1 What is pooling ？（Pooling）

1. Pooling, Pooling layer , Also known as down sampling 、 Convergence layer , It is the process of re selecting samples from samples . To reduce the data dimension .
2. Pooling There are two main categories ：① Maximum pooling （Max pooling）② Mean pooling （Avg pooling）
   The former is to take size×size The maximum . The latter is based on the average .

2 adaptivePooling and GeneralPooling

• AdaptivePooling, Adaptive pooling layer . Using this pooling method , nucleus （kernal） And step length （stride） Is a function based on the original size of the input 、 Target size Automatic calculation Coming out .

• GeneralPooling, Non adaptive pooling layer , You need to specify the size and step size of the core .

3 Code calls

• You don't need to implement the underlying code yourself , Can be called directly torch.nn.AdaptiveAvgPool2d(X)

1. Call in non adaptive mode

#  Maximum pooling 
out = nn.MaxPool2d(kernel_size=2,stride=2)  

#  Mean pooling 
out = nn.AvgPool2d(kernel_size=2,stride=2)  

2. Call of adaptive mode
   You just need to specify output_size, A parameter represents W and H identical , You can also customize it （H,W）

#  Adaptive maximum pooling 
out=nn.AdaptiveAvgPool2d(output_size=100)

#  Adaptive mean pooling 
out=nn.AdaptiveMaxPool2d((10, 15))