[vtk-developers] vtkImageData, C versus Fortran ordering

[Andras Lasso]

> do you think there are any major performance issues with supporting arbitrary orientations?

So far it seems that most of the cases there is no performance impact at all. Most of the cases at the lowest level a 4x4 homogeneous transformation matrix converts between voxel and physical coordinates already, so the difference is just where you can have non-zero elements in the rotation component of the matrix (in the diagonal only or everywhere).

Andras

-----Original Message-----
From: Marcus D. Hanwell [mailto:marcus.hanwell at kitware.com] 
Sent: Wednesday, July 26, 2017 3:52 PM
To: Andras Lasso <lasso at queensu.ca>
Cc: Berk Geveci <berk.geveci at kitware.com>; VTK Developers <vtk-developers at vtk.org>; David Gobbi <david.gobbi at gmail.com>
Subject: Re: [vtk-developers] vtkImageData, C versus Fortran ordering

On Thu, Jul 20, 2017 at 11:49 PM, Andras Lasso <lasso at queensu.ca> wrote:
>
> > My approach with this kind of change has been to create a new subclass of vtkDataSet. If you add some meta-data to vtkImageData that is then ignored by all or most of algorithms, you will end up with confused users.
>
> We’ve been working on this for several years, as a background task. Adding support in VTK for reading, processing, and displaying arbitrarily oriented images would simplify many VTK-based medical image computing applications. Some information about this effort is available here:
>
> https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdocs.
> google.com%2Fdocument%2Fd%2Fe%2F2PACX-1vQiHT2IpeCtfpphJ9vFtq_dQ3dJHsay
> ZTcvJZUMWM9lhJdItd1qQed0449pOpYIfE3Bqgd2uZaXN4iR%2Fpub&data=02%7C01%7C
> lasso%40queensu.ca%7C8d0890b2989445d73f0f08d4d45fbc59%7Cd61ecb3b38b142
> d582c4efb2838b925c%7C1%7C0%7C636366955023304566&sdata=6jLorpDvFzq6aciZ
> QnLJRaooJSrGsircXqrC1rCZTk4%3D&reserved=0
>
> I know that it is a sensitive topic and a large number of files are involved (around 400 core classes + test/examples), but in many cases the changes are trivial (changing a few lines in a class to indicate that it can accept oriented image) and it is not necessary to update all classes at once, but gradually, as the need emerges.
>
> I would love to hear from anyone who would be interested in having orientation support, have any comments or suggestions, or might be willing to contribute.
>
Thanks for the link, there is certainly some overlap (although this goes a little further) with what I was thinking of. I think it would be good to investigate the feasibility of this approach, do you think there are any major performance issues with supporting arbitrary orientations?

Marcus