深度学习 第十讲 卷积神经网络（CNN）基础篇

import torch

in_channels,out_channels = 5,10
##输入通道数量  输出通道数量
width,height = 100,100

kernel_size = 3
#也可以用[5,3]这样的元组
#卷积核的大小，也就是filter的大小
batch_size = 1

input = torch.randn(batch_size,
                    in_channels,
                    width,
                    height)

conv_layer = torch.nn.Conv2d(in_channels,
                             out_channels,
                             kernel_size = kernel_size)

output = conv_layer(input)

print(input.shape)
#[1, 5, 100, 100] batch，5个通道，100 *100
print(output.shape)
print(conv_layer.weight.shape)
#[10, 5, 3, 3]  输入，输出，卷积核大小

在这里面imput的输入通道（in_channels）和conv_layer的in_channels要相同

1-如果想让input和output维度相同–padding填充
也就是在外面进行填充，将55填成77

代码

import torch
input = [3,4,6,5,7,
         2,4,6,8,2,
         1,6,7,8,4,
         9,7,4,6,2,
         3,7,5,4,1]
input = torch.Tensor(input).view(1,1,5,5)
###转换成tensor，维度是 batch，channel，5*5
conv_layer = torch.nn.Conv2d(1,1,kernel_size=3,padding=1,bias = False)
###在这里输入是一个通道，输出也是一个通道，bias = False不加偏置量
kernel = torch.Tensor([1,2,3,4,5,6,7,8,9]).view(1,1,3,3)
#输出通道，输入通道，3*3
conv_layer.weight.data = kernel.data
##卷积层的权重进行初始化
output = conv_layer(input)
print(output)

2-有效降低高度和宽度–stride = 2，可以增加步长

conv_layer = torch.nn.Conv2d(1,1,kernel_size=3,stride=2,bias = False)

最大池化层–默认stride = 2，在每一组中找出最大值

代码

import torch
input = [3,4,6,5,
         2,4,6,8,
         1,6,7,8,
         9,7,4,6]
input = torch.Tensor(input).view(1,1,4,4)
maxpooling_layer = torch.nn.MaxPool2d(kernel_size=2)
##默认stride = 2,卷积核大小也为2
output = maxpooling_layer(input)
print(output)

实现一个简单的神经网络


代码

class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = torch.nn.Conv2d(1,10,kernel_size=5)
        self.conv2 = torch.nn.Conv2d(10,20,kernel_size=5)
        self.pooling = torch.nn.MaxPool2d(2)
        ##卷积核大小为2
        self.fc = torch.nn.Linear(320,10)

    def forward(self,x):
        ###将（n，1,28,28）展开为（n，784）
        batch_size = x.size(0)
        ##取出n，batch
        x = F.relu(self.pooling(self.conv1(x)))
        x = F.relu(self.pooling(self.conv2(x)))
        x = x.view(batch_size,-1)
        ##拉伸成单列
        x = self.fc(x)
        return x