.csv'), index=False)
epoch_df.to_csv(os.path.join(log_dir, 'epoch_loss.csv'), index=False)
batch_val = batch_val.append(pd.DataFrame({'epoch': [epoch] * len(val_losses),
'batch': np.arange(1, len(val_losses)+1),
'val_loss': val_losses}))
epoch_val = epoch_val.append(pd.DataFrame({'epoch': [epoch],
'val_loss': [np.mean(val_losses)]}))
batch_val = batch_val[['epoch', 'batch', 'val_loss']]
epoch_val = epoch_val[['epoch', 'val_loss']]
batch_val.to_csv(os.path.join(log_dir, 'batch_val_loss.csv'), index=False)
epoch_val.to_csv(os.path.join(log_dir, 'epoch_val_loss.csv'), index=False)
epoch_ori = epoch_ori.append(pd.DataFrame({'epoch': [epoch],
'ori_loss': [np.mean(ori_losses)]}))
epoch_ori = epoch_ori[['epoch', 'ori_loss']]
epoch_ori.to_csv(os.path.join(log_dir, 'epoch_ori_loss.csv'), index=False)
# Save losses to Tensorboard
self.write_log(tensorboard, 'generator_loss', np.mean(gen_losses), epoch)
model_dir = os.path.join(self.save_dir, 'model')
if not os.path.exists(model_dir):
os.makedirs(model_dir)
self.generator.save_weights(os.path.join(model_dir, 'generator_epoch_{}.hdf5'.format(epoch)))
'''if self.epochs > 20:
if epoch % 10 == 0:
# self.discriminator.save_weights(os.path.join(model_dir, 'discriminator_epoch_{}.hdf5'.format(epoch)))
else:
if epoch % 5 == 0:
self.generator.save_weights(os.path.join(model_dir, 'generator_epoch_{}.hdf5'.format(epoch)))
# self.discriminator.save_weights(os.path.join(model_dir, 'discriminator_epoch_{}.hdf5'.format(epoch)))'''
data.close()
if __name__ == '__main__':
gan = SR_UnetGAN()
gan.train()