pytorch实现DarkNet53网络

DarkNet53网络结构

DarkNet53网络结构如下图所示，主要由两个基本单元组成，一个卷积单元
Convolutional，一个残差单元Residual

• Convolutional由一个卷积层一个BN层和一个leakRelu层组成
• Residual单元完成两次卷积，并进行加残差后返回相同维度的特征图
  DarkNet53网络总共由52个卷积层，最后还有一个1x1的卷积层将前52层提取到的特征进行融合和输出
  

pytorch实现

Convolutional单元

一个Convolutional单元中由一个卷积层一个BN层和一个leakRelu层组成

class Convolutional(nn.Module):
    def __init__(self, c_in, c_out, k, s, p, bias=False):
        """
        自定义一个卷积块，实现卷积+归一化+激活
        :param c_in: 输入通道数
        :param c_out: 输出通道数
        :param k: 卷积核尺寸
        :param s: 卷积核的步距
        :param p: padding
        :param bias: 偏置，使用BN层时设置为False
        """
        super(Convolutional, self).__init__()
        self.conv = nn.Sequential(
            nn.Conv2d(c_in, c_out, k, s, p, bias),
            nn.BatchNorm2d(c_out),
            nn.LeakyReLU(0.1),
        )

    def forward(self, x):
        return self.conv(x)

Residual单元

一个Residual单元里有一个11的卷积层和一个33的卷积层，11的卷积层将输入通道数降低一半，再用33的卷积层将输入通道数升回到原来的通道数，在前向传播中，最终输出的既保留原始信息，又融入了提取到的特征

class ConvResidual(nn.Module):
    def __init__(self, c_in):
        """
        自定义残差单元，只需给出通道数，该单元完成两次卷积，并进行加残差后返回相同维度的特征图
        :param c_in: 输入通道数
        """
        c = c_in // 2
        super(ConvResidual, self).__init__()
        self.conv = nn.Sequential(
            Convolutional(c_in, c, 1, 1, 0),  
            Convolutional(c, c_in, 3, 1, 1),  
        )

    def forward(self, x):
        return x + self.conv(x)  

实现DarkNet53网络

实现了这两个基本单元，就可以按照上图的网络结构，搭建DarkNet53网络，

class Darknet53(nn.Module):
    def __init__(self):
        super(Darknet53, self).__init__()
        self.conv1 = Convolutional(3, 32, 3, 1, 1)  
        self.conv2 = Convolutional(32, 64, 3, 2, 1)
        self.conv3_4 = ConvResidual(64)  
        self.conv5 = Convolutional(64, 128, 3, 2, 1)
        self.conv6_9 = nn.Sequential(  
            ConvResidual(128),
            ConvResidual(128),
        )
        self.conv10 = Convolutional(128, 256, 3, 2, 1)
        self.conv11_26 = nn.Sequential(  
            ConvResidual(256),
            ConvResidual(256),
            ConvResidual(256),
            ConvResidual(256),
            ConvResidual(256),
            ConvResidual(256),
            ConvResidual(256),
            ConvResidual(256),
        )
        self.conv27 = Convolutional(256, 512, 3, 2, 1)
        self.conv28_43 = nn.Sequential(  
            ConvResidual(512),
            ConvResidual(512),
            ConvResidual(512),
            ConvResidual(512),
            ConvResidual(512),
            ConvResidual(512),
            ConvResidual(512),
            ConvResidual(512),
        )
        self.conv44 = Convolutional(512, 1024, 3, 2, 1)
        self.conv45_52 = nn.Sequential(  
            ConvResidual(1024),
            ConvResidual(1024),
            ConvResidual(1024),
            ConvResidual(1024),
        )

    def forward(self, x):
        conv1 = self.conv1(x)
        conv2 = self.conv2(conv1)
        conv3_4 = self.conv3_4(conv2)
        conv5 = self.conv5(conv3_4)
        conv6_9 = self.conv6_9(conv5)
        conv10 = self.conv10(conv6_9)
        conv11_26 = self.conv11_26(conv10)
        conv27 = self.conv27(conv11_26)
        conv28_43 = self.conv28_43(conv27)
        conv44 = self.conv44(conv28_43)
        conv45_52 = self.conv45_52(conv44)
        return conv45_52, conv28_43, conv11_26