[Insight-developers] Re: TensorRelativeAnisotropyImageFilterTest

[Luis Ibanez]

Hi Bill,

Thanks for pointing this out.

I added safeguards for preventing divisions by zero and
computation of square roots of negative numbers.

You are right, it seems that the Hessian is not a good
input for this test. We should figure out a way of
generating positive definite tensors.


Any suggestions are welcome...


     Luis


-------------------------------------
Lorensen, William E (Research) wrote:
> Luis, The subject test is failing. On borland it crashes. On other
> platforms if computes 1.#inf or someother bad values. The RA code
> divides by the trace, which is zero sometimes in the test images.
> We're not sure if the Hessian is guaranteed to be positive definite.
> At any rate, the test needs help.
> 
> Bill
> 
>