Edge Handling in the second convolusion process
Why does the first convolution process result in losing 2 pixels on each side, but only 1 pixel on each side during the second time?
1 answers ( 0 marked as helpful)
Hi,
This is because during the first convolution we are using a 5x5 kernel/filter, while the second time around the kernel is 3x3.
We are losing pixels on the edges because otherwise the kernel would be "sticking out", so the amount of pixels we are losing is basically how many pixels are there to the left and right of the "midpoint" of the kernel - 2 pixels on all sides for 5x5 filters and 1 pixel for the 3x3 filter.
Hope this helps!
Best,
Nikola, 365 Team