【pytorch】nn. AdaptiveMaxPool2d

Official document address

output_size – the target output size of the image of the form H x W. Can be a tuple (H, W) or a single H for a square image H x H. H and W can be either a int, or None which means the size will be the same as that of the input.

return_indices – if True, will return the indices along with the outputs. Useful to pass to nn.MaxUnpool2d. Default: False

m = torch.nn.AdaptiveMaxPool2d((5, 7))
input = torch.randn(1, 64, 8, 9)
output = m(input) 
print(output.shape)  # torch.Size([1, 64, 5, 7])
 
# target output size of 7x7 (square)
m = nn.AdaptiveAvgPool2d(7)
input = torch.randn(1, 64, 10, 9)
print(m(input).shape)  # [1, 64, 7, 7]
# target output size of 10x7
m = nn.AdaptiveAvgPool2d((None, 7))
input = torch.randn(1, 64, 10, 9) # [1, 64, 10, 7]
output = m(input)
print(output.shape)

import torch.nn as nn
import torch

input = torch.tensor([[[[2, 4, 8, 15],
                        [3, 6, 9, 19],
                        [7, 22, 5, 12],
                        [1, 66, 1, 77]]]], dtype=torch.float64)

m = nn.AdaptiveMaxPool2d((3, 3))
output = m(input)
print(input)
print(output)

# padding  I made it up myself , Convenient observation 
# tensor([[[[0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
# [0.0, 2., 4., 8., 15., 0.0],
# [0.0, 3., 6., 9., 19., 0.0],
# [0.0, 7., 22., 5., 12., 0.0],
# [0.0, 1., 66., 1., 77., 0.0],
# [0.0, 0.0, 0.0, 0.0, 0.0, 0.0]]]], dtype=torch.float64)

# tensor([[[[ 6., 9., 19.],
# [22., 22., 19.],
# [66., 66., 77.]]]], dtype=torch.float64)

---

Global average pooling GAP（Global Average Pooling） The advantage is that ：

Categories on Feature Map The connection between them is more intuitive （ Compared with the black box with full connection layer ）,Feature Map It is also easier to convert into classification probability , Because in GAP No parameters need to be adjusted in , Therefore, the problem of over fitting is avoided .

GAP Summarizes spatial information , Therefore, it is more robust to the spatial transformation of input , Therefore, the following full connection layers in convolutional networks , Mostly with GAP Replace .

Global pooling occurs in Kreas There are corresponding layers in , Maximize the pool layer for the global （GlobalMaxPooling2D）, But in Pytorch Although there is no corresponding pool layer in , But you can use Pytorch Adaptive pooling layer in （==AdaptiveMaxPool2d(1) perhaps AdaptiveMaxPool2d(1)== To achieve ）