-
Notifications
You must be signed in to change notification settings - Fork 3
/
net.py
167 lines (143 loc) · 5.8 KB
/
net.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
import torch
from torch import nn
import torch.nn.functional as F
import vgg
def convblock(in_,out_,ks,st,pad):
return nn.Sequential(
nn.Conv2d(in_,out_,ks,st,pad),
nn.BatchNorm2d(out_),
nn.ReLU(inplace=True)
)
class GFB(nn.Module):
def __init__(self,in_1,in_2):
super(GFB, self).__init__()
self.ca1 = CA(2*in_1)
self.conv1 = convblock(2*in_1,128, 3, 1, 1)
self.conv_globalinfo = convblock(512,128,3, 1, 1)
self.ca2 = CA(in_2)
self.conv_curfeat =convblock(in_2,128,3,1,1)
self.conv_out= convblock(128,in_2,3,1,1)
def forward(self, pre1,pre2,cur,global_info):
cur_size = cur.size()[2:]
pre = self.ca1(torch.cat((pre1,pre2),1))
pre =self.conv1(F.interpolate(pre,cur_size,mode='bilinear',align_corners=True))
global_info = self.conv_globalinfo(F.interpolate(global_info,cur_size,mode='bilinear',align_corners=True))
cur_feat =self.conv_curfeat(self.ca2(cur))
fus = pre + cur_feat + global_info
return self.conv_out(fus)
class GlobalInfo(nn.Module):
def __init__(self):
super(GlobalInfo, self).__init__()
self.ca = CA(1024)
self.de_chan = convblock(1024,256,3,1,1)
self.b0 = nn.Sequential(
nn.AdaptiveMaxPool2d(13),
nn.Conv2d(256,128,1,1,0,bias=False),
nn.ReLU(inplace=True)
)
self.b1 = nn.Sequential(
nn.AdaptiveMaxPool2d(9),
nn.Conv2d(256,128,1,1,0,bias=False),
nn.ReLU(inplace=True)
)
self.b2 = nn.Sequential(
nn.AdaptiveMaxPool2d(5),
nn.Conv2d(256, 128, 1, 1, 0,bias=False),
nn.ReLU(inplace=True)
)
self.b3 = nn.Sequential(
nn.AdaptiveMaxPool2d(1),
nn.Conv2d(256, 128, 1, 1, 0,bias=False),
nn.ReLU(inplace=True)
)
self.fus = convblock(768,512,1,1,0)
def forward(self, rgb,t):
x_size=rgb.size()[2:]
x=self.ca(torch.cat((rgb,t),1))
x=self.de_chan(x)
b0 = F.interpolate(self.b0(x),x_size,mode='bilinear',align_corners=True)
b1 = F.interpolate(self.b1(x),x_size,mode='bilinear',align_corners=True)
b2 = F.interpolate(self.b2(x),x_size,mode='bilinear',align_corners=True)
b3 = F.interpolate(self.b3(x),x_size,mode='bilinear',align_corners=True)
out = self.fus(torch.cat((b0,b1,b2,b3,x),1))
return out
class CA(nn.Module):
def __init__(self,in_ch):
super(CA, self).__init__()
self.avg_weight = nn.AdaptiveAvgPool2d(1)
self.max_weight = nn.AdaptiveMaxPool2d(1)
self.fus = nn.Sequential(
nn.Conv2d(in_ch, in_ch // 2, 1, 1, 0),
nn.ReLU(),
nn.Conv2d(in_ch // 2, in_ch, 1, 1, 0),
)
self.c_mask = nn.Sigmoid()
def forward(self, x):
avg_map_c = self.avg_weight(x)
max_map_c = self.max_weight(x)
c_mask = self.c_mask(torch.add(self.fus(avg_map_c), self.fus(max_map_c)))
return torch.mul(x, c_mask)
class FinalScore(nn.Module):
def __init__(self):
super(FinalScore, self).__init__()
self.ca =CA(256)
self.score = nn.Conv2d(256, 1, 1, 1, 0)
def forward(self,f1,f2,xsize):
f1 = torch.cat((f1,f2),1)
f1 = self.ca(f1)
score = F.interpolate(self.score(f1), xsize, mode='bilinear', align_corners=True)
return score
class Decoder(nn.Module):
def __init__(self):
super(Decoder, self).__init__()
self.global_info =GlobalInfo()
self.score_global = nn.Conv2d(512, 1, 1, 1, 0)
self.gfb4_1 = GFB(512,512)
self.gfb3_1= GFB(512,256)
self.gfb2_1= GFB(256,128)
self.gfb4_2 = GFB(512, 512) #1/8
self.gfb3_2 = GFB(512, 256)#1/4
self.gfb2_2 = GFB(256, 128)#1/2
self.score_1=nn.Conv2d(128, 1, 1, 1, 0)
self.score_2 = nn.Conv2d(128, 1, 1, 1, 0)
self.refine =FinalScore()
def forward(self,rgb,t):
xsize=rgb[0].size()[2:]
global_info =self.global_info(rgb[4],t[4]) # 512 1/16
d1=self.gfb4_1(global_info,t[4],rgb[3],global_info)
d2=self.gfb4_2(global_info, rgb[4], t[3], global_info)
#print(d1.shape,d2.shape)
d3= self.gfb3_1(d1, d2,rgb[2],global_info)
d4 = self.gfb3_2(d2, d1, t[2], global_info)
d5 = self.gfb2_1(d3, d4, rgb[1], global_info)
d6 = self.gfb2_2(d4, d3, t[1], global_info) #1/2 128
score_global = self.score_global(global_info)
score1=self.score_1(F.interpolate(d5,xsize,mode='bilinear',align_corners=True))
score2 = self.score_2(F.interpolate(d6, xsize, mode='bilinear', align_corners=True))
score =self.refine(d5,d6,xsize)
return score,score1,score2,score_global
class Mnet(nn.Module):
def __init__(self):
super(Mnet,self).__init__()
self.rgb_net= vgg.a_vgg16()
self.t_net= vgg.a_vgg16()
self.decoder=Decoder()
for m in self.modules():
if isinstance(m, nn.Conv2d):
m.weight.data.normal_(0, 0.01)
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
def forward(self,rgb,t):
rgb_f= self.rgb_net(rgb)
t_f= self.t_net(t)
score,score1,score2,score_g =self.decoder(rgb_f,t_f)
return score,score1,score2,score_g
def load_pretrained_model(self):
st=torch.load("vgg16.pth")
st2={}
for key in st.keys():
st2['base.'+key]=st[key]
self.rgb_net.load_state_dict(st2)
self.t_net.load_state_dict(st2)
print('loading pretrained model success!')