PyTorch torch.nn.AdaptiveMaxPool2d Function

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

torch.nn.AdaptiveMaxPool2dIt is an adaptive max pooling module in PyTorch.

It pools the input to a specified size, retaining the maximum value rather than the average.

Function Definition

torch.nn.AdaptiveMaxPool2d(output_size, return_indices=False)

Parameters

  • output_size: Output size
  • return_indices: Whether to return indices

Usage Examples

Example 1: Basic Usage

Example

import torch
import torch.nn as nn

# Global max pooling
gap = nn.AdaptiveMaxPool2d(1)

# Inputs with different sizes
x1 = torch.randn(1, 64, 32, 32)
x2 = torch.randn(1, 64, 16, 16)

print("32x32 ->", gap(x1).shape)
print("16x16 ->", gap(x2).shape)

Example 2: Returning Indices

Example

import torch
import torch.nn as nn

gap = nn.AdaptiveMaxPool2d(1, return_indices=True)

x = torch.randn(1, 64, 8, 8)
output, indices = gap(x)

print("Output shape:", output.shape)
print("Index shape:", indices.shape)
print("Index values:", indices.item())

Example 3: Comparison with AdaptiveAvgPool2d

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)

avgpool = nn.AdaptiveAvgPool2d(1)
maxpool = nn.AdaptiveMaxPool2d(1)

print("Input:\n", x[0, 0])
print("Average pooling:", avgpool(x).item())
print("Max pooling:", maxpool(x).item())

Use Cases

  • Global max pooling: Extract the most significant features
  • Feature aggregation: Preserve key information
  • Classification networks: Replace FC layers

Tip: Max pooling retains the most significant features, while average pooling is smoother.

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

Other Extensions