deeplabv3plus的Encoder模块代码

以下是deeplabv3plus的Encoder模块代码：

class Encoder(nn.Module):
    def __init__(self, in_channels, out_channels, output_stride=16):
        super(Encoder, self).__init__()
        if output_stride == 8:
            stride_1, stride_2, stride_3 = 1, 2, 1
            dilation_1, dilation_2, dilation_3 = 1, 1, 2
        else:
            stride_1, stride_2, stride_3 = 2, 2, 2
            dilation_1, dilation_2, dilation_3 = 1, 1, 1

        self.layer1 = nn.Sequential(
            nn.Conv2d(in_channels, 32, kernel_size=3, stride=stride_1, padding=dilation_1, dilation=dilation_1, bias=False),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),
            nn.Conv2d(32, 32, kernel_size=3, stride=1, padding=1, bias=False),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),
            nn.Conv2d(32, 64, kernel_size=3, stride=1, padding=1, bias=False),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )

        self.layer2 = nn.Sequential(
            nn.Conv2d(64, 128, kernel_size=3, stride=stride_2, padding=dilation_2, dilation=dilation_2, bias=False),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1, bias=False),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=3, stride=1, padding=1, bias=False),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True)
        )

        self.layer3 = nn.Sequential(
            nn.Conv2d(128, 256, kernel_size=3, stride=stride_3, padding=dilation_3, dilation=dilation_3, bias=False),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1, bias=False),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1, bias=False),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True)
        )

        self.layer4 = nn.Sequential(
            nn.Conv2d(256, 512, kernel_size=3, stride=1, padding=dilation_3, dilation=dilation_3, bias=False),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=dilation_3, dilation=dilation_3, bias=False),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=dilation_3, dilation=dilation_3, bias=False),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True)
        )

        self.layer5 = nn.Sequential(
            nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=dilation_3, dilation=dilation_3, bias=False),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=dilation_3, dilation=dilation_3, bias=False),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, out_channels, kernel_size=1, stride=1, bias=False)
        )

    def forward(self, x):
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = self.layer5(x)
        return x

该模块包含5个卷积层，每个卷积层包含多个卷积、BN和ReLU操作。它的输入是`in_channels`通道的特征图，输出是`out_channels`通道的特征图。`output_stride`参数控制特征图的输出步幅。该模块可以将特征图降采样为1/8或1/16。