-
Notifications
You must be signed in to change notification settings - Fork 7
/
visual_model_selector.py
268 lines (239 loc) · 10.6 KB
/
visual_model_selector.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
264
265
266
from __future__ import absolute_import, division
import importlib
from tensorflow.keras.layers import Input
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Model
from dense_classifier import get_classifier
from downscaling_cnn import get_downscaling_model
class ModelFactory:
"""
Model facotry for Keras default models
"""
def __init__(self):
self.models_ = dict(
VGG16=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.vgg16",
last_conv_layer="block5_conv3",
),
VGG19=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.vgg19",
last_conv_layer="block5_conv4",
),
DenseNet121=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.densenet",
last_conv_layer="bn",
),
DenseNet169=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.densenet",
last_conv_layer="bn",
),
DenseNet201=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.densenet",
last_conv_layer="bn",
),
Xception=dict(
input_shape=(299, 299, 3),
module_name="tensorflow.keras.applications.xception",
last_conv_layer="block14_sepconv2_bn",
),
ResNet50=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.resnet",
last_conv_layer="activation_49",
),
ResNet50V2=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.resnet_v2",
last_conv_layer="activation_49",
),
ResNet101=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.resnet",
last_conv_layer="activation_49",
),
ResNet101V2=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.resnet_v2",
last_conv_layer="activation_49",
),
ResNet152=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.resnet",
last_conv_layer="activation_49",
),
ResNet152V2=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.resnet_v2",
last_conv_layer="activation_49",
),
InceptionV3=dict(
input_shape=(299, 299, 3),
module_name="tensorflow.keras.applications.inception_v3",
last_conv_layer="mixed10",
),
InceptionResNetV2=dict(
input_shape=(299, 299, 3),
module_name="tensorflow.keras.applications.inception_resnet_v2",
last_conv_layer="conv_7b_ac",
),
NASNetMobile=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.nasnet",
last_conv_layer="activation_188",
),
NASNetLarge=dict(
input_shape=(331, 331, 3),
module_name="tensorflow.keras.applications.nasnet",
last_conv_layer="activation_260",
),
MobileNet=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.mobilenet",
last_conv_layer="conv_pw_13_bn",
),
MobileNetV2=dict(
input_shape=(224, 224, 3),
module_name="tensorflow.keras.applications.mobilenet_v2",
last_conv_layer="Conv_1_bn",
),
EfficientNetB0=dict(
input_shape=(224, 224, 3),
module_name="efficientnet.tfkeras",
last_conv_layer="Conv_1_bn",
),
EfficientNetB1=dict(
input_shape=(224, 224, 3),
module_name="efficientnet.tfkeras",
last_conv_layer="Conv_1_bn",
),
EfficientNetB2=dict(
input_shape=(224, 224, 3),
module_name="efficientnet.tfkeras",
last_conv_layer="Conv_1_bn",
),
EfficientNetB3=dict(
input_shape=(224, 224, 3),
module_name="efficientnet.tfkeras",
last_conv_layer="Conv_1_bn",
),
EfficientNetB4=dict(
input_shape=(224, 224, 3),
module_name="efficientnet.tfkeras",
last_conv_layer="Conv_1_bn",
),
EfficientNetB5=dict(
input_shape=(224, 224, 3),
module_name="efficientnet.tfkeras",
last_conv_layer="Conv_1_bn",
),
EfficientNetB6=dict(
input_shape=(224, 224, 3),
module_name="efficientnet.tfkeras",
last_conv_layer="Conv_1_bn",
),
EfficientNetB7=dict(
input_shape=(224, 224, 3),
module_name="efficientnet.tfkeras",
last_conv_layer="Conv_1_bn",
),
)
def get_last_conv_layer(self, model_name):
return self.models_[model_name]["last_conv_layer"]
def get_input_size(self, model_name):
return self.models_[model_name]["input_shape"][:2]
def load_chexnet_weights(self, base_model, img_input, weights_path):
