dolfin team mailing list archive

[Kent Andre <kent-and@xxxxxxxxx>]

On to., 2009-01-29 at 14:59 +0000, A Navaei wrote:
> 2009/1/29 Kent Andre <kent-and@xxxxxxxxx>:
> > On to., 2009-01-29 at 14:03 +0000, A Navaei wrote:
> >> 2009/1/29 Kent Andre <kent-and@xxxxxxxxx>:
> >> > On ti., 2009-01-27 at 12:34 +0000, A Navaei wrote:
> >> >> Kent: Luca is right about the lack of wrapping for double type. I was
> >> >> not aware that you were using the debian package available in ubuntu,
> >> >> that binary (which has been nicely packaged) is now a bit out of date
> >> >> and uses an old design of wrapitk while only the types that Luca
> >> >> mentioned are wrapped. If you have access to a fast server, compiling
> >> >> wrapitk is not that bad, simply turn python wrapping and double type
> >> >> supporting on in the cmake gui, and you're good to go. If you turn the
> >> >> cpack debian option on, it will also automatically generate a ready to
> >> >> use debian package.
> >> >>
> >> >
> >> >
> >> >
> >> > So I have to used the svn version of both ITK and wrapitk ? Or can I
> >> > download ITK 3.10 and compile with doubles. Anyway, I guess I the image
> >> > pixels are of integer type and I don't need to convert these to
> >> > doubles.
> >>
> >> itk 3.10 is buggy and some classes will not compile when wrapping,
> >> please use the development version and follow the instructions in
> >> http://code.google.com/p/wrapitk/ . (Note that when you compile itk,
> >> only a small fraction of the library is actually compiled, the rest
> >> are simply templates which need to be instantiated. wrapitk
> >> instantiates and compiles every single class, and that's where the
> >> buggs are discovered.)
> >>
> >> It doesn't matter if your input images are not of double type. Since
> >> your output images must be of double type, in order to connect them to
> >> dolfin, you must have double type wrapping on. If you want to use
> >> numpy as a glue, you might be able to cast the interface numpy array
> >> from int to double. Let me know if there are any problems.
> >
> >
> > So I modified your code slightly to get it work using ubuntu packages,
> > (I will go to the newer version  ... :) )
> >
> > from dolfin import *
> > from numpy import *
> > import itk
> >
> > dim = 2
> > inType = itk.Image[itk.UC, dim]
> >
> > reader = itk.ImageFileReader[inType].New(FileName="bowling.jpg")
> > reader.Update()
> >
> > itk_py_converter = itk.PyBuffer[inType]
> > image_array = itk_py_converter.GetArrayFromImage( reader.GetOutput() )
> >
> >
> > I have a 150x150 pixel images (in jpg) and running the above code  takes
> > like 10 seconds ? What is wrong ?
> >
> 
> Since I don't have the old version installed, I am unable to
> re-produce the problem. itk to numpy is simply done by pointer
> assignment (http://code.google.com/p/wrapitk/source/browse/trunk/ExternalProjects/PyBuffer/itkPyBuffer.txx#34),
> there is no loop involved in this. (Of course I cannot speak for the
> old version. Note that, that version uses another design which is
> completely different to our new wrapping design).
> 
> I recommend jumping to the latest version. If you are not in a hurry
> and compilation is a problem, you can use the debian package I'm going
> to create sometime soon.
> 
> 
> -Ali

Let me know when you have created debian packages. It seems I have to
rebuild everything, even cmake. Compilation is in principle no problem, 
but I already compile a decent number of packages on a regular basis and
being non-bleeding edge on packages I don't know well seems tempting. 

I might download and compile, but not today. 

Kent