[Insight-users] Orthogonality tolerance test

[Andriy Fedorov]

Luis,

I think orthogonalization is the best resolution. I will follow the code 
you reference. I am not sure changing the tolerance is an option, 
because it may result in transforms that will not be readable by tools 
that use ITK with the original tolerance setting.

Thank you for the reply.

Andriy Fedorov




insight-users-request at itk.org wrote on 07/02/2009 06:56 PM:
> Date: Thu, 2 Jul 2009 18:38:34 -0400
> From: Luis Ibanez <luis.ibanez at kitware.com>
> Subject: 
> To: Andriy Fedorov <fedorov at bwh.harvard.edu>
> Cc: ITK Users <insight-users at itk.org>
> Message-ID:
> 	<f7abd23c0907021538v123d840ci56a5418d96d4ff9b at mail.gmail.com>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Hi Andy,
>
> Thanks for pointing this out.
>
> We have struggled in the past with the selection
> of values to be used as tolerances...
>
> We tend to put tolerances that are too stringent,
> and sometimes unrealistic.  I recall that recently
> Hans Johnson fixed several classes that were
> using tolerances that were too exigent.
>
> You may be pointing out to a case that also needs
> reevaluation.
>
> Any suggestions on what could be a realistic value,
> that is still making sure that we don't take in matrices
> that are too deformed ?
>
> Also, another option to consider is that we have code
> that computes the Orthogonal part of a Matrix that
> is not fully orthogonal. This is done via Polar decomposition,
>
> See for example
> http://www.itk.org/pipermail/insight-users/2006-August/019025.html
>
> You could use this code for cleaning up that Matrix,
> before passing it to the Rigid Transform.
>
>
> Please let us know what you think is a better approach,
>
>
>         Thanks
>
>
>              Luis
>
>
> ------------------------------------------------------------------------
> On Thu, Jul 2, 2009 at 11:43 AM, Andriy Fedorov <fedorov at bwh.harvard.edu>wrote:
>
>   
>> Hi,
>>
>> What is the motivation for setting the tolerance in matrix orthogonality
>> test to 1e-10 in itkRigid3DTransform.cxx?
>>
>> I have a transform produced by FSL FLIRT tool, which is orthogonal with
>> tolerance 1e-5. I am trying to re-shape this transform to an ITK
>> transform (FSL uses different origin conventions), so that I can import
>> it into Slicer. With the current tolerance setting this is not possible,
>> the code breaks when I am trying to set the matrix.
>>
>> Is the current check with 1e-10 tolerance level a correct and necessary
>> precaution? Isn't it too strict for practical purposes?
>>
>> Thanks
>>
>> Andriy Fedorov
>> _____________________________________
>> Powered by www.kitware.com
>>
>> Visit other Kitware open-source projects at
>> http://www.kitware.com/opensource/opensource.html
>>
>> Please keep messages on-topic and check the ITK FAQ at:
>> http://www.itk.org/Wiki/ITK_FAQ
>>
>> Follow this link to subscribe/unsubscribe:
>> http://www.itk.org/mailman/listinfo/insight-users
>>
>>