-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathmodel.py
161 lines (132 loc) · 5.75 KB
/
model.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
import torch.nn as nn
import torchvision
import torch
# backbone
class PANnet(nn.Module):
def __init__(self):
super(PANnet, self).__init__()
self.model = torchvision.models.resnet18(pretrained=True)
self.conv1 = nn.Sequential(
nn.Conv2d(64, 128, kernel_size=1),
nn.BatchNorm2d(128),
nn.ReLU()
)
self.conv2 = nn.Sequential(
nn.Conv2d(128, 128, kernel_size=1),
nn.BatchNorm2d(128),
nn.ReLU()
)
self.conv3 = nn.Sequential(
nn.Conv2d(256, 128, kernel_size=1),
nn.BatchNorm2d(128),
nn.ReLU()
)
self.conv4 = nn.Sequential(
nn.Conv2d(512, 128, kernel_size=1),
nn.BatchNorm2d(128),
nn.ReLU()
)
self.upscale = nn.Sequential(
nn.Conv2d(128, 128, kernel_size=3, groups=128, padding=1),
nn.Conv2d(128, 128, kernel_size=1),
nn.BatchNorm2d(128),
nn.ReLU()
)
self.upsample = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False)
self.upsample1 = nn.Upsample(scale_factor=4, mode='bilinear', align_corners=False)
self.upsample2 = nn.Upsample(scale_factor=8, mode='bilinear', align_corners=False)
self.downscale = nn.Sequential(
nn.Conv2d(128, 128, kernel_size=3, groups=128, stride=2, padding=1),
nn.Conv2d(128, 128, kernel_size=1),
nn.BatchNorm2d(128),
nn.ReLU()
)
self.output = nn.Sequential(
nn.Conv2d(512, 6, kernel_size=1),
nn.Sigmoid()
)
self._initialize_weights()
def _initialize_weights(self):
for m in self.modules():
print(self.modules)
if isinstance(m, nn.Conv2d):
nn.init.xavier_normal_(m.weight)
if m.bias is not None:
nn.init.constant_(m.bias, 0.0)
def get_input(self, input):
output = self.model.conv1(input)
output = self.model.bn1(output)
output = self.model.relu(output)
output = self.model.maxpool(output)
input_layer1 = self.model.layer1(output)
input_layer2 = self.model.layer2(input_layer1)
input_layer3 = self.model.layer3(input_layer2)
input_layer4 = self.model.layer4(input_layer3)
input_layer1 = self.conv1(input_layer1)
input_layer2 = self.conv2(input_layer2)
input_layer3 = self.conv3(input_layer3)
input_layer4 = self.conv4(input_layer4)
return [input_layer1, input_layer2, input_layer3, input_layer4]
def up_scale_enhancement(self, layers):
layer1, layer2, layer3, layer4 = layers
upscale_layer4 = layer4
upscale_layer3 = self.upsample(upscale_layer4)
upscale_layer3 = torch.add(upscale_layer3, layer3)
upscale_layer3 = self.upscale(upscale_layer3)
upscale_layer2 = self.upsample(upscale_layer3)
upscale_layer2 = torch.add(upscale_layer2, layer2)
upscale_layer2 = self.upscale(upscale_layer2)
upscale_layer1 = self.upsample(upscale_layer2)
upscale_layer1 = torch.add(upscale_layer1, layer1)
upscale_layer1 = self.upscale(upscale_layer1)
return [upscale_layer1, upscale_layer2, upscale_layer3, upscale_layer4]
def down_scale_enhancement(self, layers):
layer1, layer2, layer3, layer4 = layers
output_layer1 = layer1
output_layer2 = self.upsample(layer2)
output_layer2 = torch.add(output_layer1, output_layer2)
output_layer2 = self.downscale(output_layer2)
output_layer3 = self.upsample(layer3)
output_layer3 = torch.add(output_layer2, output_layer3)
output_layer3 = self.downscale(output_layer3)
output_layer4 = self.upsample(layer4)
output_layer4 = torch.add(output_layer3, output_layer4)
output_layer4 = self.downscale(output_layer4)
return [output_layer1, output_layer2, output_layer3, output_layer4]
def feature_fusion_module(self, input_layers, output_layers):
input_layer1, input_layer2, input_layer3, input_layer4 = input_layers
output_layer1, output_layer2, output_layer3, output_layer4 = output_layers
layer1 = torch.add(input_layer1, output_layer1)
layer2 = torch.add(input_layer2, output_layer2)
layer3 = torch.add(input_layer3, output_layer3)
layer4 = torch.add(input_layer4, output_layer4)
layer2 = self.upsample(layer2)
layer3 = self.upsample1(layer3)
layer4 = self.upsample2(layer4)
final_feature_map = torch.cat([layer1, layer2, layer3, layer4], 1)
return final_feature_map
def output_text_region(self, feature):
return self.output(feature)
# return text_region
def output_kernel(self, feature):
return self.output(feature)
def output_similarity_vector(self, feature):
return self.output(feature)
def forward(self, input):
input_layers = self.get_input(input)
layers = self.up_scale_enhancement(input_layers)
output_layers = self.down_scale_enhancement(layers)
final_feature_map = self.feature_fusion_module(input_layers, output_layers)
final_feature_map = self.upsample1(final_feature_map)
output = self.output(final_feature_map)
output_text_region = output[:,0,:,:].unsqueeze(1)
output_kernel = output[:, 1, :, :].unsqueeze(1)
output_similarity_vector = output[:, 2:, :, :]
return output_text_region, output_kernel, output_similarity_vector
if __name__ == '__main__':
import cv2
from torchsummary import summary
model = PANnet()
input = torch.randn(size=(1, 3, 224, 224), dtype=torch.float32)
# print(model)
summary(model, (3, 224, 224), device='cpu')