pytorch forward两个参数

以channel Attention Block为例子

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