U-Net代码真实还原（基于pytorch）_unet代码复现

首先，UNet框架图

我们先来实现一下 conv3X3卷积块，也就是下图部分：

def contracting_block(in_channels, out_channels):
    block = torch.nn.Sequential(
        nn.Conv2d(kernel_size=(3,3), in_channels=in_channels, out_channels=out_channels),
        nn.ReLU(),
        nn.BatchNorm2d(out_channels),
        nn.Conv2d(kernel_size=(3,3), in_channels=out_channels, out_channels=out_channels),
        nn.ReLU(),
        nn.BatchNorm2d(out_channels)
    )
    return block
1
2
3
4
5
6
7
8
9
10

然后实现一下中间的桥结构

        bottleneck = nn.Sequential(
            nn.Conv2d(kernel_size=(3,3), in_channels=512, out_channels=1024),
            nn.ReLU(),
            nn.BatchNorm2d(1024),
            nn.Conv2d(kernel_size=(3,3), in_channels=1024, out_channels=1024),
            nn.ReLU(),
            nn.BatchNorm2d(1024)
        )
1
2
3
4
5
6
7
8

上采样部分有一个拼接，需要从编码部分裁剪出来和解码部分一样尺寸的图片，通过torch中cat函数，将其拼接在一起，这里的思想是相对应的解码部分的图片大小减去上采样之后的图片大小，然后得到图片的差值，在编码器部分上下左右减去差值的1/2，如下代码块：其中e表示编码部分，d表示解码部分。

diffY = e.size()[2]-d.size()[2]
diffX = e.size()[3]-d.size()[3]
e = e[:,:, diffY//2:e.size()[2]-diffY//2, diffX//2:e.size()[3]-diffX//2]
1
2
3

上采样块的代码：

nn.ConvTranspose2d(in_channels, in_channels//2, kernel_size=(3,3), stride=2, padding=1, 
                                     output_padding=1)
1
2

将上面（除了第一个）代码拼接起来，形成总的解码过程

class expansive_block(nn.Module):
    def __init__(self, in_channels, mid_channels, out_channels):
        super(expansive_block, self).__init__()
        self.up = nn.ConvTranspose2d(in_channels, in_channels//2, kernel_size=(3,3), stride=2, padding=1, 
                                     output_padding=1)
        self.block = nn.Sequential(
            nn.Conv2d(kernel_size=(3,3), in_channels=in_channels, out_channels=mid_channels),
            nn.ReLU(),
            nn.BatchNorm2d(mid_channels),
            nn.Conv2d(kernel_size=(3,3), in_channels=mid_channels, out_channels=out_channels),
            nn.ReLU(),
            nn.BatchNorm2d(out_channels)
        )
    
    def forward(self, e, d):
        d = self.up(d)
        #concat
        diffY = e.size()[2]-d.size()[2]
        diffX = e.size()[3]-d.size()[3]
        e = e[:,:, diffY//2:e.size()[2]-diffY//2, diffX//2:e.size()[3]-diffX//2]
        cat = torch.cat([e, d], dim=1)
        out = self.block(cat)
        return out
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

还有最后一块1x1卷积的部分，我们单独提取出来，写一个函数块

def final_block(in_channels, out_channels):
    block = nn.Sequential(
        nn.Conv2d(kernel_size=(1,1), in_channels=in_channels, out_channels=out_channels),
        nn.ReLU(),
        nn.BatchNorm2d(out_channels),
    )
    return block
1
2
3
4
5
6
7

