PyTorch torch.nn.AvgPool2d Function

PyTorch torch.nn 参考手册PyTorch torch.nn Reference Manual

torch.nn.AvgPool2dIt is a 2D average pooling module in PyTorch.

It computes the average for each window of the input, commonly used for downsampling and feature aggregation.

Function Definition

torch.nn.AvgPool2d(kernel_size, stride=None, padding=0, ceil_mode=False, count_include_pad=True)

Parameter Description

  • kernel_size: pooling window size
  • stride: stride, defaults to kernel_size
  • padding: padding size
  • ceil_mode: whether to use ceil to compute output size

Usage Examples

Example 1: Basic Usage

Example

import torch
import torch.nn as nn

pool = nn.AvgPool2d(kernel_size=2, stride=2)

x = torch.randn(1, 1, 4, 4)
output = pool(x)

print("Input:n", x.squeeze().tolist())
print("nOutput:n", output.squeeze().tolist())
print("Shape:", x.shape, "->", output.shape)

Example 2: Global Average Pooling

Example

import torch
import torch.nn as nn

# Global average pooling
gap = nn.AdaptiveAvgPool2d(1)

x = torch.randn(4, 64, 16, 16)
out = gap(x)

print("Input:", x.shape)
print("Output:", out.shape)
print("Number of averages per channel:", out.numel() // 64)

Example 3: Comparison with MaxPool

Example

import torch
import torch.nn as nn

x = torch.tensor([[[
    [1, 2, 3, 4],
    [5, 6, 7, 8],
    [9, 10, 11, 12],
    [13, 14, 15, 16]
]]], dtype=torch.float32)

maxpool = nn.MaxPool2d(2, 2)
avgpool = nn.AvgPool2d(2, 2)

print("Input:n", x[0, 0])
print("nMaxPool:n", maxpool(x)[0, 0])
print("nAvgPool:n", avgpool(x)[0, 0])

Use Cases

  • Feature aggregation: Reduce spatial dimensions
  • Global average pooling: Replace fully connected layers
  • Smooth featuresReduce noise

PyTorch torch.nn 参考手册PyTorch torch.nn Reference Manual

Other extensions