Exercise session 3 Question

Exercise session 3 Question

by Loic Nathan Thome Busson -
Number of replies: 1

Hi,

At the end of exercise session 3, we see that "With higher mean edge weight (C), the algorithm prefers paths that are shortest in terms of pixels rather than gradient value."

Because our goal is to follow the ridge (meaning, follow locations where the gradient is high), why aren't we setting this "C" value of mean edge weight to 0? (or something really really small?). All in all, I don't understand what this constant brings to the table, for me, the higher it is, the worse is the edge prediction (at least in this example, results are better with C = 0). So, might as well set it to 0 or something really small no?

Thanks,

Loïc

In reply to Loic Nathan Thome Busson

Re: Exercise session 3 Question

by Pascal Fua -

You are right when the gradients are strong. But suppose that the path has to connect two portions of the contour with a gap in between them where the gradient is weak or non-existent, for example due to noise or occlusions. In such a case, we want the straightest possible path and that's what C "brings to the table". As often in vision, it's a compromise.