-
-
Notifications
You must be signed in to change notification settings - Fork 2
/
CNN.py
180 lines (144 loc) · 7.64 KB
/
CNN.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
import keras
import tensorflow as tf
from keras import backend as K
from keras.backend.tensorflow_backend import set_session
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.8
config.gpu_options.allow_growth = True
set_session(tf.Session(config=config))
K.clear_session()
import numpy as np
import h5py
import random
from scipy.stats import norm
from keras.models import Model, load_model
from keras import regularizers, Input, optimizers, layers
from keras.layers import Activation
from keras.callbacks import History, ModelCheckpoint
import os
from glob import glob
nb_epochs = 300
steps_train = 100
steps_valid = 50
size_block = 16 # (16*16*16)
initial_LR = 0.01
Decay = 0.8 #Decay applied every 10epochs
adim_8 = 342.553
if not os.path.exists('save_model'):
os.makedirs('save_model')
def pad_fields(passive, target, adim, size_block_x, size_block_yz):
padded_passive = np.pad(passive, ((0, size_block_x), (size_block_yz // 2, size_block_yz // 2), (size_block_yz // 2, size_block_yz // 2)), mode="edge")
#periodicity y
padded_passive[:, :size_block_yz//2] = padded_passive[:, -2*(size_block_yz//2) : -(size_block_yz//2)]
padded_passive[:, -(size_block_yz//2):] = padded_passive[:, (size_block_yz//2) : 2*(size_block_yz//2)]
#periodicity z
padded_passive[:, :, :size_block_yz//2] = padded_passive[:, :, -2*(size_block_yz//2) : -(size_block_yz//2)]
padded_passive[:, :, -(size_block_yz//2):] = padded_passive[:, :, (size_block_yz//2) : 2*(size_block_yz//2)]
target = target / adim
padded_target = np.pad(target, ((0, size_block_x), (size_block_yz // 2, size_block_yz // 2), (size_block_yz // 2, size_block_yz // 2)), mode="edge")
#periodicity y
padded_target[:, :size_block_yz//2] = padded_target[:, -2*(size_block_yz//2) : -(size_block_yz//2)]
padded_target[:, -(size_block_yz//2):] = padded_target[:, (size_block_yz//2) : 2*(size_block_yz//2)]
#periodicity z
padded_target[:, :, :size_block_yz//2] = padded_target[:, :, -2*(size_block_yz//2) : -(size_block_yz//2)]
padded_target[:, :, -(size_block_yz//2):] = padded_target[:, :, (size_block_yz//2) : 2*(size_block_yz//2)]
return(padded_passive, padded_target)
def generator_3D(begin_distrib, end_distrib, size_num, rotate, flip): #valid_or_train= 1 for validation, 2 for train
while 1:
crops_per_block = 10
batch_size = size_num*crops_per_block
liste_input = np.zeros((batch_size, size_block, size_block, size_block, 1))
liste_target = np.zeros((batch_size, size_block, size_block, size_block, 1))
for num in range (size_num):
num_DNS = np.random.randint(1, 3)#DNS 1 or 2
liste_name = sorted(glob("DATA/DNS" + str(num_DNS) + "*.h5"))
choosen_num = np.random.randint(begin_distrib, end_distrib)
my_file = h5py.File(liste_name[choosen_num], 'r')
passive = my_file['filt_8'].value
target = my_file['filt_grad_8'].value
padded_passive, padded_target = pad_fields(passive, target, adim_8, size_block, size_block)
for k in range (crops_per_block):
abs_x = np.random.randint(0, 64-size_block+1)
abs_y = np.random.randint(0, 32)#with padding
abs_z = np.random.randint(0, 32)
liste_input[num*crops_per_block + k, :, :, :, 0]=padded_passive[abs_x:abs_x+size_block, abs_y:abs_y+size_block, abs_z:abs_z+size_block]
liste_target[num*crops_per_block + k, :, :, :, 0]=padded_target[abs_x:abs_x+size_block, abs_y:abs_y+size_block, abs_z:abs_z+size_block]
if rotate:
rot_times = np.random.randint(4, size=batch_size)
rot_ax = np.random.randint(3, size=batch_size)
