AI Blog

AI Blog

by Michele Laurelli

Padding

/ˈpædɪŋ/
Technique
Definition

Adding extra pixels around the border of input data to control output size in convolution operations.

Padding preserves spatial dimensions and prevents information loss at borders. 'Valid' padding = no padding, 'Same' padding = output size matches input. Common: zero-padding.

Examples

1

Zero padding for same output size

2

Valid padding for reduction

3

Preserving border information

Michele Laurelli - AI Research & Engineering