[Insight-developers] itk vs vnl / Vector & Matrix

[Stephen R. Aylward]

Hi,

I think Luis' last point - that we pass itkVectors is nearly a
requirement, not just a preference - to keep the API consistent
looking.  Right?

Stephen

Luis Ibanez wrote:
> 
> Robert,
> 
> Well there is no clear cut rule on this...
> 
> The following could serve as basic guidelines:
> 
> 1) itkVector   defines its dimension at compile time
>    vnl_vector  defines its dimension at run time
> 
>   as a consequence vnl_vector has an overload in
>   memory of 8 bytes with respect to the itkVector.
> 
>   So, if you are planning to create millions of them,
>   your are better off with itk::Vector.
> 
>   On the other hand, vnl_vector is the native type for
>   vnl mathematical operations...  so, if you care more
>   about math, vnl_vector is a better option. Math in
>   itkVector is limited to arithmetic ops and product by
>   an itkMatrix. Scalar products and normalization are
>   defined for the itkVector too.
> 
> 2) itk::Matrix and vnl_matrix are more closely related
>    the itk::Matix has a vnl_matrix_fix internally.
>   Both have dimension defined at compile time (though
>   vnl_matrix_fixed is just hidding a run time assignment)
> 
>   itk::Matrix was added as a convenience wrapper of the
>   vnl_matrix in order to avoid conversion between itkVectors
>   and vnl_vector just for performing a vector*matrix product.
> 
>   It was a bit shameful to require ten lines of code for
>   multiplying a vector by a matrix ...    :-)
> 
> We have tryied to pass itk::Vector and itk::Matrix around
> in the API of itk classes but this is just a preference.
> 
> Luis
> 
> ----------------------------------------------------------------
> 
> Robert J Tamburo wrote:
> 
> >On the subject of itkPoint ......... when should someone use itkVector vs.
> >vnl_vector or itkMatrix vs. vnl_matrix. I kind of just "wing" it based on
> >what functionality I need (i.e. I'll use vnl_vector if I need dot products),
> >but I'm sure there's some logic in choosing the more appropriate class.
> >
> >Rob
> >
> 
> _______________________________________________
> Insight-developers mailing list
> Insight-developers@public.kitware.com
> http://public.kitware.com/mailman/listinfo/insight-developers

-- 
===============================================
Dr. Stephen R. Aylward
Assistant Professor of Radiology
Adjunct Assistant Professor of Computer Science
http://caddlab.rad.unc.edu
aylward@unc.edu
(919) 966-9695