class MyNet(nn.Module): def __init__(self): super(MyNet, self).__init__() self.vgg16 = vgg16(pretrained=True) self.resnet18 = resnet18(pretrained=True) self.vgg16.classifier = nn.Identity() self.resnet18.fc = nn.Identity() self.fc = nn.Linear(25600, 2) def forward(self, x): x1 = self.vgg16(x) x2 = self.resnet18(x) x1 = x1.view(x1.size(0), -1) x2 = x2.view(x2.size(0), -1) x = torch.cat((x1, x2), dim=1) x = self.fc(x) return x 将以上代码加入DANet注意力机制

可以将 DAnet 注意力机制加入到 MyNet 的 forward 函数中，如下所示：

import torch
import torch.nn as nn

class DAnet(nn.Module):
    def __init__(self, in_channels, out_channels, reduction=16):
        super(DAnet, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.conv1 = nn.Conv2d(in_channels, in_channels // reduction, kernel_size=1, stride=1, padding=0)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = nn.Conv2d(in_channels // reduction, out_channels, kernel_size=1, stride=1, padding=0)
        self.sigmoid = nn.Sigmoid()

    def forward(self, x):
        b, c, _, _ = x.size()
        y = self.avg_pool(x)
        y = self.conv1(y)
        y = self.relu(y)
        y = self.conv2(y)
        y = self.sigmoid(y)
        return x * y.expand_as(x)

class MyNet(nn.Module):
    def __init__(self):
        super(MyNet, self).__init__()
        self.vgg16 = vgg16(pretrained=True)
        self.resnet18 = resnet18(pretrained=True)
        self.vgg16.classifier = nn.Identity()
        self.resnet18.fc = nn.Identity()
        self.danet1 = DAnet(512, 512)
        self.danet2 = DAnet(512, 512)
        self.fc = nn.Linear(25600, 2)

    def forward(self, x):
        x1 = self.vgg16.features(x)
        x1 = self.danet1(x1)
        x1 = self.vgg16.avgpool(x1)
        x1 = torch.flatten(x1, 1)
        
        x2 = self.resnet18.conv1(x)
        x2 = self.resnet18.bn1(x2)
        x2 = self.resnet18.relu(x2)
        x2 = self.resnet18.maxpool(x2)
        x2 = self.resnet18.layer1(x2)
        x2 = self.resnet18.layer2(x2)
        x2 = self.resnet18.layer3(x2)
        x2 = self.resnet18.layer4(x2)
        x2 = self.danet2(x2)
        x2 = self.resnet18.avgpool(x2)
        x2 = torch.flatten(x2, 1)

        x = torch.cat((x1, x2), dim=1)
        x = self.fc(x)
        return x

在 forward 函数中，我们首先对 VGG16 和 ResNet18 的特征提取部分进行计算，然后将 VGG16 的特征图 x1 和 ResNet18 的特征图 x2 分别输入到两个 DAnet 注意力机制中进行加权，最后将加权后的两个特征图拼接在一起，并通过全连接层进行分类。