The GeoMap is a unified representation for segmentation results, which comprises both their geometry and topology. The GeoMap is intended to be used not only for the final segmentation result, but in particular for intermediate working states of a complex segmentation process chain. Hence, it offers operations that modify the tesselation, for instance merging two neighboring regions. This particular implementation is using a polygonal representation, and consists of a core C++ library, Python bindings, and more complex operations (and GUI components) being implemented in Python only.

A list of corresponding scientific publications can be found at: http://www.citeulike.org/user/hans_meine/tag/geomap

This repository contains ten years worth of GeoMap code.

The code was ported from boost::python to boost::python V2, from vigra/interactive to vigra/interactive2, partially to vigra/interactive3, and finally (at least much of it) to the current, released vigranumpy bindings. Furthermore, the GUI stuff was ported from Qt3 to Qt4. (And we could soon start the next port to Qt5.) The code was initially managed with CVS, then in Subversion, Mercurial, and now Git. It was always tried to preserve the history during conversion, but of course chances are that it suffered.

However, the latest porting effort to vigranumpy (and maybe even to Qt4) is not finished, i.e. not all code may be ported or tested yet.

Furthermore, during the preparation of this Git repository to be published, a lot of cruft has been removed, and many subdirectories were introduced. (Imagine all files in one directory, intertwined with at least the same number of files from one-off experiments and student thesises, and you are close to the status I was looking at this morning.)

Compilation was fixed, but the python files have not been updated and tested yet. Nevertheless, we have a public repository now, which at last makes collaboration possible! :-)

Interesting frontend bits that come to my mind (many of the often needed and more or less stable functions and classes are even documented):

• general helper functions are in maputils.py
• sub-pixel watersheds are initialized from within maputils.py
• crack-edge maps (and mid-cracks etc.) are initialized using crackConvert.py
• display GeoMaps on top of images using mapdisplay.py
• (Constrained) Delaunay Triangulation is in delaunay.py
• Chordal Axis Transform (CAT) is in delaunay.py, too
• a simple segmentation GUI / frontend is in workplace.py

Licensed under MIT license.