*Memos:
My post explains Pooling Layer.
My post explains AvgPool1d().
My post explains MaxPool1d().
My post explains MaxPool2d().
My post explains MaxPool3d().
My post explains requires_grad.
AvgPool2d() can get the 3D or 4D tensor of the one or more values computed by 2D average pooling from the 3D or 4D tensor of one or more elements as shown below:
*Memos:
- The 1st argument for initialization is
kernel_size(Required-Type:intortupleorlistofint). *It must be1 <= x. - The 2nd argument for initialization is
stride(Optional-Default:None-Type:intortupleorlistofint):
*Memos:
- It must be
1 <= x. - If
None,kernel_sizeis set.
- It must be
- The 3rd argument for initialization is
padding(Optional-Default:0-Type:intortupleorlistofint). *It must be0 <= x. - The 4th argument for initialization is
ceil_mode(Optional-Default:False-Type:bool). - The 5th argument for initialization is
count_include_pad(Optional-Default:True-Type:bool). - The 6th argument for initialization is
divisor_override(Optional-Default:None-Type:int). - The 1st argument is
input(Required-Type:tensorofintorfloat). - The tensor's
requires_gradwhich isFalseby default is not set toTruebyAvgPool2d().
import torch
from torch import nn
tensor1 = torch.tensor([[[8., -3., 0., 1., 5., -2.]]])
tensor1.requires_grad
# False
avgpool2d = nn.AvgPool2d(kernel_size=1)
tensor2 = avgpool2d(input=tensor1)
tensor2
# tensor([[[8., -3., 0., 1., 5., -2.]]])
tensor2.requires_grad
# False
avgpool2d
# AvgPool2d(kernel_size=1, stride=1, padding=0)
avgpool2d.kernel_size
# 1
avgpool2d.stride
# 1
avgpool2d.padding
# 0
avgpool2d.ceil_mode
# False
avgpool2d.count_include_pad
# True
avgpool2d.divisor_override
# None
avgpool2d = nn.AvgPool2d(kernel_size=1, stride=None, padding=0,
ceil_mode=False, count_include_pad=True,
divisor_override=None)
avgpool2d(input=tensor1)
# tensor([[[8., -3., 0., 1., 5., -2.]]])
avgpool2d = nn.AvgPool2d(kernel_size=2, padding=1)
avgpool2d(input=tensor1)
# tensor([[[2.0000, -0.7500, 1.5000, -0.5000]]])
avgpool2d = nn.AvgPool2d(kernel_size=2, padding=1, divisor_override=2)
avgpool2d(input=tensor1)
# tensor([[[4.0000, -1.5000, 3.0000, -1.0000]]])
avgpool2d = nn.AvgPool2d(kernel_size=3, padding=1)
avgpool2d(input=tensor1)
# tensor([[[0.5556, 0.6667]]])
avgpool2d = nn.AvgPool2d(kernel_size=4, padding=2)
avgpool2d(input=tensor1)
# tensor([[[0.3125, 0.2500]]])
avgpool2d = nn.AvgPool2d(kernel_size=5, padding=2)
avgpool2d(input=tensor1)
# tensor([[[0.2000, 0.1600]]])
avgpool2d = nn.AvgPool2d(kernel_size=6, padding=3)
avgpool2d(input=tensor1)
# tensor([[[0.1389, 0.1111]]])
avgpool2d = nn.AvgPool2d(kernel_size=7, padding=3)
avgpool2d(input=tensor1)
# tensor([[[0.1224]]])
etc.
my_tensor = torch.tensor([[[8.], [-3.], [0.]],
[[1.], [5.], [-2.]]])
avgpool2d = nn.AvgPool2d(kernel_size=1)
avgpool2d(input=my_tensor)
# tensor([[[8.], [-3.], [0.]],
# [[1.], [5.], [-2.]]])
avgpool2d = nn.AvgPool2d(kernel_size=2, padding=1)
avgpool2d(input=my_tensor)
# tensor([[[2.0000], [-0.7500]],
# [[0.2500], [0.7500]]])
avgpool2d = nn.AvgPool2d(kernel_size=2, padding=1, divisor_override=2)
avgpool2d(input=my_tensor)
# tensor([[[4.0000], [-1.5000]],
# [[0.5000], [1.5000]]])
avgpool2d = nn.AvgPool2d(kernel_size=3, padding=1)
avgpool2d(input=my_tensor)
# tensor([[[0.5556]], [[0.6667]]])
avgpool2d = nn.AvgPool2d(kernel_size=4, padding=2)
avgpool2d(input=my_tensor)
# tensor([[[0.3125]], [[0.3750]]])
avgpool2d = nn.AvgPool2d(kernel_size=5, padding=2)
avgpool2d(input=my_tensor)
# tensor([[[0.2000]], [[0.1600]]])
avgpool2d = nn.AvgPool2d(kernel_size=6, padding=3)
avgpool2d(input=my_tensor)
# tensor([[[0.1389]], [[0.1111]]])
avgpool2d = nn.AvgPool2d(kernel_size=7, padding=3)
avgpool2d(input=my_tensor)
# tensor([[[0.1020]], [[0.0816]]])
etc.
my_tensor = torch.tensor([[[[8.], [-3.], [0.]],
[[1.], [5.], [-2.]]]])
avgpool2d = nn.AvgPool2d(kernel_size=1)
avgpool2d(input=my_tensor)
# tensor([[[[8.], [-3.], [0.]],
# [[1.], [5.], [-2.]]]])
my_tensor = torch.tensor([[[[8], [-3], [0]],
[[1], [5], [-2]]]])
avgpool2d = nn.AvgPool2d(kernel_size=1)
avgpool2d(input=my_tensor)
# tensor([[[[8], [-3], [0]],
# [[1], [5], [-2]]]])