Deep Learning · Convolutions

CNN Stride & Padding

A kernel slides over the grid — stride sets the jump, padding adds a border. Watch the output size follow the formula.

stride 1 padding 0
input 5×5 + padding → feature map
output
3×3
position
—
output size
out = ⌊(5−3+0)/1⌋+1 = 3
kernel overlapkernel 3×3padding = 0
tip: stride = how far the kernel jumps · padding = zeros added around the border
conv.py