[Insight-users] Interpolation and centered pixel coordinates

[Dženan Zukić]

Hi Luis,

vec1 and vec2 are osg::Vec3f classes from openscenegraph.org. And it occurs
rarely, maybe once every few million executions.

Regards,
Dženan

2010/6/17 Luis Ibanez <luis.ibanez at kitware.com>

> Hi Dženan,
>
> Thanks for sharing your findings.
>
> This is quite interesting.
>
> Did you log a bug entry for this problem ?
>
> I'm trying to find the exact place with this
> conversion of:
>
>
> double cosFi=vec1*vec2/(vec1.length()*vec2.length());
>
> Is this something that you found in ITK ?
> or is this part or your application's code ?
>
>
>    Please let us know,
>
>
>            Thanks
>
>
>                 Luis
>
>
> ------------------------------------------------------
> 2010/6/16 Dženan Zukić <dzenanz at gmail.com>
>
> Hi Luis,
>>
>> I found out what the problem is.
>>
>> The way I call evaluate is OK:
>>
>> image->TransformPointFromPhysicalToContinouosIndex(p, ind, ok);
>> if (ok)
>>   pixel=interpolator->EvaluateAtContinuousIndex(ind);
>>
>> but my original point p had NaN values in it, which were caused by code
>> which looked something like this:
>>
>> double cosFi=vec1*vec2/(vec1.length()*vec2.length());
>> //cosine of angle between vectors
>> p+=acos(cosFi)*...;
>>
>> Due to numerical round-off errors it would rarely occur that cosFi>1
>> (something like 1.00000002534), and acos would produce NaN, which would then
>> propagate all the way to poor interpolator which would then complain.
>> Shortly after starting to use interpolator I also started using different
>> sort of input files, which made my program do much more iterations of the
>> above numerically unstable vector code and made manifestations of that
>> problem more often.
>>
>> Since the problem was manifesting itself in interpolator, and just before
>> calling interpolator I was getting its input data from
>> TransformPhysicalPointToContinuousIndex, and code for this transformation is
>> very convoluted, I suspected it might be wrong and I had to ask.
>>
>> I am also quite busy these last few weeks (I barely managed to find time
>> to fix that bug which I adopted :D), so I soon forgot I asked this question,
>> and of course didn't write that I found out the problem (which was quite
>> far, in calculation of normals and curvatures of the mesh).
>>
>> Thanks for the answer,
>> Dženan
>>
>> 2010/6/16 Luis Ibanez <luis.ibanez at kitware.com>
>>
>>
>>> Dženan,
>>>
>>> We did quite a heavy testing when we fixed
>>> the problem of centered pixel coordinates.
>>>
>>> The Dashboard was green when we released
>>> ITK 3.18, which give us some confidence on
>>> the correctness of the code.
>>>
>>> --
>>>
>>> Please note that it is your responsibility to call
>>>
>>>      interpolator->IsInsideBuffer( point )
>>>
>>> before you call
>>>
>>>      interpolator->Evaluate( point );
>>>
>>>
>>> That is, your code should take the form:
>>>
>>>
>>>    if(    interpolator->IsInsideBuffer( p1 ) )
>>>       {
>>>       p2 = interpolator->Evaluate( p1 );
>>>       }
>>>
>>>
>>> --
>>>
>>> If you still find run-time errors, please
>>> post to the list a minimal code example
>>> that displays the problem.
>>>
>>>
>>>     Thanks
>>>
>>>
>>>         Luis
>>>
>>>
>>> ----------------------------------------------------------
>>> 2010/6/11 Dženan Zukić <dzenanz at gmail.com>
>>>
>>>> Hi everyone,
>>>>
>>>> I am running into some crashes in
>>>> \InsightToolkit-3.18.0\Code\Common\itkLinearInterpolateImageFunction.txx
>>>> (line 150), and I wonder could it be due to recent move to consistent use of
>>>> centered pixels (physical coordinates going from -spacing/2 to
>>>> (n-1)*spacing+spacing/2, instead of 0 to n*spacing). Did anyone have
>>>> encounters with similar problems? More specifically, does
>>>> TransformPhysicalPointToContinuousIndex in
>>>> \InsightToolkit-3.18.0\Code\Common\itkImageBase.h honor the centered pixel
>>>> coordinates?
>>>>
>>>> Regards,
>>>> Dženan
>>>>
>>>> _____________________________________
>>>> 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.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
>>>>
>>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20100617/033ae706/attachment-0001.htm>