Unet网络结构搭建

from typing import Dict
import torch
import torch.nn as nn


class DoubleConv(nn.Sequential): #定义两个串联卷积模块
    def __init__(self, in_channels, out_channels, mid_channels=None):
        if mid_channels is None: #如果没有设置mid_channels，则mid_channels = out_channels
            mid_channels = out_channels
        super(DoubleConv, self).__init__(
            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)
        )


class Down(nn.Sequential): #定义下采样模块，最大池化+两个串联卷积模块
    def __init__(self, in_channels, out_channels):
        super(Down, self).__init__(
            nn.MaxPool2d(2, stride=2),
            DoubleConv(in_channels, out_channels)
        )


class Up(nn.Module):
    def __init__(self, in_channels, out_channels, bilinear=True):
        super(Up, self).__init__()
        if bilinear: #双线性插值
            self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)
            self.conv = DoubleConv(in_channels, out_channels, in_channels // 2)
        else: #转置卷积
            self.up = nn.ConvTranspose2d(in_channels, in_channels // 2, kernel_size=2, stride=2)
            self.conv = DoubleConv(in_channels, out_channels)

    # Up正向传播过程，先进行上采样，在进行拼接，拼接之后在经过DoubleConv
    def forward(self, x1: torch.Tensor, x2: torch.Tensor) -> torch.Tensor:
        x1 = self.up(x1) #此处只经过Upsample或ConvTranspose2d
        x = torch.cat([x2, x1], dim=1) #x2, x1进行拼接
        x = self.conv(x) #拼接之后在经过DoubleConv
        return x


class OutConv(nn.Sequential):
    def __init__(self, in_channels, num_classes):
        super(OutConv, self).__init__(
            nn.Conv2d(in_channels, num_classes, kernel_size=1) #1*1卷积调整最后的通道数
        )


class UNet(nn.Module):
    def __init__(self,
                 in_channels: int = 3,
                 num_classes: int = 2,
                 bilinear: bool = True,
                 base_c: int = 64):
        super(UNet, self).__init__()
        self.in_channels = in_channels
        self.num_classes = num_classes
        self.bilinear = bilinear

        self.in_conv = DoubleConv(in_channels, base_c)
        self.down1 = Down(base_c, base_c * 2)
        self.down2 = Down(base_c * 2, base_c * 4)
        self.down3 = Down(base_c * 4, base_c * 8)
        factor = 2 if bilinear else 1
        self.down4 = Down(base_c * 8, base_c * 16 // factor)
        self.up1 = Up(base_c * 16, base_c * 8 // factor, bilinear)
        self.up2 = Up(base_c * 8, base_c * 4 // factor, bilinear)
        self.up3 = Up(base_c * 4, base_c * 2 // factor, bilinear)
        self.up4 = Up(base_c * 2, base_c, bilinear)
        self.out_conv = OutConv(base_c, num_classes)

    def forward(self, x: torch.Tensor) -> Dict[str, torch.Tensor]:
        x1 = self.in_conv(x) # 480*480*3 -> 480*480*64 -> 480*480*64
        x2 = self.down1(x1) # 240*240*64 -> 240*240*128 -> 240*240*128
        x3 = self.down2(x2) # 120*120*128 -> 120*120*256 -> 120*120*256
        x4 = self.down3(x3) # 60*60*256 -> 60*60*512 -> 60*60*512
        x5 = self.down4(x4) # 30*30*512 -> 30*30*512 -> 30*30*512
        x = self.up1(x5, x4) # 60*60*512 -> 60*60*1024 -> 60*60*512 -> 60*60*256
        x = self.up2(x, x3) # 120*120*256 -> 120*120*512 -> 120*120*256 -> 120*120*128
        x = self.up3(x, x2) # 240*240*128 -> 240*240*256 -> 240*240*128 -> 240*240*64
        x = self.up4(x, x1) # 480*480*64 -> 480*480*128 -> 480*480*64 -> 480*480*64
        logits = self.out_conv(x) # 480*480*64-> 480*480*num_classes

        return {"out": logits}

reference

【使用Pytorch搭建U-Net网络并基于DRIVE数据集训练(语义分割)】 https://www.bilibili.com/video/BV1rq4y1w7xM?share_source=copy_web&vd_source=95705b32f23f70b32dfa1721628d5874