U-net 代码阅读笔记
论文阅读笔记可以参考这篇博客
代码链接:https://github.com/milesial/Pytorch-UNet/blob/master/unet/unet_model.py
先来看一下它的网络结构:
先来看一下代码的整体结构,构建了一个UNet类,四个downsampling layers,四个upsampling layers和一个输出网络层。
class UNet(nn.Module):
def __init__(self, n_channels, n_classes, bilinear=True):
super(UNet, self).__init__()
self.n_channels = n_channels
self.n_classes = n_classes
self.bilinear = bilinear
self.inc = DoubleConv(n_channels, 64) # 输入层
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
1.DoubleConv 类
先来看一下网络输入进来的时候,经过的两个3x3卷积层。较为简单,但是值得注意的是,原论文中使用的是unpadded convolutions,但是代码中使用了padding=1。查看了以下代码的issue,获得了下面的回答:
With padding the edges are not lost from the convolution, so it should always be better. I suppose if you want to strictly follow the paper, you should remove the padding, and lose 1 pixel for each edge and for each convolution.
意思就是说,有padding的话,对边界信息会好,不会丢失边界信息。
每个3x3卷积后面都紧跟BN和ReLU。
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),
nn.BatchNorm2d(mid_channels),
nn.ReLU(inplace=True),
nn.Conv2d(mid_channels, out_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True)
)
def forward(self, x):
return self.double_conv(x)
2.Downsampling layers
降采样层由Down这个类所构成。主要是实现2x2的max-pooling。等待max-pooling之后,紧跟着就是上面两组3x3 convolutions。因此直接调用了上面的DoubleConv类。
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)
3.Upsampling layers
上采样层由up类实现。可以看到,作者实现了两种方式
- bilinear upsampling
- Deconvolution upsampling
从作者调用的角度来看,使用的是第一种方法,线性插值的方法。
net = UNet(n_channels=3, n_classes=2, bilinear=True)
且上采样之后,计算与Contracting path相对应特征图相差的大小,然后进行padding。最后进行拼接操作。
class Up(nn.Module):
"""Upscaling then double conv"""
def __init__(self, in_channels, out_channels, bilinear=True):
super().__init__()
# would be a nice idea if the upsampling could be learned too,
# but my machine do not have enough memory to handle all those weights
# 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:
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)
4. Output layer
输出层由一个1x1卷积组成,通道数就是类别的个数。
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)
Conclusion
后续需要详细了解deconvolution和upsample的区别。