pytorch初探-搭建神经网络（图像篇）

这里举例简单的卷积网络，代码如下：

import torch.nn as nn
class myCNN(nn.Module):
    def __init__(self,numclass =62,charnum =4):
        super(myCNN, self).__init__()#继承
        self.numclass = numclass
        self.charnum = charnum
        self.conv = nn.Sequential(
            #这里搭建一层为例，input=3*120*40，output =16*60*20
            nn.Conv2d(3,16,2,padding=(1,1)),#卷积，输入通道3，输出通道16，卷积核2*2，边界1填充
            nn.MaxPool2d(2,2),#池化
            nn.BatchNorm2d(16),#归一化，参数为上一步输出通道数
            nn.ReLU()#激活函数
        )
        self.fc =nn.Linear(16*60*20,self.numclass*self.charnum)
    def forward(self, x):
        x = self.conv(x)
        x = x.view(-1,16*60*20)
        x = self.fc(x)
        return x