Week3 quiz question

I thought stride ensures size shrinks 16x, which generates a name for FCN (FCN16) on the final step in decoder… Am I wrong?

It seems you are wrong, yes! Kernel size is also important in size reduction…

Hi Gent:
when I retook the test, I also selected kernel_size just to pass grader, and I passed it. But I still disagree.
My understanding is like this: if original tensor is n by n, the size of conv2d’d tensor is (n+2p-f)/s + 1 by (n+2p-f)/s. Here f is filters (f,f) and s is stride (s,s) and p is padding (p,p). If you pad “exact”, 2p cancels f (p will be chosen to cancel f). So it is stride that will guarantee size shrink by 8x or 16x or whatever. And the size shrink of the last conv2d on the tensor in FCN gives name to FCN8/16 etc model.
Am I wrong?
Thank you