以channel Attention Block為例子
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
|
class CAB(nn.Module): def __init__(self, in_channels, out_channels): super(CAB, self).__init__() self.global_pooling = nn.AdaptiveAvgPool2d(output_size=1) self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=1, padding=0) self.relu = nn.ReLU() self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=1, stride=1, padding=0) self.sigmod = nn.Sigmoid() def forward(self, x): x1, x2 = x # high, low x = torch.cat([x1,x2],dim=1) x = self.global_pooling(x) x = self.conv1(x) x = self.relu(x) x = self.conv2(x) x = self.sigmod(x) x2 = x * x2 res = x2 + x1 return res |
以上這篇pytorch forward兩個參數實例就是小編分享給大家的全部內容了，希望能給大家一個參考，也希望大家多多支持服務器之家。
原文鏈接：https://blog.csdn.net/weixin_41950276/article/details/89069659
