Linknet网络结构
Linknet网络结构
LinkNet图像语义分割
像素级的图像语义分割，不仅需要精确，还需要高效（例如：自动驾驶）
具体结构
整体结构
一个输入层+4个编码层+4个解码层+1个输出层
编码层
解码层
创建Linknet模型
思路：编写不同的block在最后输出阶段将其链接
1、编写 卷积模块 （卷积 + 激活 + BN）
2、编写 反卷积模块 （反卷积 + 激活 + BN）
3、编码器（4*卷积模块）
4、解码器（卷积模块+反卷积模块+卷积模块）
5、实现整体的网络结构 （卷积模型+反卷积模型+解码器+编码器）
卷积模块
class Convblock (nn.Module):
def __init__(self, in_channels, out_channels,
k_size=3,
stride=1,
padding=1):
super(Convblock, self).__init__()
self.conv_relu = nn.Sequential(
nn.Conv2d(in_channels, out_channels,
kernel_size=k_size,
stride=stride,
padding=padding),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True)
)
def forward(self, x):
x = self.conv_relu(x)
return x
反卷积模块
#反卷积模块
class Deconvblock (nn.Module):
def __init__(self,in_channels,out_channels,
k_size = 3,
stride = 2,
padding = 1,
output_padding = 1):
super(Deconvblock,self).__init__()
self.deconv = nn.ConvTranspose2d(in_channels,out_channels,kernel_size=k_size,stride=stride,padding=padding,output_padding=output_padding)
self.bn = nn.BatchNorm2d(out_channels)
def forward(self,x,is_act = True):
x = self.deconv(x)
if is_act:
x = torch.relu(self.bn(x))
return x
编码器
#编码器模块
class Encodeblock (nn.Module):
def __init__(self,in_channels,out_channels):
super(Encodeblock,self).__init__()
self.conv1 = Convblock(in_channels,out_channels,stride=2)
self.conv2 = Convblock(out_channels,out_channels)
self.conv3 = Convblock(out_channels,out_channels)
self.conv4 = Convblock(out_channels,out_channels)
self.short_cut = Convblock(in_channels,out_channels,stride=2)
def forward(self,x):
out1 = self.conv1(x)
out1 = self.conv1(out1)
short_cut = self.short_cut(x)
out2 = self.conv3(out1+short_cut)
out2 = self.conv4(out2)
return out2+out1
解码器
#解码器模块
class Encodeblock(nn.Module):
def __init__(self, in_channels, out_channels):
super(Encodeblock, self).__init__()
self.conv1_1 = Convblock(in_channels, out_channels, stride=2)
self.conv1_2 = Convblock(out_channels, out_channels)
self.conv2_1 = Convblock(out_channels, out_channels)
self.conv2_2 = Convblock(out_channels, out_channels)
self.shortcut = Convblock(in_channels, out_channels, stride=2)
def forward(self, x):
out1 = self.conv1_1(x)
out1 = self.conv2_1(out1)
residue = self.shortcut(x)
out2 = self.conv2_1(out1 + residue)
out2 = self.conv2_2(out2)
return out2 + out1
最终模型的编写（注意每一步的输出，中间有类Resnet结构）
#模型编写
class Net (nn.Module):
def __init__(self):
super(Net,self).__init__()
self.input_conv = Convblock(3,64,k_size=7,stride=2,padding=3)
self.input_maxpool = nn.MaxPool2d(kernel_size=(2,2))
self.encode1 = Encodeblock(64,64)
self.encode2 = Encodeblock(64,128)
self.encode3 = Encodeblock(128,256)
self.encode4 = Encodeblock(256,512)
self.decode4 = Decodeblock(512,256)
self.decode3 = Decodeblock(256,128)
self.decode2 = Decodeblock(128,64)
self.decode1 = Decodeblock(64,64)
self.deconv_out1 = Deconvblock(64,32)
self.conv_out = Convblock(32,32)
self.deconv_out2 = Deconvblock(32,2,k_size=2,padding=0,output_padding=0)
def forward(self,x):
x = self.input_conv(x)
x = self.input_maxpool(x)
e1 = self.encode1(x)
e2 = self.encode2(e1)
e3 = self.encode3(e2)
e4 = self.encode4(e3)
d4 = self.decode4(e4)
d3 = self.decode3(d4+e3)
d2 = self.decode2(d3+e2)
d1 = self.decode1(d2+e1)
f1 = self.deconv_out1(d1)
f2 = self.conv_out(f1)
f3 = self.deconv_out2(f2)
return f3