【代码分析】Unet-Pytorch

1：unet_parts.py

主要包含：

【1】double conv，双层卷积

【2】down，下采样

【3】up，上采样

【4】out conv，输出卷积

""" Parts of the U-Net model """

import torch
import torch.nn as nn
import torch.nn.functional as F


class DoubleConv(nn.Module):
    """(convolution => [BN] => ReLU) * 2"""

    def __init__(self, in_channels, out_channels, mid_channels=None):
        super().__init__()
        if not mid_channels:
            mid_channels = out_channels
        self.double_conv = nn.Sequential(
            nn.Conv2d(in_channels, mid_channels, kernel_size=3, padding=1, bias=False),
            nn.BatchNorm2d(mid_channels),
            nn.ReLU(inplace=True),
            nn.Conv2d(mid_channels, out_channels, kernel_size=3, padding=1, bias=False),
            nn.BatchNorm2d(out_channels),
            nn.ReLU(inplace=True)
        )

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


class Down(nn.Module):
    """Downscaling with maxpool then double conv"""

    def __init__(self, in_channels, out_channels):
        super().__init__()
        self.maxpool_conv = nn.Sequential(
            nn.MaxPool2d(2),
            DoubleConv(in_channels, out_channels)
        )

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


class Up(nn.Module):
    """Upscaling then double conv"""

    def __init__(self, in_channels, out_channels, bilinear=True):
        super().__init__()

        # if bilinear, use the normal convolutions to reduce the number of channels
        if bilinear:
            self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)
            self.conv = DoubleConv(in_channels, out_channels, in_channels // 2)
        else:
            # // 是整除运算
            self.up = nn.ConvTranspose2d(in_channels, in_channels // 2, kernel_size=2, stride=2)
            self.conv = DoubleConv(in_channels, out_channels)

    def forward(self, x1, x2):
        x1 = self.up(x1)
        # input is CHW
        diffY = x2.size()[2] - x1.size()[2]
        diffX = x2.size()[3] - x1.size()[3]

        x1 = F.pad(x1, [diffX // 2, diffX - diffX // 2,
                        diffY // 2, diffY - diffY // 2])
        # if you have padding issues, see
        # https://github.com/HaiyongJiang/U-Net-Pytorch-Unstructured-Buggy/commit/0e854509c2cea854e247a9c615f175f76fbb2e3a
        # https://github.com/xiaopeng-liao/Pytorch-UNet/commit/8ebac70e633bac59fc22bb5195e513d5832fb3bd
        x = torch.cat([x2, x1], dim=1)
        return self.conv(x)


class OutConv(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(OutConv, self).__init__()
        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=1)

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

【1】double conv

=》卷积。卷积核是3*3，填充是1

=》批归一化。

=》ReLU。激活函数

=》卷积。卷积核是3*3，填充是1

=》批归一化。

=》ReLU。激活函数

【2】down

=》最大池化。池化核是2*2

=》double conv。

【3】up

=》上采样。可选择upsample + double conv 和 transpose + double conv

=》计算尺寸差异。

=》填充x1。使得x1和x2对齐

=》拼接x2和x1。按照dim=1，也就是channel通道拼接

=》double conv。

【4】out conv

=》卷积。卷积核是1*1

2：unet_model.py

主要包含：UNet完整架构

""" Full assembly of the parts to form the complete network """

from .unet_parts import *


class UNet(nn.Module):
    def __init__(self, n_channels, n_classes, bilinear=False):
        super(UNet, self).__init__()
        self.n_channels = n_channels
        self.n_classes = n_classes
        self.bilinear = bilinear

        self.inc = (DoubleConv(n_channels, 64))
        self.down1 = (Down(64, 128))
        self.down2 = (Down(128, 256))
        self.down3 = (Down(256, 512))
        factor = 2 if bilinear else 1
        self.down4 = (Down(512, 1024 // factor))
        self.up1 = (Up(1024, 512 // factor, bilinear))
        self.up2 = (Up(512, 256 // factor, bilinear))
        self.up3 = (Up(256, 128 // factor, bilinear))
        self.up4 = (Up(128, 64, bilinear))
        self.outc = (OutConv(64, n_classes))

    def forward(self, x):
        x1 = self.inc(x)
        x2 = self.down1(x1)
        x3 = self.down2(x2)
        x4 = self.down3(x3)
        x5 = self.down4(x4)
        x = self.up1(x5, x4)
        x = self.up2(x, x3)
        x = self.up3(x, x2)
        x = self.up4(x, x1)
        logits = self.outc(x)
        return logits

    def use_checkpointing(self):
        self.inc = torch.utils.checkpoint(self.inc)
        self.down1 = torch.utils.checkpoint(self.down1)
        self.down2 = torch.utils.checkpoint(self.down2)
        self.down3 = torch.utils.checkpoint(self.down3)
        self.down4 = torch.utils.checkpoint(self.down4)
        self.up1 = torch.utils.checkpoint(self.up1)
        self.up2 = torch.utils.checkpoint(self.up2)
        self.up3 = torch.utils.checkpoint(self.up3)
        self.up4 = torch.utils.checkpoint(self.up4)
        self.outc = torch.utils.checkpoint(self.outc)

其中，use_checkpointing的作用是丢弃中间计算结果，加快训练速度。

上面的代码可以结合下图分析

前向传播过程：

        x1 = self.inc(x)

通过double conv双层卷积，输入通道为图像自身的，输出通道为64

        x2 = self.down1(x1)

通过down下采样，输入通道为64，输出通道为128

        x3 = self.down2(x2)

通过down下采样，输入通道为128，输出通道为256

        x4 = self.down3(x3)

通过down下采样，输入通道为256，输出通道为512

        x5 = self.down4(x4)

通过down下采样，输入通道为512，输出通道为1024（非bilinear，后续上采样也是如此）

        x = self.up1(x5, x4)

通过up上采样，输入通道为1024，输出通道为512

这个地方concat的对象是x4，也就是下采样输出通道为512的时候的特征

        x = self.up2(x, x3)

通过up上采样，输入通道为512，输出通道为256

这个地方concat的对象是x，也就是原图（后续也是原图）

其实这里和原作者的跳跃连接有点不太一样，代码库的作者直接省事用了原图进行拼接

        x = self.up3(x, x2)

通过up上采样，输入通道为256，输出通道为128

        x = self.up4(x, x1)

通过up上采样，输入通道为128，输出通道为64

        logits = self.outc(x)

通过out conv输出卷积，输入通道为64，输出通道为2，也就是分割为背景和物体2个类别的像素

3：完整代码

可以在github上通过git clone下载

milesial/Pytorch-UNet: PyTorch implementation of the U-Net for image semantic segmentation with high quality images (github.com)