-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathmodel_covid.py
232 lines (177 loc) · 9.29 KB
/
model_covid.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
import torch.nn as nn
from torchvision import models
import torch.nn.functional as F
class Flatten(nn.Module):
def forward(self, input):
return input.view(input.size(0), -1)
class PEPX(nn.Module):
def __init__(self, n_input, n_out):
super(PEPX, self).__init__()
self.network = nn.Sequential(nn.Conv2d(in_channels=n_input, out_channels=n_input // 2, kernel_size=1),
nn.Conv2d(in_channels=n_input // 2, out_channels=int(3 * n_input / 4),
kernel_size=1),
nn.Conv2d(in_channels=int(3 * n_input / 4), out_channels=int(3 * n_input / 4),
kernel_size=3, groups=int(3 * n_input / 4), padding=1),
nn.Conv2d(in_channels=int(3 * n_input / 4), out_channels=n_input // 2,
kernel_size=1),
nn.Conv2d(in_channels=n_input // 2, out_channels=n_out, kernel_size=1))
def forward(self, mapping_filters):
return self.network(mapping_filters)
class CovidNet(nn.Module):
# Inputs an image and ouputs the prediction for the class
def __init__(self, n_classes):
super(CovidNet, self).__init__()
filters = {
'pexp1_1': [64, 256],
'pexp1_2': [256, 256],
'pexp1_3': [256, 256],
'pexp2_1': [256, 512],
'pexp2_2': [512, 512],
'pexp2_3': [512, 512],
'pexp2_4': [512, 512],
'pexp3_1': [512, 1024],
'pexp3_2': [1024, 1024],
'pexp3_3': [1024, 1024],
'pexp3_4': [1024, 1024],
'pexp3_5': [1024, 1024],
'pexp3_6': [1024, 1024],
'pexp4_1': [1024, 2048],
'pexp4_2': [2048, 2048],
'pexp4_3': [2048, 2048],
}
# CONV 7X7
self.add_module('conv1_7x7', nn.Conv2d(in_channels=3, out_channels=64, kernel_size=7, stride=2, padding=3))
# add all PEPEX
for name, sizes in filters.items():
self.add_module(name, PEPX(sizes[0], sizes[1]))
# conv 1x1 top layers
self.add_module('conv1_1x1', nn.Conv2d(in_channels=64, out_channels=256, kernel_size=1))
self.add_module('conv2_1x1', nn.Conv2d(in_channels=256, out_channels=512, kernel_size=1))
self.add_module('conv3_1x1', nn.Conv2d(in_channels=512, out_channels=1024, kernel_size=1))
self.add_module('conv4_1x1', nn.Conv2d(in_channels=1024, out_channels=2048, kernel_size=1))
# Final block
self.add_module('flatten', Flatten())
self.add_module('fc1', nn.Linear(7 * 7 * 2048, 1024))
self.add_module('fc2', nn.Linear(1024, 256))
self.add_module('classifier', nn.Linear(256, n_classes))
# grad cam
self.gradients = None
def forward(self, img):
# CHUNK 0
out_conv7 = F.max_pool2d(F.relu(self.conv1_7x7(img)), 2)
# CHUNK 1
out_conv1_1x1 = self.conv1_1x1(out_conv7)
pepx11 = self.pexp1_1(out_conv7)
pepx12 = self.pexp1_2(pepx11 + out_conv1_1x1)
pepx13 = self.pexp1_3(pepx12 + pepx11 + out_conv1_1x1)
# CHUNK 2
out_conv2_1x1 = F.max_pool2d(self.conv2_1x1(pepx12 + pepx11 + pepx13 + out_conv1_1x1), 2)
pepx21 = self.pexp2_1(
F.max_pool2d(pepx13, 2) + F.max_pool2d(pepx11, 2) + F.max_pool2d(pepx12, 2) + F.max_pool2d(out_conv1_1x1,
2))
pepx22 = self.pexp2_2(pepx21 + out_conv2_1x1)
pepx23 = self.pexp2_3(pepx22 + pepx21 + out_conv2_1x1)
pepx24 = self.pexp2_4(pepx23 + pepx21 + pepx22 + out_conv2_1x1)
# CHUNK 3
out_conv3_1x1 = F.max_pool2d(self.conv3_1x1(pepx22 + pepx21 + pepx23 + pepx24 + out_conv2_1x1), 2)
pepx31 = self.pexp3_1(
F.max_pool2d(pepx24, 2) + F.max_pool2d(pepx21, 2) + F.max_pool2d(pepx22, 2) + F.max_pool2d(pepx23,
2) + F.max_pool2d(
out_conv2_1x1, 2))
pepx32 = self.pexp3_2(pepx31 + out_conv3_1x1)
pepx33 = self.pexp3_3(pepx31 + pepx32 + out_conv3_1x1)
pepx34 = self.pexp3_4(pepx31 + pepx32 + pepx33 + out_conv3_1x1)
pepx35 = self.pexp3_5(pepx31 + pepx32 + pepx33 + pepx34 + out_conv3_1x1)
pepx36 = self.pexp3_6(pepx31 + pepx32 + pepx33 + pepx34 + pepx35 + out_conv3_1x1)
# CHUNK 4
out_conv4_1x1 = F.max_pool2d(
