[Insight-developers] Dashboard Looking Good : MultiResImageRegistrationTest (addtional baseline)

[Luis Ibanez]

Hi Bill,

Sounds good.
I just committed a modified CMakeLists.txt file with a relaxed
tolerance of 10,000 for the MultiResImageRegistration2 tests.


    Luis


------------------------
Bill Lorensen wrote:
> Luis,
> 
> Maybe we should increase the number of compared pixels for that test.
> I just bumped it to 10000 and the test passes on my mingw build. The
> computed offsets sure look great on all platforms, even the ones that
> are failing.
> 
> As for proximal, it is an experimental build. I would ignore it. Our
> official Borland builds are fine.
> 
> Bill
> 
> On Sun, Apr 13, 2008 at 10:34 AM, Luis Ibanez <luis.ibanez at kitware.com> wrote:
> 
>> Hi Bill,
>>
>>
>> Yes, we seem to need a second baseline for
>>
>>
>>      MultiResImageRegistration2Test2
>>
>>
>> The test is failing due to comparisons with the baseline:
>>
>>http://www.cdash.org/CDash/testSummary.php?project=2&name=MultiResImageRegistration2Test2&date=20080413
>>
>>
>>  in the following systems
>>
>> dash19.kitware  Win32-bcc32             Failed  48.52
>> SINTEF.LabPC.XP ITKcvs_CMake247_VS2003  Failed  68.14
>> dashsun1.kitware        SunOS-CC-5.6    Failed  121.00
>> james.uiowa     Linux-g++4-4.1-opt      Failed  34.51
>> dash13.kitware  Win32-vs60              Failed  33.50
>> Proximal        Win32-bcc5.5-Static     Not Run 0.00
>>
>>
>>
>> Proximal is failing to build ITK due to the following error
>> ..\..\..\..\Sources\Insight\Utilities\openjpeg\j2k_lib.c(30) :
>> fatal error C1083: Cannot open include file: 'sys/time.h':
>> No such file or directory
>>
>> So, that's a separate issue.
>>
>>
>>
>> The numerical convergences of the test are pretty good though:
>>
>>   dash19.kitware      X = 12.9823 Y = 16.9992
>>   SINTEF.LabPC.X      X = 12.982  Y = 16.9989
>>   dashsun1.kitware    X = 12.9819 Y = 16.9985
>>   james.uiowa         X = 12.982  Y = 16.9995
>>   dash13.kitware      X = 12.9845 Y = 17
>>
>> The ideal values are   X = 13.0000 Y = 17.0000
>>
>>
>>
>> The difference images of the first four systems
>> look quite close, while the one from dash13 (VS60)
>> looks different. I think part of the problem is
>> that we are resampling with the linear interpolator
>> and once we pass from 16.999 to 17.0001 that results
>> in a change of 1 pixel position when resampling.
>>
>> I just added a second baseline (actually third, since
>> there was already a second baseline ...1.png) by using
>> the result of dash19.kitware and we may have to add a
>> third (fourth) for dash14.kitware (VS60) later...
>>
>>
>>   Luis
>>
>>
>> -----------------------
>> Bill Lorensen wrote:
>>
>>
>>>Luis,
>>>
>>>The changes you made to the multi res registration examples are
>>>producing great results. Looks like we need a second regression image
>>>for the second multi res registration example.
>>>
>>>The tests DeformableRegistration15Test1 and
>>>DeformableRegistration15Test2 are failing on all platforms they run
>>>on. Looks like they use the data from BrainWeb (Test1) or Kilian
>>>(Test2).
>>>
>>>It's not clear to me what BrainWeb data is required to run the tests.
>>>And I don't think Kilian's data should be in a test checked into ITK
>>>unless we can add his data to the ITK distribution.
>>>
>>>Bill
>>>
>>>
>>
>