最后我们来实现总的U-Net网络：

class UNet(nn.Module):
    
    def __init__(self, in_channel, out_channel):
        super(UNet, self).__init__()
        #Encode
        self.conv_encode1 = contracting_block(in_channels=in_channel, out_channels=64)
        self.conv_pool1 = nn.MaxPool2d(kernel_size=2, stride=2)
        self.conv_encode2 = contracting_block(in_channels=64, out_channels=128)
        self.conv_pool2 = nn.MaxPool2d(kernel_size=2, stride=2)
        self.conv_encode3 = contracting_block(in_channels=128, out_channels=256)
        self.conv_pool3 = nn.MaxPool2d(kernel_size=2, stride=2)
        self.conv_encode4 = contracting_block(in_channels=256, out_channels=512)
        self.conv_pool4 = nn.MaxPool2d(kernel_size=2, stride=2)
        
        #Bottleneck
        self.bottleneck = nn.Sequential(
            nn.Conv2d(kernel_size=(3,3), in_channels=512, out_channels=1024),
            nn.ReLU(),
            nn.BatchNorm2d(1024),
            nn.Conv2d(kernel_size=(3,3), in_channels=1024, out_channels=1024),
            nn.ReLU(),
            nn.BatchNorm2d(1024)
        )
        
        # Decode
        self.conv_decode4 = expansive_block(1024, 512, 512)
        self.conv_decode3 = expansive_block(512, 256, 256)
        self.conv_decode2 = expansive_block(256, 128, 128)
        self.conv_decode1 = expansive_block(128, 64, 64)
        
        self.final_layer = final_block(64, out_channel)
    
    def forward(self, x):
        # Encode
        encode_block1 = self.conv_encode1(x);print('encode_block1:', encode_block1.size())
        encode_pool1 = self.conv_pool1(encode_block1);print('encode_pool1:', encode_pool1.size())
        encode_block2 = self.conv_encode2(encode_pool1);print('encode_block2:', encode_block2.size())
        encode_pool2 = self.conv_pool2(encode_block2);print('encode_pool2:', encode_pool2.size())
        encode_block3 = self.conv_encode3(encode_pool2);print('encode_block3:', encode_block3.size())
        encode_pool3 = self.conv_pool3(encode_block3);print('encode_pool3:', encode_pool3.size())
        encode_block4 = self.conv_encode4(encode_pool3);print('encode_block4:', encode_block4.size())
        encode_pool4 = self.conv_pool4(encode_block4);print('encode_pool4:', encode_pool4.size())

        # Bottleneck
        bottleneck = self.bottleneck(encode_pool4);print('bottleneck:', bottleneck.size())
        
        # Decode
        decode_block4 = self.conv_decode4(encode_block4, bottleneck);print('decode_block4:', decode_block4.size())
        decode_block3 = self.conv_decode3(encode_block3, decode_block4);print('decode_block3:', decode_block3.size())
        decode_block2 = self.conv_decode2(encode_block2, decode_block3);print('decode_block2:', decode_block2.size())
        decode_block1 = self.conv_decode1(encode_block1, decode_block2);print('decode_block1:', decode_block1.size())
        
        final_layer = self.final_layer(decode_block1)
        return final_layer
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54

我们来测试一下我们的代码是否能通：

image = torch.rand((1, 3, 572, 572))
image.size()
1
2

torch.Size([1, 3, 572, 572])

unet = UNet(in_channel=3, out_channel=1)
mask = unet(image)
1
2

encode_block1: torch.Size([1, 64, 568, 568])
encode_pool1: torch.Size([1, 64, 284, 284])
encode_block2: torch.Size([1, 128, 280, 280])
encode_pool2: torch.Size([1, 128, 140, 140])
encode_block3: torch.Size([1, 256, 136, 136])
encode_pool3: torch.Size([1, 256, 68, 68])
encode_block4: torch.Size([1, 512, 64, 64])
encode_pool4: torch.Size([1, 512, 32, 32])
bottleneck: torch.Size([1, 1024, 28, 28])
decode_block4: torch.Size([1, 512, 52, 52])
decode_block3: torch.Size([1, 256, 100, 100])
decode_block2: torch.Size([1, 128, 196, 196])
decode_block1: torch.Size([1, 64, 388, 388])

对代码有疑问欢迎探讨！