[Insight-users] about TPS and EBS

[Luis Ibanez]

Hi Zhangzhijun,

ITK ThinPlateSpline and ElasticBody splines are implemented
using a common framework under the class itk::KernelTransform.

This classes have not been optimized for speed yet.  So, it expected
that VTK's version could be running faster.  We are interested in
pushing performance to the same level anyways      :-)


However,
1) The numerical results should be the same between itk and vtk
2) The time of execution for itk::TPS and itk::EBS should be the
    same because the Point transformation is done by a common
    code in KernelTransform. If you found EBS to be slower than
    the TPS... there should be a problem with the code or with the
    use of the class.

Did you made this comparison in a compact application that you
could send to us ?

Otherwise we can make some tests in our side in order to identify
time consumption. One of the suspect things done in KernelTransfom
is the common extraction of submatrices from the WMatrix, which
can be a costly operation.


I'll look in to that.


Thanks

Luis

========================================================

zhangzhijun wrote:

>  
>
> Dear all:
>
>         I want to use the thin plate spline and elastic body spline 
> implementation in itk
>
> but I found that the speed of the thin plate spline is very slow, much 
> slower than
>
> that in VTK and the ebs is even slower, also, the result is different 
> with VTK.
>
> I use totally 80 point for a 181*217*181 data warp, I dont' know 
> whether the number
>
> of points for the warp is ok.
>
>  
>
>  
>
> regards,
>
> zhang zhijun
>