ax_to_plane = {0: (1, 2), 1: (0, 2), 2: (0, 1)}
for i, (ti, ax) in enumerate(zip(rot_times, rot_ax)):
liste_input[i,] = np.rot90(liste_input[i,], k=ti, axes=ax_to_plane[ax])
liste_target[i,] = np.rot90(liste_target[i,], k=ti, axes=ax_to_plane[ax])
if flip:
for i, flip_by in enumerate(np.random.randint(6, size=batch_size)):
if flip_by <= 2:
liste_input[i] = np.flip(liste_input[i], axis=flip_by)
liste_target[i] = np.flip(liste_target[i], axis=flip_by)
indice = np.arange(liste_input.shape[0])
np.random.shuffle(indice)
liste_input = liste_input[indice]
liste_target = liste_target[indice]
yield liste_input, liste_target
def CNN():
num_channels = 1
num_mask_channels = 1
img_shape = (None, None, None, 1)
inputs = Input(shape = img_shape)
conv1 = layers.Conv3D(32, 3, padding='same')(inputs)
conv1 = layers.BatchNormalization()(conv1)
conv1 = Activation('relu')(conv1)
conv1 = layers.Conv3D(32, 3, padding='same')(conv1)
conv1 = layers.BatchNormalization()(conv1)
conv1 = Activation('relu')(conv1)
pool1 = layers.MaxPooling3D(pool_size=(2, 2, 2))(conv1)
conv2 = layers.Conv3D(64, 3, padding='same')(pool1)
conv2 = layers.BatchNormalization()(conv2)
conv2 = Activation('relu')(conv2)
conv2 = layers.Conv3D(64, 3, padding='same')(conv2)
conv2 = layers.BatchNormalization()(conv2)
conv2 = Activation('relu')(conv2)
pool2 = layers.MaxPooling3D(pool_size=(2, 2, 2))(conv2)
conv3 = layers.Conv3D(128, 3, padding='same')(pool2)
conv3 = layers.BatchNormalization()(conv3)
conv3 = Activation('relu')(conv3)
conv3 = layers.Conv3D(128, 3, padding='same')(conv3)
conv3 = layers.BatchNormalization()(conv3)
conv3 = Activation('relu')(conv3)
conv3 = layers.UpSampling3D(size=(2, 2, 2))(conv3)
up4 = layers.concatenate([conv3, conv2])
conv4 = layers.Conv3DTranspose(64, 3, padding='same')(up4)
conv4 = layers.BatchNormalization()(conv4)
conv4 = Activation('relu')(conv4)
conv4 = layers.Conv3DTranspose(64, 3, padding='same')(conv4)
conv4 = layers.BatchNormalization()(conv4)##conv ou crop
conv4 = Activation('relu')(conv4)
conv4 = layers.Conv3DTranspose(64, 1, padding='same')(conv4)
conv4 = layers.UpSampling3D(size=(2, 2, 2))(conv4)
up5 = layers.concatenate([conv4, conv1])
conv5 = layers.Conv3DTranspose(32, 3, padding='same')(up5)
conv5 = layers.BatchNormalization()(conv5)
conv5 = Activation('relu')(conv5)
conv5 = layers.Conv3DTranspose(32, 3, padding='same')(conv5)
conv5 = layers.BatchNormalization()(conv5)##conv ou crop
conv5 = Activation('relu')(conv5)
conv5 = layers.Conv3DTranspose(1, 1, padding='same', activation='relu')(conv5)
model = Model(inputs=inputs, outputs=conv5)
#model.summary()
return(model)
def scheduler(epoch):
initial_lrate = K.get_value(model.optimizer.lr)
if epoch%10==9 :
lrate = initial_lrate * Decay
return lrate
return initial_lrate
if __name__ == "__main__":
train_generator = generator_3D(1, 41, 4, True, True)#first 40 fields for training cf paper
valid_generator = generator_3D(41, 51, 1, False, False)#Last 10 for validation
model=CNN()
new_Adam = optimizers.Adam(lr=initial_LR, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)
model.compile(optimizer=new_Adam, loss='mse', metrics = ['mae'])
lrate = keras.callbacks.LearningRateScheduler(scheduler, verbose=1)
filepath="save_model/unet_{epoch:03d}_with_loss_{val_loss:.4f}.h5"
checkpoint = ModelCheckpoint(filepath, monitor='val_loss', verbose=1, save_best_only=True)
history = model.fit_generator(train_generator, steps_per_epoch=steps_train, epochs=nb_epochs, validation_data=valid_generator, callbacks=[lrate, checkpoint], validation_steps = steps_valid)