-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_repository.py
236 lines (200 loc) · 11.1 KB
/
model_repository.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
import tensorflow as tf
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Conv2DTranspose, concatenate, BatchNormalization
import sys
from tensorflow.keras.metrics import Recall, Precision
from utils.custom_metrics import mean_iou
from utils.constract_loss_function import ConstructLossFunction
from utils.construct_optimizer import ConstructOptimizer
class ModelRepository:
def __init__(self,
model_name,
dim,
input_channels,
batch_size,
srcnn_count=0,
optimizer="SGD",
l_rate=0.001,
decay=1e-6,
momentum=0.9,
nesterov=True,
loss="dice"):
'''
Collection of deep learning models for image segmentation.
:param model_name:name of the model that'll run.
:param dim: 2D dimension of the input image
:param input_channels: channel count of the input image
:param batch_size: size of a batch
:param srcnn_count: number of srcnn layers - first srcnn_count images are applied, later not applied or ignored.
:param optimizer: Optimizer
:param l_rate: Learning rate.
:param decay: decay parameter for SGD optimizer
:param momentum: momentum parameter for SGD optimizer
:param nesterov: nesterov parameter for SGD optimizer
:param loss: loss function parameter
'''
self.model_name = model_name
self.dim = dim
self.input_channels = input_channels
self.batch_size = batch_size
self.srcnn_count = srcnn_count
self.optimizer = optimizer
self.l_rate = l_rate
self.decay = decay
self.momentum = momentum
self.nesterov = nesterov
self.loss = loss
self.loss_function = ConstructLossFunction(loss_function_name=self.loss).get_loss_function()
self.optimizer = ConstructOptimizer(optimizer_name=self.optimizer, l_rate = self.l_rate, decay = self.decay,
momentum = self.momentum, nesterov=self.nesterov).get_optimizer()
# model to run
if self.model_name == "test_model":
self.test_model()
elif self.model_name == "unet":
self.unet(self.dim, self.input_channels, self.batch_size)
elif self.model_name == "srcnn_unet":
self.srcnn_unet(self.dim, self.input_channels, self.batch_size, self.srcnn_count)
else:
print(self.model_name + " not defined yet.")
sys.exit()
def get_model(self):
return self.model
def test_model(self):
self.model = tf.keras.Sequential()
# model.add(tf.keras.layers.Flatten())
self.model.add(tf.keras.layers.Dense(100, activation='relu'))
self.model.add(tf.keras.layers.Dense(100, activation='relu'))
self.model.add(tf.keras.layers.Dense(100, activation='relu'))
self.model.add(tf.keras.layers.Dense(10000, activation='softmax'))
self.model.compile(optimizer=self.optimizer,
loss=self.loss_function,
metrics=['accuracy'])
def unet(self, dim, input_channels, batch_size):
inputs_layer = tf.keras.layers.Input((dim[0], dim[1], input_channels), batch_size=batch_size)
conv1 = Conv2D(64, (3, 3), activation="relu", padding="same")(inputs_layer)
conv1 = BatchNormalization()(conv1)
conv1 = Conv2D(64, (3, 3), activation="relu", padding="same")(conv1)
conv1 = BatchNormalization()(conv1)
pool1 = MaxPooling2D((2, 2))(conv1)
conv2 = Conv2D(128, (3, 3), activation="relu", padding="same")(pool1)
conv2 = BatchNormalization()(conv2)
