0
$\begingroup$

I was reading a paper and this paragraph said that:

The ground truth score is calculated based on the intersectionover- union (IoU) of the perturbed image and the ground truth one. Since we would like to distinguish among IoU values close to 1.0, we use $IOU^3$ as the ground truth score.

I couldn't find any references to this, nor mentioned in the paper. Is it just simply the calculated IoU to the power of 3? Or is it a special kind of IoU calculation method? What does this mean?

$\endgroup$
3
  • $\begingroup$ Link? Are you sure 3 is not a reference? $\endgroup$
    – sfotiadis
    Jun 13 at 11:36
  • $\begingroup$ openaccess.thecvf.com/content/WACV2022/papers/… $\endgroup$
    – Ryan
    Jun 13 at 11:37
  • 1
    $\begingroup$ Yeah it's not a reference, it's under section 3.1, score regression network $\endgroup$
    – Ryan
    Jun 13 at 11:38

2 Answers 2

1
$\begingroup$

Since we would like to distinguish among IoU values close to 1.0, we use IOU3 as the ground truth score for the SRN.

It seems to be just IoU to the power of 3. They use the cube function because they want their regressor to pick up small differences in IoU when IoU is near 1.0. They could have used any other function that is sufficiently steep close to 1.0. I haven't seen power of 3 being used very often but it's pretty straightforward to implement and I assume it works better than the square in this problem for them to use it.

$\endgroup$
1
$\begingroup$

From context, I would say: Yes, it's IoU to the power of 3, since they want to have larger differences between values close to 1. Obviously, the difference between 0.9^3 and 1.0^3 is larger than between 0.9 and 1.0.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .