[Insight-developers] CurvatureAnisotropicDiffusionTest failin g

[Miller, James V (Research)]

Josh, 

Have the defaults been updated to "reasonable" values for the
TimeStep?



> -----Original Message-----
> From: Joshua Cates [mailto:cates@sci.utah.edu]
> Sent: Thursday, March 20, 2003 1:01 PM
> To: Luis Ibanez
> Cc: Lorensen, William E (Research); Insight-developers (E-mail)
> Subject: Re: [Insight-developers] CurvatureAnisotropicDiffusionTest
> failing
> 
> 
> Hi Bill and Luis,
> 
> Additionally, I made some adjustments yesterday to the 
> CurvatureAD filter
> which clamp the maximum timestep to guarantee stability.  I 
> would suggest
> generating a baseline image for the CurvatureAD filter using 
> a time step
> of 0.125 on the 2D image (0.0625 on 3D image).  Double those for the 
> GradientAD filter.  In retrospect, exposing the time step to 
> users was a 
> dangerous idea.
> 
> I'll add more documentation to these filters to better 
> explain the time 
> step parameter.
> 
> Thanks,
> 
> Josh.
> 
> ______________________________
>  Josh Cates			
>  School of Computer Science	
>  University of Utah
>  Email: cates@sci.utah.edu
>  Phone: (801) 587-7697
>  URL:   http://www.sci.utah.edu/~cates
> 
> 
> On Thu, 20 Mar 2003, Luis Ibanez wrote:
> 
> > Hi Bill,
> > 
> > It's great to have the regression testing working !
> > 
> > ---
> > 
> > Looking at the images,
> > 
> http://www.itk.org/Testing/Sites/starsend.kitware.com/WinXP-VC
> 7/20030320-1144-Experimental/Results/__Examples_Filtering_Curv
> atureAnisotropicDiffusionImageFilterTest.html
> > 
> > Here is what seems to happen,
> > 
> > Some day ago a glitch was found in the anisotropic diffusion
> > filters that was preventing the TimeStep from being used.
> > 
> > That is, the User settings were not being honored. Instead,
> > a default time step was always being used.  Josh rapidly
> > fixed this bug, and as a consequence all the code that was
> > using anisotropic filters suddenly started to actually use
> > the TimeStep parameters. It happened that most of these settings
> > were incorrectly tuned, often leading to numerical instabilities.
> > 
> > The Baselineimage for this filter seems to have been generated
> > just after the bug was fixed but before the new parameters were
> > tunned. The 'chekers'-like  appearance of the image is due to
> > the numerical instabilities originated from too large values
> > of the time step.
> > 
> > So, to summarize:
> > 
> >    Please update the Baseline image with the current one.
> >    The smooth image appearing as "Testimage" is the right one.
> > 
> > 
> > BTW, this may be happening with other tests...
> > 
> > 
> >      Luis
> > 
> > 
> > -----------------
> > Lorensen, William E (Research) wrote:
> > > Luis and Josh,
> > > The subject regression test from Examples/Filtering is 
> failing. I think the
> > > changes Josh made yesterday to the filter have had an 
> effect. Do you want me
> > > check in a new baseline image or is there a problem?
> > > 
> > > 
> http://public.kitware.com/Insight/Testing/HTML/TestingResults/
Sites/coffee.k
> >
itware/Linux-gcc296/20030320-0500-Nightly/Results/__Examples_Filtering_Curva
> > tureAnisotropicDiffusionImageFilterTest.html
> > 
> > Bill
> > 
> > _______________________________________________
> > Insight-developers mailing list
> > Insight-developers@public.kitware.com
> > http://public.kitware.com/mailman/listinfo/insight-developers
> > 
> 
> 
> 

_______________________________________________
Insight-developers mailing list
Insight-developers@public.kitware.com
http://public.kitware.com/mailman/listinfo/insight-developers