conv2 = Conv2D(128, (3, 3), activation="relu", padding="same")(conv2)
conv2 = BatchNormalization()(conv2)
pool2 = MaxPooling2D((2, 2))(conv2)
conv3 = Conv2D(256, (3, 3), activation="relu", padding="same")(pool2)
conv3 = BatchNormalization()(conv3)
conv3 = Conv2D(256, (3, 3), activation="relu", padding="same")(conv3)
conv3 = BatchNormalization()(conv3)
pool3 = MaxPooling2D((2, 2))(conv3)
conv4 = Conv2D(512, (3, 3), activation="relu", padding="same")(pool3)
conv4 = BatchNormalization()(conv4)
conv4 = Conv2D(512, (3, 3), activation="relu", padding="same")(conv4)
conv4 = BatchNormalization()(conv4)
pool4 = MaxPooling2D((2, 2))(conv4)
conv_middle = Conv2D(1024, (3, 3), activation="relu", padding="same")(pool4)
conv_middle = BatchNormalization()(conv_middle)
conv_middle = Conv2D(1024, (3, 3), activation="relu", padding="same")(conv_middle)
conv_middle = BatchNormalization()(conv_middle)
conv_t4 = Conv2DTranspose(512, (2, 2), strides=(2, 2), padding="same")(conv_middle)
conc4 = concatenate([conv_t4, conv4])
up_conv4 = Conv2D(512, (3, 3), activation="relu", padding="same")(conc4)
up_conv4 = BatchNormalization()(up_conv4)
up_conv4 = Conv2D(512, (3, 3), activation="relu", padding="same")(up_conv4)
up_conv4 = BatchNormalization()(up_conv4)
conv_t3 = Conv2DTranspose(256, (2, 2), strides=(2, 2), padding="same")(up_conv4)
conc3 = concatenate([conv_t3, conv3])
up_conv3 = Conv2D(256, (3, 3), activation="relu", padding="same")(conc3)
up_conv3 = BatchNormalization()(up_conv3)
up_conv3 = Conv2D(256, (3, 3), activation="relu", padding="same")(up_conv3)
up_conv3 = BatchNormalization()(up_conv3)
conv_t2 = Conv2DTranspose(128, (2, 2), strides=(2, 2), padding="same")(up_conv3)
conc2 = concatenate([conv_t2, conv2])
up_conv2 = Conv2D(128, (3, 3), activation="relu", padding="same")(conc2)
up_conv2 = BatchNormalization()(up_conv2)
up_conv2 = Conv2D(128, (3, 3), activation="relu", padding="same")(up_conv2)
up_conv2 = BatchNormalization()(up_conv2)
conv_t1 = Conv2DTranspose(64, (2, 2), strides=(2, 2), padding="same")(up_conv2)
conc1 = concatenate([conv_t1, conv1])
up_conv1 = Conv2D(64, (3, 3), activation="relu", padding="same")(conc1)
up_conv1 = BatchNormalization()(up_conv1)
up_conv1 = Conv2D(64, (3, 3), activation="relu", padding="same")(up_conv1)
up_conv1 = BatchNormalization()(up_conv1)
output_layer = Conv2D(2, (1, 1), padding="same", activation="sigmoid")(up_conv1)
self.model = tf.keras.Model(inputs=inputs_layer, outputs=output_layer)
self.model.compile(optimizer=self.optimizer,
loss=self.loss_function,
metrics=["accuracy",
Precision(),
Recall(),
mean_iou])
def srcnn_unet(self, dim, input_channels, batch_size, srcnn_count):
input_layers = []
input_conc_layers = []
for i in range(0, srcnn_count):
temp_input_layer = tf.keras.layers.Input((dim[0], dim[1], 1), batch_size=batch_size)
input_layers.append(temp_input_layer)
srcnn = Conv2D(64, (9, 9), activation='relu', padding="same")(temp_input_layer)
srcnn = Conv2D(32, (1, 1), activation='relu', padding="same")(srcnn)
srcnn = Conv2D(1, (5, 5), activation='relu', padding="same")(srcnn)
input_conc_layers.append(srcnn)
if input_channels - srcnn_count > 0:
