-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
133 lines (115 loc) · 5.01 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
import paddle
from paddle.vision.models import vgg16
import math
from weight_init import weight_init_
class extractor(paddle.nn.Layer):
def __init__(self, pretrained):
super(extractor, self).__init__()
vgg16_bn = vgg16(pretrained=False, batch_norm=True)
self.features = vgg16_bn.features
def forward(self, x):
out = []
for m in self.features:
x = m(x)
if isinstance(m, paddle.nn.MaxPool2D):
out.append(x)
return out[1:]
class merge(paddle.nn.Layer):
def __init__(self):
super(merge, self).__init__()
self.conv1 = paddle.nn.Conv2D(in_channels=1024, out_channels=128,
kernel_size=1)
self.bn1 = paddle.nn.BatchNorm2D(num_features=128)
self.relu1 = paddle.nn.ReLU()
self.conv2 = paddle.nn.Conv2D(in_channels=128, out_channels=128,
kernel_size=3, padding=1)
self.bn2 = paddle.nn.BatchNorm2D(num_features=128)
self.relu2 = paddle.nn.ReLU()
self.conv3 = paddle.nn.Conv2D(in_channels=384, out_channels=64,
kernel_size=1)
self.bn3 = paddle.nn.BatchNorm2D(num_features=64)
self.relu3 = paddle.nn.ReLU()
self.conv4 = paddle.nn.Conv2D(in_channels=64, out_channels=64,
kernel_size=3, padding=1)
self.bn4 = paddle.nn.BatchNorm2D(num_features=64)
self.relu4 = paddle.nn.ReLU()
self.conv5 = paddle.nn.Conv2D(in_channels=192, out_channels=32,
kernel_size=1)
self.bn5 = paddle.nn.BatchNorm2D(num_features=32)
self.relu5 = paddle.nn.ReLU()
self.conv6 = paddle.nn.Conv2D(in_channels=32, out_channels=32,
kernel_size=3, padding=1)
self.bn6 = paddle.nn.BatchNorm2D(num_features=32)
self.relu6 = paddle.nn.ReLU()
self.conv7 = paddle.nn.Conv2D(in_channels=32, out_channels=32,
kernel_size=3, padding=1)
self.bn7 = paddle.nn.BatchNorm2D(num_features=32)
self.relu7 = paddle.nn.ReLU()
for m in self.sublayers():
if isinstance(m, paddle.nn.Conv2D):
weight_init_(m.weight, "kaiming_normal_", mode='fan_out', nonlinearity='relu')
if m.bias is not None:
init_Constant = paddle.nn.initializer.Constant(value=0)
init_Constant(m.bias)
elif isinstance(m, paddle.nn.BatchNorm2D):
init_Constant = paddle.nn.initializer.Constant(value=1)
init_Constant(m.weight)
init_Constant = paddle.nn.initializer.Constant(value=0)
init_Constant(m.bias)
def forward(self, x):
y = paddle.nn.functional.interpolate(x=x[3], scale_factor=2, mode=
'bilinear', align_corners=True)
y = paddle.concat(x=(y, x[2]), axis=1)
y = self.relu1(self.bn1(self.conv1(y)))
y = self.relu2(self.bn2(self.conv2(y)))
y = paddle.nn.functional.interpolate(x=y, scale_factor=2, mode=
'bilinear', align_corners=True)
y = paddle.concat(x=(y, x[1]), axis=1)
y = self.relu3(self.bn3(self.conv3(y)))
y = self.relu4(self.bn4(self.conv4(y)))
y = paddle.nn.functional.interpolate(x=y, scale_factor=2, mode=
'bilinear', align_corners=True)
y = paddle.concat(x=(y, x[0]), axis=1)
y = self.relu5(self.bn5(self.conv5(y)))
y = self.relu6(self.bn6(self.conv6(y)))
y = self.relu7(self.bn7(self.conv7(y)))
return y
class output(paddle.nn.Layer):
def __init__(self, scope=512):
super(output, self).__init__()
self.conv1 = paddle.nn.Conv2D(in_channels=32, out_channels=1,
kernel_size=1)
self.sigmoid1 = paddle.nn.Sigmoid()
self.conv2 = paddle.nn.Conv2D(in_channels=32, out_channels=4,
kernel_size=1)
self.sigmoid2 = paddle.nn.Sigmoid()
self.conv3 = paddle.nn.Conv2D(in_channels=32, out_channels=1,
kernel_size=1)
self.sigmoid3 = paddle.nn.Sigmoid()
self.scope = 512
for m in self.sublayers():
if isinstance(m, paddle.nn.Conv2D):
weight_init_(m.weight, "kaiming_normal_", mode='fan_out', nonlinearity='relu')
if m.bias is not None:
init_Constant = paddle.nn.initializer.Constant(value=0)
init_Constant(m.bias)
def forward(self, x):
score = self.sigmoid1(self.conv1(x))
loc = self.sigmoid2(self.conv2(x)) * self.scope
angle = (self.sigmoid3(self.conv3(x)) - 0.5) * math.pi
geo = paddle.concat(x=(loc, angle), axis=1)
return score, geo
class EAST(paddle.nn.Layer):
def __init__(self, pretrained=False):
super(EAST, self).__init__()
self.extractor = extractor(pretrained)
self.merge = merge()
self.output = output()
def forward(self, x):
return self.output(self.merge(self.extractor(x)))
if __name__ == '__main__':
m = EAST()
x = paddle.randn(shape=[1, 3, 256, 256])
score, geo = m(x)
print(score.shape)
print(geo.shape)