-
Notifications
You must be signed in to change notification settings - Fork 3
/
Train_network.py
214 lines (159 loc) · 7.32 KB
/
Train_network.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
import numpy as np
import cv2
import os
import itertools
from keras.models import Sequential, Model
from keras.layers import Dense, Input, Conv2D, MaxPooling2D, concatenate, Conv2DTranspose, UpSampling2D, Flatten
from keras import backend as K
from keras.optimizers import Adam
from PIL import Image
model = Sequential()
train_images="C:\\Users\\Nexus\\Desktop\\Diplom\\train_full"
train_images_full="C:\\Users\\Nexus\\Desktop\\Diplom\\train_full"
smooth = 1.
def dice_coef(y_true, y_pred):
y_true_f = K.flatten(y_true)
y_pred_f = K.flatten(y_pred)
intersection = K.sum(y_true_f * y_pred_f)
return (2. * intersection + smooth) / (K.sum(y_true_f) + K.sum(y_pred_f) + smooth)
def dice_coef_loss(y_true, y_pred):
return -dice_coef(y_true, y_pred)
def load_image(path):
return cv2.imread(path, cv2.IMREAD_GRAYSCALE)
def feature_matrix_con(img):
return (img / 255.0)
def feature_matrix_full(img):
window = (5, 5)
nbrs = [cv2.getRectSubPix(img, window, (y, x)).ravel()
for x, y in itertools.product(range(np.shape(img)[0]), range(np.shape(img)[1]))]
median5 = cv2.medianBlur(img, 5).ravel()
median25 = cv2.medianBlur(img, 25).ravel()
grad = np.abs(cv2.Sobel(img, cv2.CV_16S, 1, 1, ksize=3).ravel())
div = np.abs(cv2.Sobel(img, cv2.CV_16S, 2, 2, ksize=3).ravel())
misc = np.vstack((median5, median25, grad, div)).transpose()
features = np.hstack((np.asarray(nbrs), misc))
return (features / 255.0).astype('float32')
def target_matrix_con(img):
return (img / 255.0)
def target_matrix_full(img):
return (img / 255.0).astype('float32').ravel()[:, None]
def load_train_set_con(file_list):
xs = []
ys = []
for fname in file_list:
xs.append(feature_matrix_con(load_image(os.path.join('C:\\Users\\Nexus\\Desktop\\Diplom\\train', fname))))
ys.append(target_matrix_con(load_image(os.path.join('C:\\Users\\Nexus\\Desktop\\Diplom\\train_cleaned', fname))))
return np.vstack(xs), np.vstack(ys)
def load_train_set_full(file_list):
xs = []
ys = []
for fname in file_list:
xs.append(feature_matrix_full(load_image(os.path.join('C:\\Users\\Nexus\\Desktop\\Diplom\\train_full', fname))))
ys.append(target_matrix_full(load_image(os.path.join('C:\\Users\\Nexus\\Desktop\\Diplom\\train_cleaned', fname))))
return np.vstack(xs), np.vstack(ys)
def blockshaped(arr, nrows, ncols):
h, w = arr.shape
return (arr.reshape(h//nrows, nrows, -1, ncols)
.swapaxes(1,2)
.reshape(-1, nrows, ncols))
def unet(img_channels, image_rows, image_cols):
inputs = Input((img_channels, image_rows, image_cols))
conv1 = Conv2D(32, (3, 3), activation='tanh', padding='same')(inputs)
conv1 = Conv2D(32, (3, 3), activation='tanh', padding='same')(conv1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = Conv2D(64, (3, 3), activation='tanh', padding='same')(pool1)
conv2 = Conv2D(64, (3, 3), activation='tanh', padding='same')(conv2)
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
conv3 = Conv2D(128, (3, 3), activation='tanh', padding='same')(pool2)
conv3 = Conv2D(128, (3, 3), activation='tanh', padding='same')(conv3)
up8 = concatenate([Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same')(conv3), conv2], axis=3)
conv8 = Conv2D(64, (3, 3), activation='tanh', padding='same')(up8)
