-
Notifications
You must be signed in to change notification settings - Fork 24
/
Copy pathtrain_ensembler.py
176 lines (128 loc) · 4.69 KB
/
train_ensembler.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
import numpy as np
import tensorflow as tf
from models import *
from losses import *
class_weights = np.load('class_weights.npy')
# Optimizers
ensembler_optimizer = tf.keras.optimizers.Adam(2e-4, beta_1=0.5)
imShape = (128, 128, 128, 40)
gtShape = (128, 128, 128, 4)
E = ensembler(imShape, gtShape, class_weights, kernel_size=3)
# Load the Vox2Vox models
gen1 = Generator()
gen1.load_weights('./RESULTS/Generator1.h5')
print('Vox2Vox generator 1 loaded.')
gen2 = Generator()
gen2.load_weights('./RESULTS/Generator2.h5')
print('Vox2Vox generator 2 loaded.')
gen3 = Generator()
gen3.load_weights('./RESULTS/Generator3.h5')
print('Vox2Vox generator 3 loaded.')
gen4 = Generator()
gen4.load_weights('./RESULTS/Generator4.h5')
print('Vox2Vox generator 4 loaded.')
gen5 = Generator()
gen5.load_weights('./RESULTS/Generator5.h5')
print('Vox2Vox generator 5 loaded.')
gen6 = Generator()
gen6.load_weights('./RESULTS/Generator6.h5')
print('Vox2Vox generator 6 loaded.')
gen7 = Generator()
gen7.load_weights('./RESULTS/Generator7.h5')
print('Vox2Vox generator 7 loaded.')
gen8 = Generator()
gen8.load_weights('./RESULTS/Generator8.h5')
print('Vox2Vox generator 8 loaded.')
gen9 = Generator()
gen9.load_weights('./RESULTS/Generator9.h5')
print('Vox2Vox generator 9 loaded.')
gen10 = Generator()
gen10.load_weights('./RESULTS/Generator10.h5')
print('Vox2Vox generator 10 loaded.')
# ## Training
def get_scores(X, y):
SCORES = np.empty((X.shape[0], 128, 128, 128, 40))
y_pred_1 = gen1.predict(X)
SCORES[:,:,:,:,0:4] = y_pred_1
del(y_pred_1)
y_pred_2 = gen2.predict(X)
SCORES[:,:,:,:,4:8] = y_pred_2
del(y_pred_2)
y_pred_3 = gen3.predict(X)
SCORES[:,:,:,:,8:12] = y_pred_3
del(y_pred_3)
y_pred_4 = gen4.predict(X)
SCORES[:,:,:,:,12:16] = y_pred_4
del(y_pred_4)
y_pred_5 = gen5.predict(X)
SCORES[:,:,:,:,16:20] = y_pred_5
del(y_pred_5)
y_pred_6 = gen6.predict(X)
SCORES[:,:,:,:,20:24] = y_pred_6
del(y_pred_6)
y_pred_7 = gen7.predict(X)
SCORES[:,:,:,:,24:28] = y_pred_7
del(y_pred_7)
y_pred_8 = gen8.predict(X)
SCORES[:,:,:,:,28:32] = y_pred_8
del(y_pred_8)
y_pred_9 = gen9.predict(X)
SCORES[:,:,:,:,32:36] = y_pred_9
del(y_pred_9)
y_pred_10 = gen10.predict(X)
SCORES[:,:,:,:,36:40] = y_pred_10
del(X, y_pred_1, y_pred_2, y_pred_3, y_pred_4, y_pred_5, y_pred_6, y_pred_7, y_pred_8, y_pred_9, y_pred_10)
# pre-proc zero-center
SCORES -= 0.5
return SCORES
@tf.function
def train_step(image, target):
with tf.GradientTape() as ens_tape:
scores = get_scores(image, target)
ens_output = E(scores, training=True)
dice_loss = diceLoss(target, ens_output, class_weights)
ensembler_gradients = ens_tape.gradient(dice_loss, E.trainable_variables)
ensembler_optimizer.apply_gradients(zip(ensembler_gradients, E.trainable_variables))
return dice_loss
@tf.function
def test_step(image, target):
scores = get_scores(image, target)
ens_output = E(scores, training=False)
dice_loss = diceLoss(target, ens_output, class_weights)
return dice_loss
def fit(train_gen, valid_gen, epochs):
path = './RESULTS'
if os.path.exists(path)==False:
os.mkdir(path)
Nt = len(train_gen)
prev_loss = np.inf
epoch_dice_loss = tf.keras.metrics.Mean()
epoch_dice_loss_val = tf.keras.metrics.Mean()
for e in range(epochs):
print('Epoch {}/{}'.format(e+1,epochs))
b = 0
for Xb, yb in train_gen:
b += 1
loss = train_step(Xb, yb)
epoch_dice_loss.update_state(loss)
stdout.write('\rBatch: {}/{} - dice_loss: {:.4f}'.format(b, Nt, epoch_dice_loss.result()))
stdout.flush()
for Xb, yb in valid_gen:
loss_val = test_step(Xb, yb)
epoch_dice_loss_val.update_state(loss_val)
stdout.write('\n dice_loss_val: {:.4f}'.format(epoch_dice_loss_val.result()))
stdout.flush()
# save models
print(' ')
if epoch_dice_loss_val.result() < prev_loss:
E.save_weights(path + '/Ensembler.h5')
print("Validation loss decresaed from {:.4f} to {:.4f}. Models' weights are now saved."
.format(prev_loss, epoch_dice_loss_val.result()))
prev_loss = epoch_dice_loss_val.result()
else:
print("Validation loss did not decrese from {:.4f}.".format(prev_loss))
print(' ')
# reset losses state
epoch_dice_loss.reset_states()
epoch_dice_loss_val.reset_states()
del Xb, yb