[Insight-users] More compilation problems with VectorImage

[Uwe Köhler]

Dear Karthik,

many thanks for your quick reply. This goes a bit in the direction of a bug I 
submitted earlier, where it was unclear to the user that a crash occured due 
to an unsigned pixel type in an image. When looking at the documentation it 
is rather supprising that a filter for vector valued images does not work von 
Vector images. 

I can see that for a number of filters it is not clear what they should do 
when faced with vector valued images, but many should be able to cope (apply 
the filter to every component found?). In that case the number of components 
also does not need to be fixed at compile time. It should be enough to be 
able to get the number of components. I am quite clearly missing a very 
important point here. I have used ITK for a number of years now, but trying 
to introduce people to it, the fact that the there is no documentation of the 
compatability of filters with types of input images always seems to pose a 
rather large problem. Could this be documented for every filter and adaptor 
(see my post about VectorImageToImageAdaptor and Iterators).

I know that it fails already during compilation, however, I would prefer not 
to find out the hard way. 

Cheers

Uwe

Am Freitag, 10. März 2006 17:29 schrieb Karthik Krishnan:
> On Fri, 2006-03-10 at 09:53 +0100, Dr. Uwe Köhler wrote:
> > Following up on my earlier email, I got further compilation problems
> > related to the use of VectorImages:
> >
> > I was unable to use VectorRescaleIntensityImageFilter with a
> >
> > /usr/local/include/InsightToolkit/BasicFilters/itkVectorRescaleIntensityI
> >mageFilter.h:37: error: ‘Dimension’ is not a member of
> > ‘itk::VariableLengthVector<int>’
>
> Clearly this class can be used *only* with Images with pixel types that
> have the PixelType::Dimension typedef available.
> such as itk::Vector where the length is a fixed value
>
> That's the way the class was written. There is nothing that can be done
> to change it, without breaking the API. Its easy to write a filter that
> does the job, but this filter will not do.
>
> Maybe that's an another reason for Concept Checking in ITK just like
> Boost.
>
> > Who can tell me which filters work with VectorImage or do I have to try
> > them out?
>
> Personlly, I've used a few filters I needed to in the past, such as
> RecursiveGaussianFilter for smoothing etc..
>
> As before, filters that do not make an assumption that the Dimension is
> fixed, such as the ones you just tried, and filters that do not assume
> that pixels are scalar should generally work just fine..
>
> > Cheers
> >
> > Uwe