forked from ngessert/isic2019
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodels.py
250 lines (213 loc) · 10.5 KB
/
models.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
import torch
import numbers
import numpy as np
import functools
import h5py
import math
from torchvision import models
import pretrainedmodels
import torch.nn.functional as F
import types
import torch
from efficientnet_pytorch import EfficientNet
from collections import OrderedDict
import torch.nn as nn
def Dense121(config):
return models.densenet121(pretrained=True)
def Dense161(config):
return models.densenet169(pretrained=True)
def Dense169(config):
return models.densenet161(pretrained=True)
def Dense201(config):
return models.densenet201(pretrained=True)
def Resnet50(config):
return pretrainedmodels.__dict__['resnet50'](num_classes=1000, pretrained='imagenet')
def Resnet101(config):
return models.resnet101(pretrained=True)
def InceptionV3(config):
return models.inception_v3(pretrained=True)
def se_resnext50(config):
return pretrainedmodels.__dict__['se_resnext50_32x4d'](num_classes=1000, pretrained='imagenet')
def se_resnext101(config):
return pretrainedmodels.__dict__['se_resnext101_32x4d'](num_classes=1000, pretrained='imagenet')
def se_resnet50(config):
return pretrainedmodels.__dict__['se_resnet50'](num_classes=1000, pretrained='imagenet')
def se_resnet101(config):
return pretrainedmodels.__dict__['se_resnet101'](num_classes=1000, pretrained='imagenet')
def se_resnet152(config):
return pretrainedmodels.__dict__['se_resnet152'](num_classes=1000, pretrained='imagenet')
def resnext101(config):
return pretrainedmodels.__dict__['resnext101_32x4d'](num_classes=1000, pretrained='imagenet')
def resnext101_64(config):
return pretrainedmodels.__dict__['resnext101_64x4d'](num_classes=1000, pretrained='imagenet')
def senet154(config):
return pretrainedmodels.__dict__['senet154'](num_classes=1000, pretrained='imagenet')
def polynet(config):
return pretrainedmodels.__dict__['polynet'](num_classes=1000, pretrained='imagenet')
def dpn92(config):
return pretrainedmodels.__dict__['dpn92'](num_classes=1000, pretrained='imagenet+5k')
def dpn68b(config):
return pretrainedmodels.__dict__['dpn68b'](num_classes=1000, pretrained='imagenet+5k')
def nasnetamobile(config):
return pretrainedmodels.__dict__['nasnetamobile'](num_classes=1000, pretrained='imagenet')
def resnext101_32_8_wsl(config):
return torch.hub.load('facebookresearch/WSL-Images', 'resnext101_32x8d_wsl')
def resnext101_32_16_wsl(config):
return torch.hub.load('facebookresearch/WSL-Images', 'resnext101_32x16d_wsl')
def resnext101_32_32_wsl(config):
return torch.hub.load('facebookresearch/WSL-Images', 'resnext101_32x32d_wsl')
def resnext101_32_48_wsl(config):
return torch.hub.load('facebookresearch/WSL-Images', 'resnext101_32x48d_wsl')
def efficientnet_b0(config):
return EfficientNet.from_pretrained('efficientnet-b0',num_classes=config['numClasses'])
def efficientnet_b1(config):
return EfficientNet.from_pretrained('efficientnet-b1',num_classes=config['numClasses'])
def efficientnet_b2(config):
return EfficientNet.from_pretrained('efficientnet-b2',num_classes=config['numClasses'])
def efficientnet_b3(config):
return EfficientNet.from_pretrained('efficientnet-b3',num_classes=config['numClasses'])
def efficientnet_b4(config):
return EfficientNet.from_pretrained('efficientnet-b4',num_classes=config['numClasses'])
def efficientnet_b5(config):
return EfficientNet.from_pretrained('efficientnet-b5',num_classes=config['numClasses'])
def efficientnet_b6(config):
return EfficientNet.from_pretrained('efficientnet-b6',num_classes=config['numClasses'])
def efficientnet_b7(config):
return EfficientNet.from_pretrained('efficientnet-b7',num_classes=config['numClasses'])
def modify_meta(mdlParams,model):
# Define FC layers
if len(mdlParams['fc_layers_before']) > 1:
model.meta_before = nn.Sequential(nn.Linear(mdlParams['meta_array'].shape[1],mdlParams['fc_layers_before'][0]),
nn.BatchNorm1d(mdlParams['fc_layers_before'][0]),
nn.ReLU(),
nn.Dropout(p=mdlParams['dropout_meta']),
nn.Linear(mdlParams['fc_layers_before'][0],mdlParams['fc_layers_before'][1]),
nn.BatchNorm1d(mdlParams['fc_layers_before'][1]),
nn.ReLU(),
nn.Dropout(p=mdlParams['dropout_meta']))
else:
model.meta_before = nn.Sequential(nn.Linear(mdlParams['meta_array'].shape[1],mdlParams['fc_layers_before'][0]),
