[Insight-users] Gradient computation bug in SingleValuedVnlCostFunctionAdaptor when scales are used (patch included)

[Tom Vercauteren]

Hi all,

I think I found a bug in
SingleValuedVnlCostFunctionAdaptor::ConvertExternalToInternalGradient.
Here is a brief description:

SingleValuedVnlCostFunctionAdaptor can be used to wrap an itk cost
function into a vnl cost function.

Additionally the user can specify a set of scales such that the
function to be optimized is F(x) = f(x./s) instead of f(x).

The gradient of F, grad(F)(x) is thus given by grad(F)(x) = grad(f)(x./s)./s

Instead of that, the cost function adpator returns grad(f)(x./s)


I have filed a bug report in the bug tracker:
http://www.itk.org/Bug/view.php?id=6887

Let me know if I am missing something.

Best regards,
Tom Vercauteren