[SciPy-User] watershed question
Thu Apr 29 14:38:26 CDT 2010
Hi Stéfan and Zach,
thank you for your answers. So it seems that
ndimage.watershed_ift is quite buggy, maybe some warnings should be
added to its docstring? We can't afford people spending too much time
trying to use the function if it doesn't work.
Using the cellprofile/cpmath package is a neat trick, I tried it
and it works perfectly. It even works for 3-D arrays (using the
fast_watershed function)! Too bad that it's GPL-licensed and it's not
possible to integrate the code in the image processing scikit :(.
On Wed, Apr 28, 2010 at 01:18:11PM -0400, Zachary Pincus wrote:
> > Unless I'm also missing something obvious, the code returns an invalid
> > result. I even adjusted the depths of the two "pits", but always one
> > region overruns the other---not what I would expect to happen. I
> > haven't delved into the ndimage code at all, but I wonder weather we
> > shouldn't implement one of the simpler algorithms as part of
> > scikits.image.segment for comparison?
> Cellprofiler has a watershed algorithm, I believe. And like most of
> the cellprofiler stuff, the implementation seems pretty high-quality
> and well-thought-out.
> I wound up extracting the cpmath sub-package, and (after a few
> setup.py changes) it works great standalone with just scipy and numpy
> as dependencies:
> SciPy-User mailing list
More information about the SciPy-User