nn.BatchNorm1d(mdlParams['fc_layers_before'][0]),
nn.ReLU(),
nn.Dropout(p=mdlParams['dropout_meta']))
# Define fc layers after
if len(mdlParams['fc_layers_after']) > 0:
if 'efficient' in mdlParams['model_type']:
num_cnn_features = model._fc.in_features
elif 'wsl' in mdlParams['model_type']:
num_cnn_features = model.fc.in_features
else:
num_cnn_features = model.last_linear.in_features
model.meta_after = nn.Sequential(nn.Linear(mdlParams['fc_layers_before'][-1]+num_cnn_features,mdlParams['fc_layers_after'][0]),
nn.BatchNorm1d(mdlParams['fc_layers_after'][0]),
nn.ReLU())
classifier_in_features = mdlParams['fc_layers_after'][0]
else:
model.meta_after = None
classifier_in_features = mdlParams['fc_layers_before'][-1]+model._fc.in_features
# Modify classifier
if 'efficient' in mdlParams['model_type']:
model._fc = nn.Linear(classifier_in_features, mdlParams['numClasses'])
elif 'wsl' in mdlParams['model_type']:
model.fc = nn.Linear(classifier_in_features, mdlParams['numClasses'])
else:
model.last_linear = nn.Linear(classifier_in_features, mdlParams['numClasses'])
# Modify forward pass
def new_forward(self, inputs):
x, meta_data = inputs
# Normal CNN features
if 'efficient' in mdlParams['model_type']:
# Convolution layers
cnn_features = self.extract_features(x)
# Pooling and final linear layer
cnn_features = F.adaptive_avg_pool2d(cnn_features, 1).squeeze(-1).squeeze(-1)
if self._dropout:
cnn_features = F.dropout(cnn_features, p=self._dropout, training=self.training)
elif 'wsl' in mdlParams['model_type']:
cnn_features = self.conv1(x)
cnn_features = self.bn1(cnn_features)
cnn_features = self.relu(cnn_features)
cnn_features = self.maxpool(cnn_features)
cnn_features = self.layer1(cnn_features)
cnn_features = self.layer2(cnn_features)
cnn_features = self.layer3(cnn_features)
cnn_features = self.layer4(cnn_features)
cnn_features = self.avgpool(cnn_features)
cnn_features = torch.flatten(cnn_features, 1)
else:
cnn_features = self.layer0(x)
cnn_features = self.layer1(cnn_features)
cnn_features = self.layer2(cnn_features)
cnn_features = self.layer3(cnn_features)
cnn_features = self.layer4(cnn_features)
cnn_features = self.avg_pool(cnn_features)
if self.dropout is not None:
cnn_features = self.dropout(cnn_features)
cnn_features = cnn_features.view(cnn_features.size(0), -1)
# Meta part
#print(meta_data.shape,meta_data)
meta_features = self.meta_before(meta_data)
# Cat
features = torch.cat((cnn_features,meta_features),dim=1)
#print("features cat",features.shape)
if self.meta_after is not None:
features = self.meta_after(features)
# Classifier
if 'efficient' in mdlParams['model_type']:
output = self._fc(features)
elif 'wsl' in mdlParams['model_type']:
output = self.fc(features)
else:
output = self.last_linear(features)
return output
model.forward = types.MethodType(new_forward, model)
return model
model_map = OrderedDict([('Dense121', Dense121),
('Dense169' , Dense161),
('Dense161' , Dense169),
('Dense201' , Dense201),
('Resnet50' , Resnet50),
('Resnet101' , Resnet101),
('InceptionV3', InceptionV3),# models.inception_v3(pretrained=True),
('se_resnext50', se_resnext50),
('se_resnext101', se_resnext101),
('se_resnet50', se_resnet50),
('se_resnet101', se_resnet101),
('se_resnet152', se_resnet152),
('resnext101', resnext101),
('resnext101_64', resnext101_64),
('senet154', senet154),
('polynet', polynet),
('dpn92', dpn92),
('dpn68b', dpn68b),
('nasnetamobile', nasnetamobile),
('resnext101_32_8_wsl', resnext101_32_8_wsl),
('resnext101_32_16_wsl', resnext101_32_16_wsl),
('resnext101_32_32_wsl', resnext101_32_32_wsl),
('resnext101_32_48_wsl', resnext101_32_48_wsl),
('efficientnet-b0', efficientnet_b0),
('efficientnet-b1', efficientnet_b1),
('efficientnet-b2', efficientnet_b2),
('efficientnet-b3', efficientnet_b3),
('efficientnet-b4', efficientnet_b4),
('efficientnet-b5', efficientnet_b5),
('efficientnet-b6', efficientnet_b6),
('efficientnet-b7', efficientnet_b7),
])
def getModel(config):
"""Returns a function for a model
Args:
config: dictionary, contains configuration
Returns:
model: A class that builds the desired model
Raises:
ValueError: If model name is not recognized.
"""
if config['model_type'] in model_map:
func = model_map[config['model_type'] ]
@functools.wraps(func)
def model():
return func(config)
else:
raise ValueError('Name of model unknown %s' % config['model_name'] )
return model