ITKv5 Streaming/ImageSinc Impact

[blowekamp (Bradley Lowekamp)]

@Niels_Dekker

We are still busy here, trying to make elastix compatible with ITK5. We just got it compiling with a very recent version of ITK from the master, tag 9e12266 of May 9, 2019. But with this ITK version, elastix still has test failures, that need to be solved. And now, with the streaming redesign, elastix does not even compile anymore. Even though we still have ITK_LEGACY_REMOVE=OFF , for elastix.

This sounds like it is related to my recent changes. Please let me know if there is anything I can do to help with the problem you have encountered.

[Niels_Dekker]

Thanks for your extensive reply, @matt.mccormick!

@blowekamp

This sounds like it is related to my recent changes. Please let me know if there is anything I can do to help with the problem you have encountered.

Thanks for offering, @blowekamp .Indeed, the recent compilation errors I get from elastix are related to your work on Add streaming base classes and streaming statistics by blowekamp · Pull Request #855 · InsightSoftwareConsortium/ITK · GitHub One of the issues is that some elastix classes now need to override GetRadius(). I really have to dive into those classes to figure out how to properly implement such a GetRadius(). I’ll have a closer look tomorrow…

[blowekamp (Bradley Lowekamp)]

@Niels_Dekker What class are you deriving from? Do you have a link to the problematic class?

[Niels_Dekker]

@blowekamp Sorry I still haven’t finished fixing the elastix test failures related to the Spatial Object refactoring. I’m not still building elastix with an ITK version of May 9: Merge pull request #849 from thewtex/clang-detection · InsightSoftwareConsortium/ITK@9e12266 · GitHub

At ENH: Replace elastix ImageSpatialObject2 by ITK ImageSpatialObject by N-Dekker · Pull Request #147 · SuperElastix/elastix · GitHub

Of course you could try to build elastix from that pull request (elastix pr #147) against the latest ITK master. Then you would get lots, and lots of compile errors:

f:\x\src\elx\elastix\common\itkReducedDimensionBSplineInterpolateImageFunction.h(106): error C2259: ‘itk::ReducedDimensionBSplineInterpolateImageFunction<itk::Image<float,2>,double,double>’: cannot instantiate abstract class
f:\x\src\elx\elastix\common\itkReducedDimensionBSplineInterpolateImageFunction.h(106): note: due to following members:
f:\x\src\elx\elastix\common\itkReducedDimensionBSplineInterpolateImageFunction.h(106): note: ‘itk::Size<2> itk::InterpolateImageFunction<itk::Image<float,2>,double>::GetRadius(void) const’: is abstract
f:\x\src\elx\elastix\common\itkReducedDimensionBSplineInterpolateImageFunction.h(106): error C2259: ‘itk::ReducedDimensionBSplineInterpolateImageFunction<itk::Image<float,2>,double,double>’: cannot instantiate abstract class

etc.

Missing GetRadius() overrides is just one of the issues. I think elastix BeforeThreadedGenerateData() and AfterThreadedGenerateData() must also be replaced by BeforeStreamedGenerateData() and AfterStreamedGenerateData() overrides. And that’s not all… I’m not sure, I’m just guessing. So yes, your help is appreciated!

[blowekamp (Bradley Lowekamp)]

Ahh… This error has to do with the changes to the ResampleImageFilter to support streaming and the additional Interpolator::GetRadius method that is now needed. This changes occurred in this PR:

ENH: Expand ResampleImageFilter requested region by interpolator domain

[fbudin (Francois Budin)]

I had a similar error the other day while updating the ITK remote module https://github.com/InsightSoftwareConsortium/ITKGenericLabelInterpolator . This is something we should fix…

[blowekamp (Bradley Lowekamp)]

@fbudin

I had a similar error the other day while updating the ITK remote module GitHub - InsightSoftwareConsortium/ITKGenericLabelInterpolator: The basic idea of this generic interpolator for label images is to interpolate each label with an ordinary image interpolator, and return the label with the highest value. This is the idea used by the itk::LabelImageGaussianInterpolateImageFunction interpolator. Unfortunately, this class is currently limited to Gaussian interpolation. Using generic programming, our proposed interpolator extends this idea to any image interpolator. Combined with linear interpolation, this results in similar or better accuracy and much improved computation speeds on a test image. . This is something we should fix…

If compatibility is enabled we could add a base implementation which

InsightSoftwareConsortium/ITK/blob/073f17e0209bd6c7605299193c50c87865660e23/Modules/Core/ImageFunction/include/itkRayCastInterpolateImageFunction.h#L168-L176

.

[blowekamp (Bradley Lowekamp)]

@Niels_Dekker

. I think elastix BeforeThreadedGenerateData() and AfterThreadedGenerateData() must also be replaced by BeforeStreamedGenerateData() and AfterStreamedGenerateData() overrides. And that’s not all… I’m not sure, I’m just guessing. So yes, your help is appreciated!

I compiled elastix and got a couple errors related to ComputeImageExtremaFilter which is derived from the StatisticsImageFilter. Addressing those errors would require a copying much of the new implementation.

The ComputeImageExtremaFilter seems to only add the ability to specify a MaskedImage as a region to compute the statistics. The simplest thing to do would be to just use the LabelStatisticsImageFilter with the mask image as the label image.

[Niels_Dekker]

Thanks very much for trying to compile elastix!

Going home now, tomorrow I’ll have a more detailed look at your interesting findings!