[Insight-developers] User reports serious problem with BSpline Registration

[Bill Lorensen]

Folks,

A user reported issues running an ITK Wiki example for registration.
The example uses the old framework. The problems occurs if the number
of threads is > 1. I know there are thread issues with the old
registration framework.

Question: Should we spend the time to track this down for this case?
Or just punt and tell users to use threading at their own risk, with a
workaround to set the threads to 1?

Here is the example. I have already set the threads to 1 to get it to work:
http://itk.org/Wiki/ITK/Examples/Registration/ImageRegistrationMethodBSpline

BTW, we have recently added a link to download tarballs for each wiki example:
http://itk.org/Wiki/ITK/Examples/Registration/ImageRegistrationMethodBSpline#Download_and_Build_ImageRegistrationMethodBSpline

You should be able to download the tarball and try it out.

Bill


---------- Forwarded message ----------
From: Bill Lorensen <bill.lorensen at gmail.com>
Date: Tue, Dec 11, 2012 at 3:00 PM
Subject: Re: [Insight-users] Serious problem with BSpline Registration
To: "Dr. Tim Allman" <dr.tim.allman at gmail.com>
Cc: Insight Users <insight-users at itk.org>


We know that the old itk registration classes have thread issues.

I found the same problem with the example on a multi processor ubuntu
system that you reported.

I'll start a discussion on the itk developer's list to see if we
should spend time fixing the old framework (which is still in ITKv4).

Thanks for reporting this. In the meantime, I'll add an explicit
thread setting in the example.

Bill

BTW, you might notice that the wiki examples now have the ability to
download a tar file for an individual example. Have you tried it? We
have not announced it yet.

http://itk.org/Wiki/ITK/Examples/Registration/ImageRegistrationMethodBSpline#Download_and_Build_ImageRegistrationMethodBSpline

On Tue, Dec 11, 2012 at 2:41 PM, Dr. Tim Allman <dr.tim.allman at gmail.com> wrote:
> Interesting. I just tried this on the Mac and it gives the same results as
> on the dashboard. The picture is correct too. Note however the lines:
>
> THE MINIMIZATION TERMINATED WITHOUT DETECTING ERRORS.
> Optimizer stop condition = LBFGSOptimizer: Failure
>
> This seems contradictory.
>
> I'll continue work with one thread.
>
> Tim
>
>
> On 12-12-11 02:01 PM, Bill Lorensen wrote:
>>
>> Looks like a thread issue. Try this:
>> ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=1
>> export ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS
>>
>>
>> On Tue, Dec 11, 2012 at 1:20 PM, Bill Lorensen <bill.lorensen at gmail.com>
>> wrote:
>>>
>>> Yes, there seems to be something strange.
>>>
>>> The nightly linux dashboard is clean and that example runs as expected.
>>> http://open.cdash.org/testDetails.php?test=169660112&build=2707835
>>>
>>> I just built my own version on linux and it builds and produces the
>>> proper results.
>>>
>>> I'll try my Mac build later.
>>>
>>> I'll also try a Superbuild version.
>>>
>>> Bill
>>>
>>> On Tue, Dec 11, 2012 at 12:46 PM, Dr. Tim Allman
>>> <dr.tim.allman at gmail.com> wrote:
>>>>
>>>> I am trying to do 2D BSpline based registrations but with little
>>>> success. It
>>>> appears that there is a problem inside ITK which is demonstrated by the
>>>> ITKWikiExample "Registration/ImageRegistrationMethodBSpline.cxx". I get
>>>> the
>>>> same results when I compile on Linux and OSX, 32 or 64 bits. The default
>>>> Superbuild exhibits the same behaviour.
>>>>
>>>> The first thing I noticed is that the example does not give the result
>>>> presented on the web page
>>>>
>>>> http://www.itk.org/Wiki/ITK/Examples/Registration/ImageRegistrationMethodBSpline.
>>>> Further, small adjustments to the optimizer parameters make wild changes
>>>> to
>>>> the appearance of the image.
>>>>
>>>> If the tests are run, the following results:
>>>>
>>>> The following tests FAILED:
>>>>        1 - CheckPagesForImages (Failed)
>>>>      171 - Registration-ImageRegistrationMethodBSpline (Failed)
>>>>      235 - Statistics-KdTreeBasedKMeansClustering_1D (SEGFAULT)
>>>>      236 - Statistics-KdTreeBasedKMeansClustering_3D (SEGFAULT)
>>>> Errors while running CTest
>>>>
>>>> Any help in this would be much appreciated.
>>>>
>>>> --
>>>> Tim Allman, Ph.D.
>>>> 35 Margaret Street,
>>>> Guelph Ontario N1E 5R6
>>>> Canada
>>>>
>>>>
>>>> _____________________________________
>>>> Powered by www.kitware.com
>>>>
>>>> Visit other Kitware open-source projects at
>>>> http://www.kitware.com/opensource/opensource.html
>>>>
>>>> Kitware offers ITK Training Courses, for more information visit:
>>>> http://www.kitware.com/products/protraining.php
>>>>
>>>> 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
>>>>
>>>
>>>
>>> --
>>> Unpaid intern in BillsBasement at noware dot com
>>
>>
>>
>
> --
> Tim Allman, Ph.D.
> 35 Margaret Street,
> Guelph Ontario N1E 5R6
> Canada
>



--
Unpaid intern in BillsBasement at noware dot com


-- 
Unpaid intern in BillsBasement at noware dot com