[Nipy-devel] ITK question

[Gael Varoquaux]

On Fri, Aug 29, 2008 at 02:32:31PM -0700, Karl Young wrote:
> I've been casting about writing some scripts to automate some of the 
> jobs that people do in our lab using ITK and it occurred to me to check 
> in re. the latest thinking on the relationship between ITK and NiPy. I 
> certainly understand the concerns about saddling potential users with 
> such a monstrosity but on the other hand you get an awful lot (and maybe 
> even a useful framework for "easily" adding new segmentation, 
> registration,... tools).

There certainly is a huge value in ITK, and I would personaly be very
interested in these scripts, just so that I could use them to learn how
you use ITK, but also because we could improve them, and reduce the
amount of wheel reinvention.

> I was impressed with the Mayavi demos at SciPy 08 and recall that last
> year Prahbu and Gael were expressing similar concerns about VTK (i.e.
> saddling their users with the monstrosity of VTK) but seem pretty
> unconcerned about that now.

Well, I wouldn't say this :). I would say that we accept that Mayavi
wouldn't be anything without VTK, and thus we accept our dependency. We
try to make the VTK heritage as easy as possible to our users. However,
the dependency on VTK does make it hard for us to get the software to the
users (namely because we can't ship a binary of Mayavi unless we ship a
binary of VTK.

> So I was wondering if their experience had any effect on thinking about
> NiPy/ITK.

I have been spending the week with the nipy core team in Berkeley, trying
to plug the nipy images as easily as possible in Mayavi for good
visualisation. This is good, and I am very happy about this, as it will
make my life easier for visualisation. However, I would not want to make
it impossible for people to use the nice image class, with its IO, and
the interpolation function, without the Mayavi/VTK added sugar for
visualization.

I think it would be fantastic to have a well-separated class of
algorithms, that plug in nipy as well as possible, using ITK for
registration. Hopefuly the nipy core-devs can help you finding a sweet
spot in the codebase, where you can integrate this ITK code you are
talking about.

My 2 cents,

Gaël