conv8 = Conv2D(64, (3, 3), activation='tanh', padding='same')(conv8)
up9 = concatenate([Conv2DTranspose(32, (2, 2), strides=(2, 2), padding='same')(conv8), conv1], axis=3)
conv9 = Conv2D(32, (3, 3), activation='tanh', padding='same')(up9)
conv9 = Conv2D(32, (3, 3), activation='tanh', padding='same')(conv9)
conv10 = Conv2D(1, (1, 1), activation='sigmoid')(conv9)
model = Model(inputs=[inputs], outputs=[conv10])
model.compile(loss='mean_squared_error',
optimizer='adam',
metrics=['accuracy'])
return model
def train_con():
model = unet(112,112,1)
files = os.listdir(train_images)
for i in files:
print("0) ",i)
train_x, train_y = load_train_set_con([i])
train_x = cv2.resize(train_x, (448, 448))
train_y = cv2.resize(train_y, (448, 448))
train_x = blockshaped(train_x,112,112)
train_y = blockshaped(train_y,112,112)
train_x = np.expand_dims(train_x, axis=3)
train_y = np.expand_dims(train_y, axis=3)
model.fit(train_x, train_y, batch_size=16,epochs=20, validation_split=0.1, verbose=0)
model.save('weights_1.h5')
files = os.listdir(train_images)
for i in files:
print("90) ",i)
train_x, train_y = load_train_set_con([i])
train_x = np.rot90(train_x, 3)
train_y = np.rot90(train_y, 3)
train_x = cv2.resize(train_x, (448, 448))
train_y = cv2.resize(train_y, (448, 448))
train_x = blockshaped(train_x,112,112)
train_y = blockshaped(train_y,112,112)
train_x = np.expand_dims(train_x, axis=3)
train_y = np.expand_dims(train_y, axis=3)
model.fit(train_x, train_y, batch_size=16,epochs=20, validation_split=0.1, verbose=0)
model.save('weights_2.h5')
files = os.listdir(train_images)
for i in files:
print("180) ",i)
train_x, train_y = load_train_set_con([i])
train_x = np.rot90(train_x, 3)
train_x = np.rot90(train_x, 3)
train_y = np.rot90(train_y, 3)
train_y = np.rot90(train_y, 3)
train_x = cv2.resize(train_x, (448, 448))
train_y = cv2.resize(train_y, (448, 448))
train_x = blockshaped(train_x,112,112)
train_y = blockshaped(train_y,112,112)
train_x = np.expand_dims(train_x, axis=3)
train_y = np.expand_dims(train_y, axis=3)
model.fit(train_x, train_y, batch_size=16,epochs=20, validation_split=0.1, verbose=0)
model.save('weights_3.h5')
files = os.listdir(train_images)
for i in files:
print("270) ",i)
train_x, train_y = load_train_set_con([i])
train_x = np.rot90(train_x, 3)
train_x = np.rot90(train_x, 3)
train_x = np.rot90(train_x, 3)
train_y = np.rot90(train_y, 3)
train_y = np.rot90(train_y, 3)
train_y = np.rot90(train_y, 3)
train_x = cv2.resize(train_x, (448, 448))
train_y = cv2.resize(train_y, (448, 448))
train_x = blockshaped(train_x,112,112)
train_y = blockshaped(train_y,112,112)
train_x = np.expand_dims(train_x, axis=3)
train_y = np.expand_dims(train_y, axis=3)
model.fit(train_x, train_y, batch_size=16,epochs=20, validation_split=0.1, verbose=0)
model.save('weights_4.h5')
def train_image_full():
model.add(Dense(500, input_shape=(29,),activation = 'tanh'))
model.add(Dense(1, activation = 'tanh'))
model.compile(loss='mean_absolute_error',
optimizer='adam',
metrics=['accuracy'])
files = os.listdir(train_images_full)
for i in files:
train_x, train_y = load_train_set_full([i])
print(i)
model.fit(train_x, train_y, batch_size=16,epochs=20, validation_split=0.1, verbose=1)
model.save('weights_full.h5')
def main():
train_con()
train_image_full()
if __name__ == '__main__':
main()