forked from kaijieshi7/Dynamic-convolution-Pytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdynamic_conv.py
228 lines (196 loc) · 8.4 KB
/
dynamic_conv.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
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
import torch
import torch.nn as nn
import torch.nn.functional as F
class attention2d(nn.Module):
def __init__(self, in_planes, ratios, K, temperature, init_weight=True):
super(attention2d, self).__init__()
assert temperature%3==1
self.avgpool = nn.AdaptiveAvgPool2d(1)
if in_planes!=3:
hidden_planes = int(in_planes*ratios)
else:
hidden_planes = K
self.fc1 = nn.Conv2d(in_planes, hidden_planes, 1, bias=False)
self.fc2 = nn.Conv2d(hidden_planes, K, 1, bias=False)
self.temperature = temperature
if init_weight:
self._initialize_weights()
def _initialize_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
if m.bias is not None:
nn.init.constant_(m.bias, 0)
def updata_temperature(self):
if self.temperature!=1:
self.temperature -=3
print('Change temperature to:', str(self.temperature))
def forward(self, x):
x = self.avgpool(x)
x = self.fc1(x)
x = F.relu(x)
x = self.fc2(x).view(x.size(0), -1)
return F.softmax(x/self.temperature, 1)
class Dynamic_conv2d(nn.Module):
def __init__(self, in_planes, out_planes, kernel_size, ratio=0.25, stride=1, padding=0, dilation=1, groups=1, bias=True, K=4,temperature=34, init_weight=True):
super(Dynamic_conv2d, self).__init__()
assert in_planes%groups==0
self.in_planes = in_planes
self.out_planes = out_planes
self.kernel_size = kernel_size
self.stride = stride
self.padding = padding
self.dilation = dilation
self.groups = groups
self.bias = bias
self.K = K
self.attention = attention2d(in_planes, ratio, K, temperature)
self.weight = nn.Parameter(torch.Tensor(K, out_planes, in_planes//groups, kernel_size, kernel_size), requires_grad=True)
if bias:
self.bias = nn.Parameter(torch.Tensor(K, out_planes))
else:
self.bias = None
if init_weight:
self._initialize_weights()
#TODO 初始化
def _initialize_weights(self):
for i in range(self.K):
nn.init.kaiming_uniform_(self.weight[i])
def update_temperature(self):
self.attention.updata_temperature()
def forward(self, x):#将batch视作维度变量，进行组卷积，因为组卷积的权重是不同的，动态卷积的权重也是不同的
softmax_attention = self.attention(x)
batch_size, in_planes, height, width = x.size()
x = x.view(1, -1, height, width)# 变化成一个维度进行组卷积
weight = self.weight.view(self.K, -1)
# 动态卷积的权重的生成， 生成的是batch_size个卷积参数（每个参数不同）
aggregate_weight = torch.mm(softmax_attention, weight).view(-1, self.in_planes, self.kernel_size, self.kernel_size)
if self.bias is not None:
aggregate_bias = torch.mm(softmax_attention, self.bias).view(-1)
output = F.conv2d(x, weight=aggregate_weight, bias=aggregate_bias, stride=self.stride, padding=self.padding,
dilation=self.dilation, groups=self.groups*batch_size)
else:
output = F.conv2d(x, weight=aggregate_weight, bias=None, stride=self.stride, padding=self.padding,
dilation=self.dilation, groups=self.groups * batch_size)
output = output.view(batch_size, self.out_planes, output.size(-2), output.size(-1))
return output
class attention3d(nn.Module):
def __init__(self, in_planes, ratios, K, temperature):
super(attention3d, self).__init__()
assert temperature%3==1
self.avgpool = nn.AdaptiveAvgPool3d(1)
if in_planes != 3:
hidden_planes = int(in_planes * ratios)
else:
hidden_planes = K
self.fc1 = nn.Conv3d(in_planes, hidden_planes, 1, bias=False)
self.fc2 = nn.Conv3d(hidden_planes, K, 1, bias=False)
self.temperature = temperature
def updata_temperature(self):
if self.temperature!=1:
self.temperature -=3
print('Change temperature to:', str(self.temperature))
def forward(self, x):
x = self.avgpool(x)
x = self.fc1(x)
x = F.relu(x)
x = self.fc2(x).view(x.size(0), -1)
return F.softmax(x / self.temperature, 1)
class Dynamic_conv3d(nn.Module):
def __init__(self, in_planes, out_planes, kernel_size, ratio=0.25, stride=1, padding=0, dilation=1, groups=1, bias=True, K=4, temperature=34):
super(Dynamic_conv3d, self).__init__()
assert in_planes%groups==0
self.in_planes = in_planes
self.out_planes = out_planes
self.kernel_size = kernel_size
self.stride = stride
self.padding = padding
self.dilation = dilation
self.groups = groups
self.bias = bias
self.K = K
self.attention = attention3d(in_planes, ratio, K, temperature)
self.weight = nn.Parameter(torch.Tensor(K, out_planes, in_planes//groups, kernel_size, kernel_size, kernel_size), requires_grad=True)
if bias:
self.bias = nn.Parameter(torch.Tensor(K, out_planes))
else:
self.bias = None
#TODO 初始化
# nn.init.kaiming_uniform_(self.weight, )
def update_temperature(self):
self.attention.updata_temperature()
def forward(self, x):#将batch视作维度变量，进行组卷积，因为组卷积的权重是不同的，动态卷积的权重也是不同的
softmax_attention = self.attention(x)
batch_size, in_planes, depth, height, width = x.size()
x = x.view(1, -1, depth, height, width)# 变化成一个维度进行组卷积
weight = self.weight.view(self.K, -1)
# 动态卷积的权重的生成， 生成的是batch_size个卷积参数（每个参数不同）
aggregate_weight = torch.mm(softmax_attention, weight).view(-1, self.in_planes, self.kernel_size, self.kernel_size, self.kernel_size)
if self.bias is not None:
aggregate_bias = torch.mm(softmax_attention, self.bias).view(-1)
output = F.conv3d(x, weight=aggregate_weight, bias=aggregate_bias, stride=self.stride, padding=self.padding,
dilation=self.dilation, groups=self.groups*batch_size)
else:
output = F.conv3d(x, weight=aggregate_weight, bias=None, stride=self.stride, padding=self.padding,
dilation=self.dilation, groups=self.groups * batch_size)
output = output.view(batch_size, self.out_planes, output.size(-3), output.size(-2), output.size(-1))
return output
if __name__ == '__main__':
x = torch.randn(24, 3, 80, 80)
model = Dynamic_conv2d(in_planes=3, out_planes=64, kernel_size=3, ratio=0.25, padding=1,)
x = x.to('cuda:0')
model.to('cuda')
# model.attention.cuda()
# nn.Conv3d()
print(model(x).shape)
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
model.update_temperature()
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)
print(model(x).shape)