Original article was published on Deep Learning on Medium
What is Pooling ?
Pooling is an approach to down sampling. It is a technique used to reduce the dimensionality of the image obtained from the previous convolutional layer, by reducing the number of pixels in the output. A pooling layer is a new layer added after the convolutional layer. Commonly used pooling methods are Max pooling, Average pooling and Min pooling .
- Max Pooling– calculates maximum of the each block of feature map.
- Average Pooling– calculates average of the each block of feature map.
- Min Pooling– calculates minimum of the each block of feature map.
We will focus on Max pooling for this post !!
Why Max Pooling is used?
Two main reasons why max pooling is effective is :
- It reduces the amount of parameters going forward and hence computational load.
- Higher valued pixels are the most activated and hence captured in this operation.
Max Pooling Process
Since we have got a basic idea what max pooling intends to do , let’s discuss about the operation what it actually does.
We have discussed about role of kernels in the previous post . They act as filters and when they convolve over an image, they create an output corresponding to computations of the filter applied .
For max pooling, we define
- filter(or kernel) of size n*n
- stride value k (by how many pixels we want our filter to move),
For each movement of the filter from the n*n pixels block of the image under consideration at that point, maximum value is captured as the output for the next layer . Then the filter is moved by k pixels (defined as stride) to perform the same operation again.One key point to be noted that Max pooling is applied after a convolutional layer .
You will hear about Global Average Pooling (GAP layer) too while working with deep learning models , will touch on that in coming posts . Till then , stay tuned ! stay safe !