Mastering Computer Vision with TensorFlow 2.x

Pooling

Pooling is the next operation after convolution. It is used to reduce the dimensionality and feature map size (width and height) without changing the depth. The number of parameters for polling is zero. The two most popular types of pooling are as follows:

• Max pooling
• Average pooling

In max pooling, we slide the window over the feature map and take the max value of the window, while with average pooling, we take the average value in the window. Together, the convolution and pooling layers perform the task of feature extraction. The following diagram shows the max and average pooling operations being used on a 7 x 7 image:

Note how ...