About padding

In week1 of CNN in deep learning, we have discussed why to use odd filters, first reason is padding is being asymmetric.
I didn’t get second reason. about the importance of central element.

Hello @XOXO
Another reason for padding is without padding, the convolution kernel cannot access the border pixels as often as the central pixels, which may result in the loss of important edge information. Padding ensures that the convolution operation can consider border pixels as thoroughly as the central ones.


In simple terms it means that to ensure that your filter does not slide beyond the pixel of your image, pad the edges with constant values like 0 to extend it and make it capture the real pixel needed for it learning.

1 Like