self.conv4_1x1(pepx31 + pepx32 + pepx33 + pepx34 + pepx35 + pepx36 + out_conv3_1x1), 2)
pepx41 = self.pexp4_1(
F.max_pool2d(pepx31, 2) + F.max_pool2d(pepx32, 2) + F.max_pool2d(pepx32, 2) + F.max_pool2d(pepx34,
2) + F.max_pool2d(
pepx35, 2) + F.max_pool2d(pepx36, 2) + F.max_pool2d(out_conv3_1x1, 2))
pepx42 = self.pexp4_2(pepx41 + out_conv4_1x1)
pepx43 = self.pexp4_3(pepx41 + pepx42 + out_conv4_1x1)
# FINAL CHUNK
# for grad cam
activations = pepx41 + pepx42 + pepx43 + out_conv4_1x1
# h = activations.register_hook(self.activations_hook)
# flattened = self.flatten(pepx41 + pepx42 + pepx43 + out_conv4_1x1)
flattened = self.flatten(activations)
fc1out = F.relu(self.fc1(flattened))
fc2out = F.relu(self.fc2(fc1out))
logits = self.classifier(fc2out)
# probs = torch.softmax(logits, dim=1)
# winners = probs.argmax(dim=1)
return logits
def get_n_params(self):
"""
:return: number of parameters of this model
"""
pp = 0
for p in list(self.parameters()):
nn = 1
for s in list(p.size()):
nn = nn * s
pp += nn
return pp
# grad cam from this tutorial: https://medium.com/@stepanulyanin/implementing-grad-cam-in-pytorch-ea0937c31e82
# hook for the gradients of the activations
def activations_hook(self, grad):
self.gradients = grad
# method for the gradient extraction
def get_activations_gradient(self):
return self.gradients
# method to get activations
def get_activations(self, img):
# CHUNK 0
out_conv7 = F.max_pool2d(F.relu(self.conv1_7x7(img)), 2)
# CHUNK 1
out_conv1_1x1 = self.conv1_1x1(out_conv7)
pepx11 = self.pexp1_1(out_conv7)
pepx12 = self.pexp1_2(pepx11 + out_conv1_1x1)
pepx13 = self.pexp1_3(pepx12 + pepx11 + out_conv1_1x1)
# CHUNK 2
out_conv2_1x1 = F.max_pool2d(self.conv2_1x1(pepx12 + pepx11 + pepx13 + out_conv1_1x1), 2)
pepx21 = self.pexp2_1(
F.max_pool2d(pepx13, 2) + F.max_pool2d(pepx11, 2) + F.max_pool2d(pepx12, 2) + F.max_pool2d(out_conv1_1x1,
2))
pepx22 = self.pexp2_2(pepx21 + out_conv2_1x1)
pepx23 = self.pexp2_3(pepx22 + pepx21 + out_conv2_1x1)
pepx24 = self.pexp2_4(pepx23 + pepx21 + pepx22 + out_conv2_1x1)
# CHUNK 3
out_conv3_1x1 = F.max_pool2d(self.conv3_1x1(pepx22 + pepx21 + pepx23 + pepx24 + out_conv2_1x1), 2)
pepx31 = self.pexp3_1(
F.max_pool2d(pepx24, 2) + F.max_pool2d(pepx21, 2) + F.max_pool2d(pepx22, 2) + F.max_pool2d(pepx23,
2) + F.max_pool2d(
out_conv2_1x1, 2))
pepx32 = self.pexp3_2(pepx31 + out_conv3_1x1)
pepx33 = self.pexp3_3(pepx31 + pepx32 + out_conv3_1x1)
pepx34 = self.pexp3_4(pepx31 + pepx32 + pepx33 + out_conv3_1x1)
pepx35 = self.pexp3_5(pepx31 + pepx32 + pepx33 + pepx34 + out_conv3_1x1)
pepx36 = self.pexp3_6(pepx31 + pepx32 + pepx33 + pepx34 + pepx35 + out_conv3_1x1)
# CHUNK 4
out_conv4_1x1 = F.max_pool2d(
self.conv4_1x1(pepx31 + pepx32 + pepx33 + pepx34 + pepx35 + pepx36 + out_conv3_1x1), 2)
pepx41 = self.pexp4_1(
F.max_pool2d(pepx31, 2) + F.max_pool2d(pepx32, 2) + F.max_pool2d(pepx32, 2) + F.max_pool2d(pepx34,
2) + F.max_pool2d(
pepx35, 2) + F.max_pool2d(pepx36, 2) + F.max_pool2d(out_conv3_1x1, 2))
pepx42 = self.pexp4_2(pepx41 + out_conv4_1x1)
pepx43 = self.pexp4_3(pepx41 + pepx42 + out_conv4_1x1)
activations = pepx41 + pepx42 + pepx43 + out_conv4_1x1
return activations
class ResNet(nn.Module):
# Inputs an image and ouputs the prediction for the class and the projected embedding into the graph space
def __init__(self, num_class):
super(ResNet, self).__init__()
# Load pre-trained visual model
resnet = models.resnet50(pretrained=True)
self.resnet = nn.Sequential(*list(resnet.children())[:-1])
# Classifier
self.classifier = nn.Sequential(nn.Linear(2048, num_class))
def forward(self, img):
visual_emb = self.resnet(img)
visual_emb = visual_emb.view(visual_emb.size(0), -1)
logits = self.classifier(visual_emb)
return logits