[ITK Community] Bug in DefaultVectorPixelAccessor
Brian Helba
brian.helba at kitware.com
Tue Mar 11 16:59:01 EDT 2014
Hi Jan,
Note that you can update ITK's documentation via the web, without having to
bother with setting up Git locally. Class pages have an "Edit comments"
link (I'm working to make this more obvious), that allow you to suggest
changes to comments / documentation. The changes can then be reviewed by
the community via Gerrit [1] before being merged.
For DefaultVectorPixelAccessor, the page is at [2]. Any improvements you
can make would be welcome
[1] http://review.source.kitware.com
[2]
http://www.itk.org/editdoc/editcomments.php?file=itkDefaultVectorPixelAccessor.h
Best,
Brian
On Tue, Mar 11, 2014 at 10:54 AM, Jan Ehrhardt
<ehrhardt at imi.uni-luebeck.de>wrote:
> Hi Brad, hi Matt,
>
> thanks! That is a very good solution. I did not realize that an operator=
> does the job and I assumed it=ImageRegionIterator(myfeatureImage2, region)
> will not work. Sorry.
>
> So, this is the solution for my problem. Thanks to all for your help.
>
> To come back to my initial post about the DefaultVectorPixelAccessor. I
> think, the documentation of this class should be improved to avoid
> confusion about the parameters of the Get() and Set() method. As far as I
> see, the correct way to access an vector at a pixel offset is not
> intuitive:
> accessor.Get(*(pBuffer+offset), offset);
>
> Best regards,
> Jan
>
>
>
>
> On 03/11/2014 02:18 PM, Bradley Lowekamp wrote:
>
> OK,
>
> Then another approach is to take advantage that the iterators are
> default constructible. Something like this:
>
> ImageRegionIterator it2;
> if (myfeatureImage2)
> {
> it = ImageRegionIterator(myfeatureImage2, region);
> }
>
> There are a lot of options.
>
> Hope one works for you,
> Brad
>
> On Mar 11, 2014, at 9:04 AM, Jan Ehrhardt <ehrhardt at imi.uni-luebeck.de>
> wrote:
>
> Hi Brad,
>
> yes this is an alternative way, but the code inside the loop has approx.
> 200 lines (is quite complicated stuff) and I'm still testing and in your
> solution, I have to maintain the code four times...
>
> Best regards,
> Jan
>
> On 03/11/2014 01:55 PM, Bradley Lowekamp wrote:
>
> Hello,
>
> You are right that the GetPixel methods are slow, but that's why
> iterators are recommended.
>
> Currently you are trying to have you "if( featureImage2 )" in side the
> per-pixel loop. Not only does this cause you iterator initialization issue
> it also is slower of have this if statement inside the loop.
>
> Here is an alternative way to construct your logic:
>
>
> https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/Filtering/ImageFilterBase/include/itkBinaryFunctorImageFilter.hxx#L224-L297
>
> Hope this help,
> Brad
>
> On Mar 11, 2014, at 8:34 AM, Jan Ehrhardt <ehrhardt at imi.uni-luebeck.de>
> wrote:
>
> Hi Matt,
>
> yes, I currently use the ITK iterators, but I have the following problem.
>
> Standard code looks like:
>
> itk::ImageRegionIterator<ImageType1> it( radius, myImage, region );
> itk::ImageRegionIterator<ImageType2> featureIt1( radius, featureImage1,
> region );
> itk::ImageRegionIterator<ImageType3> featureIt2( radius, featureImage2,
> region );
>
> it.GotoBegin();
> featureIt1.GotoBegin();
> featureIt2.GotoBegin();
>
> while(!it.IsAtEnd())
> {
> // do some complicated processing stuff with or without feature1 and/or
> feature2
>
> ++it;
> ++featureIt1;
> ++featureIt2;
> }
>
> I only know at runtime, whether featureImage1 and/or featureImage2 are
> available. If
> featureImage1 or featureImage2 is NULL, a segmentation fault occurs.
>
> Therefore, I have currently implemented:
>
> itk::ConstNeighborhoodIterator<ImageType1> it( radius, myImage, region );
>
> it.GotoBegin();
>
> while(!it.IsAtEnd())
> {
> currentIndex=it.GetIndex();
> if( featureImage1.IsNotNull())
> {
> // use featureImage1->GetPixel(currentIndex) to process feature 1
> }
> if( featureImage2.IsNotNull())
> {
> // use featureImage2->GetPixel(currentIndex) to process feature 2
> }
>
> // do some other stuff
>
> ++it;
> }
>
> The Image::GetPixel() method is quite time-consuming and therefore I look
> for a more efficient method
> to implement this code (without writing four different versions).
>
> Any suggestions?
>
> Best regards,
> Jan
>
> On 03/11/2014 06:12 AM, Matt McCormick wrote:
>
> Hi Jan,
>
> To iterator through an Image or VectorImage, use the Iterators as
> described in the Software Guide [1]
>
> HTH,
> Matt
>
> [1] http://itk.org/ItkSoftwareGuide.pdf
>
>
> On Thu, Mar 6, 2014 at 8:23 AM, Jan Ehrhardt <ehrhardt at imi.uni-luebeck.de>wrote:
>
>> Hi folks,
>>
>> I want to iterate through a vector image without an iterator. My approach
>> was:
>>
>> MultiChannelImageType::AccessorType accessor =
>> pImage->GetPixelAccessor();
>> MultiChannelImageType::InternalPixelType* pBuffer =
>> pImage->GetBufferPointer();
>>
>> for( OffsetValueType i = from; i < to; ++i )
>> {
>> vecPixel = accessor.Get(*pBuffer, i);
>> }
>>
>> but the results are not correct (see bug(?) below). If the bug(?) in
>> DefaultVectorPixelAccessor is corrected ( m_OffsetMultiplier = l ) the
>> behaviour of standard itk iterators change. I assume, I don't understand
>> the parameters of DefaultVectorPixelAccessor::Get() correctly.
>> How can I use the DefaultVectorPixelAccessor correctly?
>>
>> best,
>> jan
>>
>>
>> On 03/06/2014 12:25 PM, Jan Ehrhardt wrote:
>>
>>> Hi folks,
>>>
>>> there is a bug in DefaultVectorPixelAccessor (ITK 4.4.0):
>>>
>>> void SetVectorLength(VectorLengthType l)
>>> {
>>> m_VectorLength = l;
>>> m_OffsetMultiplier = ( l - 1 );
>>> }
>>>
>>> DefaultVectorPixelAccessor(VectorLengthType l)
>>> {
>>> m_VectorLength = l;
>>> m_OffsetMultiplier = l - 1;
>>> }
>>>
>>> OffsetMultiplier should be m_OffsetMultiplier = l;
>>>
>>> Best regards,
>>> Jan
>>>
>>>
>> _______________________________________________
>> Community mailing list
>> Community at itk.org
>> http://public.kitware.com/cgi-bin/mailman/listinfo/community
>>
>
>
> _______________________________________________
> Community mailing list
> Community at itk.org
> http://public.kitware.com/cgi-bin/mailman/listinfo/community
>
>
>
>
>
>
> _______________________________________________
> Community mailing list
> Community at itk.org
> http://public.kitware.com/cgi-bin/mailman/listinfo/community
>
>
--
Brian Helba
Medical Imaging
Kitware, Inc.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/community/attachments/20140311/69991e16/attachment-0002.html>
More information about the Community
mailing list