non_srcnn_layers = tf.keras.layers.Input((dim[0], dim[1], 3), batch_size=batch_size)
input_layers.append(non_srcnn_layers)
input_conc_layers.append(non_srcnn_layers)
main_inputs = concatenate(input_conc_layers)
# TODO: replace code with Unet method to remove code duplication
conv1 = Conv2D(64, (3, 3), activation="relu", padding="same")(main_inputs)
conv1 = BatchNormalization()(conv1)
conv1 = Conv2D(64, (3, 3), activation="relu", padding="same")(conv1)
conv1 = BatchNormalization()(conv1)
pool1 = MaxPooling2D((2, 2))(conv1)
conv2 = Conv2D(128, (3, 3), activation="relu", padding="same")(pool1)
conv2 = BatchNormalization()(conv2)
conv2 = Conv2D(128, (3, 3), activation="relu", padding="same")(conv2)
conv2 = BatchNormalization()(conv2)
pool2 = MaxPooling2D((2, 2))(conv2)
conv3 = Conv2D(256, (3, 3), activation="relu", padding="same")(pool2)
conv3 = BatchNormalization()(conv3)
conv3 = Conv2D(256, (3, 3), activation="relu", padding="same")(conv3)
conv3 = BatchNormalization()(conv3)
pool3 = MaxPooling2D((2, 2))(conv3)
conv4 = Conv2D(512, (3, 3), activation="relu", padding="same")(pool3)
conv4 = BatchNormalization()(conv4)
conv4 = Conv2D(512, (3, 3), activation="relu", padding="same")(conv4)
conv4 = BatchNormalization()(conv4)
pool4 = MaxPooling2D((2, 2))(conv4)
conv_middle = Conv2D(1024, (3, 3), activation="relu", padding="same")(pool4)
conv_middle = BatchNormalization()(conv_middle)
conv_middle = Conv2D(1024, (3, 3), activation="relu", padding="same")(conv_middle)
conv_middle = BatchNormalization()(conv_middle)
conv_t4 = Conv2DTranspose(512, (2, 2), strides=(2, 2), padding="same")(conv_middle)
conc4 = concatenate([conv_t4, conv4])
up_conv4 = Conv2D(512, (3, 3), activation="relu", padding="same")(conc4)
up_conv4 = BatchNormalization()(up_conv4)
up_conv4 = Conv2D(512, (3, 3), activation="relu", padding="same")(up_conv4)
up_conv4 = BatchNormalization()(up_conv4)
conv_t3 = Conv2DTranspose(256, (2, 2), strides=(2, 2), padding="same")(up_conv4)
conc3 = concatenate([conv_t3, conv3])
up_conv3 = Conv2D(256, (3, 3), activation="relu", padding="same")(conc3)
up_conv3 = BatchNormalization()(up_conv3)
up_conv3 = Conv2D(256, (3, 3), activation="relu", padding="same")(up_conv3)
up_conv3 = BatchNormalization()(up_conv3)
conv_t2 = Conv2DTranspose(128, (2, 2), strides=(2, 2), padding="same")(up_conv3)
conc2 = concatenate([conv_t2, conv2])
up_conv2 = Conv2D(128, (3, 3), activation="relu", padding="same")(conc2)
up_conv2 = BatchNormalization()(up_conv2)
up_conv2 = Conv2D(128, (3, 3), activation="relu", padding="same")(up_conv2)
up_conv2 = BatchNormalization()(up_conv2)
conv_t1 = Conv2DTranspose(64, (2, 2), strides=(2, 2), padding="same")(up_conv2)
conc1 = concatenate([conv_t1, conv1])
up_conv1 = Conv2D(64, (3, 3), activation="relu", padding="same")(conc1)
up_conv1 = BatchNormalization()(up_conv1)
up_conv1 = Conv2D(64, (3, 3), activation="relu", padding="same")(up_conv1)
up_conv1 = BatchNormalization()(up_conv1)
output_layer = Conv2D(2, (1, 1), padding="same", activation="sigmoid")(up_conv1)
self.model = tf.keras.Model(inputs=input_layers, outputs=output_layer)
self.model.compile(optimizer=self.optimizer,
loss=self.loss_function,
metrics=["accuracy",
Precision(),
Recall(),
mean_iou])