- two problems without padding for [[Convolutional Neural Networks]]:
1. shrinking output (after 100 layers there's so little left in final image)
2. lack of usage of corner pixels' information
- to fix this, you pad image by 0's
- padding value = 1 means you add a single wraparound, which in the image example below means each side gets 2 extra pixels
![[CleanShot 2024-06-26 at
[email protected]|350]]