Adaptive Average Pooling 2D - Python

Introduction

The AdaptiveAvgPool2d module in PyTorch allows you to specify the output size of a 2D average pooling layer. This is unlike the traditional AvgPool2d layer where you must specify the kernel size and stride. The AdaptiveAvgPool2d layer must be specified with the desired output size, and it will automatically calculate the kernel size and stride necessary to achieve the desired output size given the input size.

Syntax

torch.nn.AdaptiveAvgPool2d(output_size)

Parameters

• output_size (tuple): Desired output size of the pooling operation. It can be a single integer, in which case the output size will be a square with side length equal to that integer, or it can be a tuple of integers in the form (height, width).

Example

import torch
import torch.nn as nn

class AdaptiveNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.adaptive_pool = nn.AdaptiveAvgPool2d((7,7))
        self.fc = nn.Linear(512*7*7, 10)
        
        def forward(self, x):
            x = self.adaptive_pool(x)
            x = x.view(x.size(0), -1)
            x = self.fc(x)
            return x

In this example, we are defining a neural network which applies adaptive average pooling to the input tensor and then passes it through a fully connected layer. The adaptive pooling layer has been specified to output a tensor with a height and width of 7. The output tensor is then flattened and fed to the fully connected layer.

Conclusion

AdaptiveAvgPool2d is a powerful module in PyTorch that allows you to easily control the output size of a 2D average pooling layer. This can be very useful in cases where you need to apply pooling to inputs of varying sizes, or you are unsure about the effect of specifying kernel size and stride.