Global Pooling in Convolutional Neural Networks

Pooling operations have been a mainstay in convolutional neural networks for some time. While processes like max pooling and average pooling have often taken more of the center stage, their less known cousins global max pooling and global average pooling have become equally as important. In this article, we will be exploring what the global variants of the two common pooling techniques entail and how they compare to one another. Many beginners in computer vision often get introduced to convolutional neural networks as the ideal neural network for image data as it retains the spatial structure of the input image while learning/extracting features from them. By doing so it is able to learn relationships between neighboring pixels and the position of objects in the image thereby making it a very powerful neural network.