predictions = Dense(14, activation="sigmoid", name="predictions")(base_model.output)
base_model = Model(inputs=img_input, outputs=predictions)
base_model.load_weights(weights_path)
print(f"loaded chexnet weights: {weights_path}")
return base_model
def pop_conv_layers(self, base_model, img_input, layers_to_pop):
for i in range(layers_to_pop):
base_model._layers.pop()
base_model.outputs = [base_model.layers[-1].output]
base_model = Model(inputs=img_input, outputs=base_model.outputs, name='Visual_Model')
return base_model
def set_trainable_layers(self, base_model, layers_to_train):
for i in range(len(base_model.layers) - layers_to_train):
base_model.layers[i].trainable = False
return base_model
def get_output_unrolled_size(self, output_layer_shape):
output_unrolled_length = 1
for dimension in output_layer_shape[1:]:
output_unrolled_length *= int(dimension)
return output_unrolled_length
def concat_models(self, downscaling_model, visual_model, classifier, img_input, base_model_img_input):
base_model_output = visual_model.layers[-1].output
if downscaling_model is not None:
downscaled_images = downscaling_model(img_input)
visual_features = visual_model(downscaled_images)
if classifier is not None:
predictions = classifier(visual_features)
loaded_model = Model(inputs=img_input, outputs=predictions, name='Pipeline')
else:
loaded_model = Model(inputs=img_input, outputs=visual_features, name='Pipeline')
else:
if classifier is not None:
predictions = classifier(base_model_output)
loaded_model = Model(inputs=base_model_img_input, outputs=predictions, name='Pipeline')
else:
loaded_model = visual_model
return loaded_model
def get_model(self, FLAGS):
if 'Efficient' in FLAGS.visual_model_name:
base_weights = "noisy-student"
elif FLAGS.use_imagenet_weights is True:
base_weights = "imagenet"
else:
base_weights = None
base_model_class = getattr(
importlib.import_module(
"{}".format(self.models_[FLAGS.visual_model_name]['module_name'])
),
FLAGS.visual_model_name)
input_shape = FLAGS.image_target_size
if input_shape is None:
input_shape = self.models_[FLAGS.visual_model_name]["input_shape"]
img_input = Input(shape=input_shape, name="image_batch")
downscaling_model = None
if FLAGS.cnn_downscaling_factor > 0:
downscaling_model = get_downscaling_model(input_shape, FLAGS.cnn_downscaling_factor,
FLAGS.cnn_downscaling_filters)
input_shape = (int(input_shape[0] / (FLAGS.cnn_downscaling_factor * 2)),
int(input_shape[1] / (FLAGS.cnn_downscaling_factor * 2)), 3)
base_model_img_input = Input(shape=input_shape)
base_model = base_model_class(
include_top=False,
input_tensor=base_model_img_input,
input_shape=input_shape,
weights=base_weights,
pooling=FLAGS.final_layer_pooling)
chexnet_classifier_exists = False
if FLAGS.use_chexnet_weights and FLAGS.visual_model_name == 'DenseNet121':
base_model = self.load_chexnet_weights(base_model, base_model_img_input, FLAGS.chexnet_weights_path)
chexnet_classifier_exists = True
if FLAGS.pop_conv_layers > 0:
base_model = self.pop_conv_layers(base_model, base_model_img_input, FLAGS.pop_conv_layers)
chexnet_classifier_exists = False
if FLAGS.conv_layers_to_train != -1:
base_model = self.set_trainable_layers(base_model, FLAGS.conv_layers_to_train)
classifier = None
if FLAGS.classes is not None and FLAGS.classes != [] and not chexnet_classifier_exists:
base_model_output = base_model.layers[-1].output
output_unrolled_length = self.get_output_unrolled_size(base_model_output.shape)
classifier = get_classifier(base_model_output.shape, FLAGS.multi_label_classification,
FLAGS.classifier_layer_sizes, len(FLAGS.classes))
loaded_model = self.concat_models(downscaling_model, base_model, classifier, img_input, base_model_img_input)
if FLAGS.show_model_summary:
loaded_model.summary()
if downscaling_model is not None:
downscaling_model.summary()
base_model.summary()
if classifier is not None:
classifier.summary()
return loaded_model