13.Sequential

下图是CIFAR 10的网络框架图

一个RGB三通道的32×32的图片经过多次卷积、池化、展平和线性层之后输出output

本节内容只作为展示sequential和模型各层的搭建过程，未假如非线性层做分类。

卷积

以conv1为例，输入为3通道32×32，输出为32通道32×32，设置in_channels = 3 out_channels = 32即可，pytorch会自动生成32个通道为3的卷积核进行卷积运算，这里还涉及到padding的计算：

https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d

Hout = Wout = 32，dilation一般取1，kernel_size = 5，stride默认取1，可以得到padding为2

[弹幕]：same padding时，p=（f-1）\2 , f是过滤器的尺寸，也是卷积核的尺寸，也就是kernel的大小，为5，所以5-1除以2等于2。

import torch
from torch import nn
from torch.nn import Conv2d, MaxPool2d, Linear
class DEMO(nn.Module):
    def __init__(self):
        super(DEMO, self).__init__()
        self.conv1 = Conv2d(in_channels=3,out_channels=32,kernel_size=5,padding=2)
        self.maxpool = MaxPool2d(kernel_size=2)
        self.conv2 = Conv2d(in_channels=32,out_channels=32,kernel_size=5,padding=2)
        self.conv3 = Conv2d(in_channels=32,out_channels=64,kernel_size=5,padding=2)
        self.flatten = torch.nn.Flatten()
        self.linear1 = Linear(in_features=1024,out_features=64)
        self.linear2 = Linear(in_features=64,out_features=10)
    def forward(self,x):
        x = self.conv1(x)
        x = self.maxpool(x)
        x = self.conv2(x)
        x = self.maxpool(x)
        x = self.conv3(x)
        x = self.maxpool(x)
        x = self.flatten(x)
        x = self.linear1(x)
        x = self.linear2(x)
        return x
demo = DEMO()
print(demo)
#我们可以用torch.ones()创建一个值全为1的shape如自己设置的假想input，来测试网络结构
#设置成CIFAR 10网络输入层的shape
input = torch.ones((64,3,32,32))
output = demo(input)
print(output.shape)

Sequential

Sequential是一个顺序容器。模块将按照它们在构造函数中传递的顺序添加到其中，结构更加直观。

https://pytorch.org/docs/stable/generated/torch.nn.Sequential.html?highlight=sequential#torch.nn.Sequential

import torch
from torch import nn
from torch.nn import Conv2d, MaxPool2d, Linear
from torch.utils.tensorboard import SummaryWriter
class DEMO(nn.Module):
    def __init__(self):
        super(DEMO, self).__init__()
        self.model = torch.nn.Sequential(
            Conv2d(in_channels=3, out_channels=32, kernel_size=5, padding=2),
            MaxPool2d(kernel_size=2),
            Conv2d(in_channels=32, out_channels=32, kernel_size=5, padding=2),
            MaxPool2d(kernel_size=2),
            Conv2d(in_channels=32, out_channels=64, kernel_size=5, padding=2),
            MaxPool2d(kernel_size=2),
            torch.nn.Flatten(),
            Linear(in_features=1024, out_features=64),
            Linear(in_features=64, out_features=10),
        )
    def forward(self,x):
        x = self.model(x)
        return x
demo = DEMO()
print(demo)
#我们可以用torch.ones()创建一个值全为1的shape如自己设置的假想input，来测试网络结构
#设置成CIFAR 10网络输入层的shape
input = torch.ones((64,3,32,32))
output = demo(input)
print(output.shape)
writer = SummaryWriter('./logs_seq')
writer.add_graph(demo,input)
writer.close()

这样我们可以通过tensorboard来查看该结构