-
Notifications
You must be signed in to change notification settings - Fork 1
/
DSS.py
127 lines (104 loc) · 4.6 KB
/
DSS.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
import torch
import torch.nn as nn
import torchvision.models as models
from tensorboardX import SummaryWriter
from torch_deform_conv.layers import ConvOffset2D
#from torch.utils.tensorboard import SummaryWriter
vgg_model_1 = models.vgg16(pretrained=True).features[:-1]
# vgg_model_1[-2] = ConvOffset2D(filters=512)
# vgg_model_1[-4] = ConvOffset2D(filters=512)
# vgg_model_1[-6] = ConvOffset2D(filters=512)
vgg_model_2 = models.vgg16(pretrained=True).features[:16]
# vgg_model_2[-2] = ConvOffset2D(filters=256)
def conv_relu(in_dim, out_dim):
return nn.Sequential(
nn.Conv2d(in_dim, out_dim, kernel_size=3, padding=1),nn.ReLU(inplace=True),)
class DSS(nn.Module):
# pretrained_model = \
# osp.expanduser('~/data/models/pytorch/fcn8s_from_caffe.pth')
#
# @classmethod
# def download(cls):
# return fcn.data.cached_download(
# url='http://drive.google.com/uc?id=0B9P1L--7Wd2vT0FtdThWREhjNkU',
# path=cls.pretrained_model,
# md5='dbd9bbb3829a3184913bccc74373afbb',
# )
def __init__(self):
super(DSS, self).__init__()
####path1
self.vgg1 = vgg_model_1
self.upscore1_6 = nn.ConvTranspose2d(512, 512, 2, stride=2, bias=False)
self.conv1_6 = conv_relu(512,512) ###########concat
self.upscore1_7 = nn.ConvTranspose2d(1024, 512, 2, stride=2, bias=False)
self.conv1_7 = conv_relu(512,512)#######concat
self.conv1_8_1 = conv_relu(768, 256)
self.conv1_8_2 = conv_relu(256, 256)
self.conv1_8_3 = conv_relu(256, 256)
self.upscore1_9 = nn.ConvTranspose2d(256, 256, 2, stride=2, bias=False)
self.conv1_9_1 = conv_relu(256, 128)
self.conv1_9_2 = conv_relu(128, 128)
self.upscore1_10 = nn.ConvTranspose2d(128, 128, 2, stride=2, bias=False)
self.conv1_10_1 = conv_relu(128, 64)
self.conv1_10_2 = conv_relu(64, 64)
self.pre_phase_1=nn.Sequential(nn.Conv2d(64, 1, kernel_size=1, padding=0),nn.Sigmoid(),)
####path2
self.vgg2 = vgg_model_2
self.upscore2_4 = nn.ConvTranspose2d(256, 256, 2, stride=2, bias=False)
self.conv2_4 = conv_relu(256, 256)
self.upscore2_5 = nn.ConvTranspose2d(384, 256, 2, stride=2, bias=False)
self.conv2_5 = conv_relu(256, 256)
self.conv2_6_1 = conv_relu(320, 64)
self.conv2_6_2 = conv_relu(64, 64)
self.pre_phase_2 = nn.Sequential(nn.Conv2d(64, 1, kernel_size=1, padding=0), nn.Sigmoid(), )
####Final prediction phase
self.pre_fina = nn.Sequential(nn.Conv2d(1, 1, kernel_size=1, padding=0), nn.Sigmoid(), )
def forward(self, x):
####path_1
feature_1 = self.vgg1(x)
feature_1_16 = self.vgg1[:16](x)
feature_1_23 = self.vgg1[:23](x)
feature_1=self.upscore1_6(feature_1)
feature_1=self.conv1_6(feature_1)
feature_1=torch.cat((feature_1,feature_1_23),1)
feature_1=self.upscore1_7(feature_1)
feature_1=self.conv1_7(feature_1)
feature_1=torch.cat((feature_1,feature_1_16),1)
feature_1=self.conv1_8_1(feature_1)
feature_1 = self.conv1_8_2(feature_1)
feature_1 = self.conv1_8_3(feature_1)
feature_1 = self.upscore1_9(feature_1)
feature_1 = self.conv1_9_1(feature_1)
feature_1 = self.conv1_9_2(feature_1)
feature_1 = self.upscore1_10(feature_1)
feature_1 = self.conv1_10_1(feature_1)
feature_1 = self.conv1_10_2(feature_1)
out_1=self.pre_phase_1(feature_1)
####path_2
feature_2 = self.vgg2(x)
feature_2_4=self.vgg2[:4](x)
feature_2_9=self.vgg2[:9](x)
feature_2=self.upscore2_4(feature_2)
feature_2 = self.conv2_4(feature_2)
feature_2 = torch.cat((feature_2, feature_2_9), 1)
feature_2 = self.upscore2_5(feature_2)
feature_2 = self.conv2_5(feature_2)
feature_2 = torch.cat((feature_2, feature_2_4), 1)
feature_2 = self.conv2_6_1(feature_2)
feature_2 = self.conv2_6_2(feature_2)
out_2=self.pre_phase_2(feature_2)
out=out_1+out_2
out=self.pre_fina(out)
return out
if __name__ == "__main__":
# device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
#image_size = 64
x = torch.rand(1, 3, 256,256)
# x.to(device)
print("x size: {}".format(x.size()))
model =DSS()
with SummaryWriter(comment='DSSNet') as w:
w.add_graph(model, (x,))
out = model(x)
print(model)
print("out size: {}".format(out.size()))