-
Notifications
You must be signed in to change notification settings - Fork 0
/
CSAM.py
137 lines (121 loc) · 4.38 KB
/
CSAM.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
import torch
import torch.nn as nn
import torch.distributions as td
def custom_max(x, dim, keepdim=True):
temp_x = x
for i in dim:
temp_x = torch.max(temp_x, dim=i, keepdim=True)[0]
if not keepdim:
temp_x = temp_x.squeeze()
return temp_x
class PositionalAttentionModule(nn.Module):
def __init__(self):
super(PositionalAttentionModule, self).__init__()
self.conv = nn.Conv2d(
in_channels=2, out_channels=1, kernel_size=(7, 7), padding=3
)
def forward(self, x):
max_x = custom_max(x, dim=(0, 1), keepdim=True)
avg_x = torch.mean(x, dim=(0, 1), keepdim=True)
att = torch.cat((max_x, avg_x), dim=1)
att = self.conv(att)
att = torch.sigmoid(att)
return x * att
class SemanticAttentionModule(nn.Module):
def __init__(self, in_features, reduction_rate=16):
super(SemanticAttentionModule, self).__init__()
self.linear = []
self.linear.append(
nn.Linear(
in_features=in_features, out_features=in_features // reduction_rate
)
)
self.linear.append(nn.ReLU())
self.linear.append(
nn.Linear(
in_features=in_features // reduction_rate, out_features=in_features
)
)
self.linear = nn.Sequential(*self.linear)
def forward(self, x):
max_x = custom_max(x, dim=(0, 2, 3), keepdim=False).unsqueeze(0)
avg_x = torch.mean(x, dim=(0, 2, 3), keepdim=False).unsqueeze(0)
max_x = self.linear(max_x)
avg_x = self.linear(avg_x)
att = max_x + avg_x
att = torch.sigmoid(att).unsqueeze(-1).unsqueeze(-1)
return x * att
class SliceAttentionModule(nn.Module):
def __init__(self, in_features, rate=4, uncertainty=True, rank=5):
super(SliceAttentionModule, self).__init__()
self.uncertainty = uncertainty
self.rank = rank
self.linear = []
self.linear.append(
nn.Linear(in_features=in_features, out_features=int(in_features * rate))
)
self.linear.append(nn.ReLU())
self.linear.append(
nn.Linear(in_features=int(in_features * rate), out_features=in_features)
)
self.linear = nn.Sequential(*self.linear)
if uncertainty:
self.non_linear = nn.ReLU()
self.mean = nn.Linear(in_features=in_features, out_features=in_features)
self.log_diag = nn.Linear(in_features=in_features, out_features=in_features)
self.factor = nn.Linear(
in_features=in_features, out_features=in_features * rank
)
def forward(self, x):
max_x = custom_max(x, dim=(1, 2, 3), keepdim=False).unsqueeze(0)
avg_x = torch.mean(x, dim=(1, 2, 3), keepdim=False).unsqueeze(0)
max_x = self.linear(max_x)
avg_x = self.linear(avg_x)
att = max_x + avg_x
if self.uncertainty:
temp = self.non_linear(att)
mean = self.mean(temp)
diag = self.log_diag(temp).exp()
factor = self.factor(temp)
factor = factor.view(1, -1, self.rank)
dist = td.LowRankMultivariateNormal(
loc=mean, cov_factor=factor, cov_diag=diag
)
att = dist.sample()
att = torch.sigmoid(att).squeeze().unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)
return x * att
class CSAM(nn.Module):
def __init__(
self,
num_slices,
num_channels,
semantic=True,
positional=True,
slice=True,
uncertainty=True,
rank=5,
):
super(CSAM, self).__init__()
self.semantic = semantic
self.positional = positional
self.slice = slice
if semantic:
self.semantic_att = SemanticAttentionModule(num_channels)
if positional:
self.positional_att = PositionalAttentionModule()
if slice:
self.slice_att = SliceAttentionModule(
num_slices, uncertainty=uncertainty, rank=rank
)
def forward(self, x):
if self.semantic:
x = self.semantic_att(x)
if self.positional:
x = self.positional_att(x)
if self.slice:
x = self.slice_att(x)
return x
if __name__ == "__main__":
x = torch.randn(1, 1, 64, 64)
model = CSAM(num_slices=1, num_channels=1)
print(model(x).shape)