segmentation.zip

import torch import torch as t import torch.nn as nn import torch.nn.functional as F class DoubleConv(nn.Module):#（两次卷积） def __init__(self, in_channels,out_channels): super().__init__() # self.some_dbb1 = DiverseBranchBlock(in_channels,out_channels,kernel_size=3,padding=1,deploy=False) # self.conv1 = nn.Conv2d(in_channels,out_channels,kernel_size=3,padding=1) self.batch1 = nn.BatchNorm2d(out_channels) self.relu1 = nn.ReLU(inplace=True) # self.some_dbb2 = DiverseBranchBlock(out_channels, out_channels, kernel_size=3, padding=1, deploy=False) self.conv2 = nn.Conv2d(out_channels,out_channels,kernel_size=3,padding=1) self.batch2 = nn.BatchNorm2d(out_channels) self.relu2 = nn.ReLU(inplace=True) # self.double_conv = nn.Sequential( # #nn.Conv2d的功能是：对由多个输入平面组成的输入信号进行二维卷积 # nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1), # #batchmorm2d做归一化处理 # nn.BatchNorm2d(out_channels), # #inplace选择是否进行覆盖运算，是否将计算得到的新值覆盖之前原来的值 # nn.ReLU(inplace=True), # nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1), # nn.BatchNorm2d(out_channels), # nn.ReLU(inplace=True) # ) def forward(self, x): x = self.conv1(x) x = self.batch1(x) x = self.relu1(x) # x = self.sa(x) x = self.conv2(x) x = self.batch2(x) x = self.relu2(x) return x class Down(nn.Module): def __init__(self, in_channels, out_channels): super().__init__() self.maxpool = nn.MaxPool2d(2) self.Doub = DoubleConv(in_channels,out_channels) #使用卷积进行2倍下采样，通道数不变 # self.maxpool_conv = nn.Sequential( # nn.MaxPool2d(2), # # DoubleConv(in_channels, out_channels) # ) def forward(self,x): x = self.maxpool(x) x = self.Doub(x) return x #return self.maxpool_conv(x) class Up(nn.Module): def __init__(self, in_channels, out_channels, bilinear=True): super().__init__() # 上采样，特征图扩大两倍，通道数减半 if bilinear: #采用双线性插值 self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True) else: #采用反卷积 # self.up = nn.DUpsampling(in_channels//2, in_channels//2, kernel_size=2, stride=2) self.up = nn.ConvTranspose2d(in_channels , out_channels , kernel_size=2, stride=2) self.conv = DoubleConv(in_channels, out_channels) def forward(self, x1, x2): #x1接收上采样数据，x2接收特征融合数据 x1 = self.up(x1) #，skip_connection两边的图像大小不一样，根据要求，需要将decoder部分 #扩展到encoder部分的大小。 diffY = x2.size()[2]-x1.size()[2]#第一个参数代表下采样 #中图片的height，第二个参数代表目标图片的大小。 diffX = x2.size()[3]-x1.size()[3]#第一个参数代表下采样 #中图片的weight，第二个参数代表目标图片的大小。 #需扩展的像素（暂时这么说，对上采样的像素进行扩展，扩展到下采样图像的大小） x1 = F.pad(x1, [diffX//2, diffX-diffX//2, diffY//2, diffY-diffY//2]) #拼接，当前采样的，和之前下采样的 x = torch.cat([x2, x1], dim=1) return self.conv(x) class OutConv(nn.Module):#最后一步，输出 def __init__(self, in_channels, out_channels): super(OutConv, self).__init__() # self.sigmoid = nn.sigmoid() self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=1) #forward方法接受张量作为输入，返回张量作为输出 def forward(self, x): x = self.conv(x) return x #return self.conv(x)