最大池化pytorch

**前置知识：

1、

复制代码

self.maxpool_2=MaxPool2d(kernel_size=3,ceil_mode=True)

复制代码

output=self.maxpool_2(input)

输入：张量的形状是（N,C,H,W）或（C,H,W）

Input: (N,C,Hin,Win)or (C,Hin,Win)
Output: (N,C,Hout,Wout)or (C,Hout,Wout)

参数：

池化核（池化窗口）大小：kernel_size (Union $* [int](https://docs.python.org/3/library/functions.html#int "int")*,* [Tuple](https://docs.python.org/3/library/typing.html#typing.Tuple "Tuple")*\[* [int](https://docs.python.org/3/library/functions.html#int "int")*,* [int](https://docs.python.org/3/library/functions.html#int "int")*$ ]) -- the size of the window to take a max over
步长：stride (Union $* [int](https://docs.python.org/3/library/functions.html#int "int")*,* [Tuple](https://docs.python.org/3/library/typing.html#typing.Tuple "Tuple")*\[* [int](https://docs.python.org/3/library/functions.html#int "int")*,* [int](https://docs.python.org/3/library/functions.html#int "int")*$ ] ) -- the stride of the window. Default value is kernel_size（默认是池化核的大小）
补边缘padding (Union $* [int](https://docs.python.org/3/library/functions.html#int "int")*,* [Tuple](https://docs.python.org/3/library/typing.html#typing.Tuple "Tuple")*\[* [int](https://docs.python.org/3/library/functions.html#int "int")*,* [int](https://docs.python.org/3/library/functions.html#int "int")*$ ]) -- Implicit negative infinity padding to be added on both sides
取整方式：ceil_mode (bool) -- when True, will use ceil instead of floor to compute the output shape（True：向上取整，保留不足的部分；False：向下取整，去除不足一份的部分）
空洞卷积dilation (Union $* [int](https://docs.python.org/3/library/functions.html#int "int")*,* [Tuple](https://docs.python.org/3/library/typing.html#typing.Tuple "Tuple")*\[* [int](https://docs.python.org/3/library/functions.html#int "int")*,* [int](https://docs.python.org/3/library/functions.html#int "int")*$ ]) -- a parameter that controls the stride of elements in the window

2、池化的作用：

从特征图中提取最有代表性的特征；防止过拟合，实现降维；保持平移不变性。

（即保留重要特征，同时减少数据量，使模型训练得更快 eg: 1080P高清------>720P高清）

**代码：

1、对单一二维矩阵进行最大池化：

input 单一二维矩阵reshape（变成3D或4D）------>nn 创建神经元------>output 计算并输出

python 复制代码

import torch
from torch import nn
from torch.nn import MaxPool2d

input=torch.tensor([
    [1,2,0,3,1],
    [0,1,2,3,1],
    [1,2,1,0,0],
    [5,2,3,1,1],
    [2,1,0,1,1] #dtype=torch.float32把整数变成小数
])

input=torch.reshape(input,(-1,1,5,5)) #-1是占位符，后续自动计算batch_size的大小
print(input.shape)

#神经元
class Xigua(nn.Module):
    def __init__(self):
        super().__init__()
        self.maxpool_2=MaxPool2d(kernel_size=3,ceil_mode=True) #保留不足的部分，也把它算进去

    def forward(self,input):
        output=self.maxpool_2(input)
        return output

xigua1=Xigua()
output=xigua1(input)
print(output)

2、对RGB图像进行池化：

input 导入并加载RGB图片数据集------>nn 创建神经元------>output 计算并记录

python 复制代码

import torch
import torchvision.datasets
from torch import nn
from torch.nn import MaxPool2d
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter

test_set=torchvision.datasets.CIFAR10(root="./dataset",train=False,transform=torchvision.transforms.ToTensor(),download=True)

dataloader=DataLoader(test_set,batch_size=64)

#神经元
class Xigua(nn.Module):
    def __init__(self):
        super().__init__()
        self.maxpool_2=MaxPool2d(kernel_size=3,ceil_mode=True) #保留不足的部分，也把它算进去

    def forward(self,input):
        output=self.maxpool_2(input)
        return output

xigua1=Xigua()
writer=SummaryWriter("logs2")
step=1
for imgs,targets in dataloader:
    print(imgs.shape)
    writer.add_images("input",imgs,step)
    imgs=xigua1(imgs)
    print(imgs.shape)
    writer.add_images("output",imgs,step)
    step=step+1
    if step>=3:
        break
writer.close()