Segment a 3D boundary map or affinity graph using various segmentation algorithms.

Code accelerated via Cython, and 0.25-0.5x as fast as equivalent C++ code (see [https://github.com/jwayne/aleks-watershed]) with several easy optimizations still unimplemented.

My goal is to use this code to rapidly develop and test different agglomeration algorithms.

Get the dependencies:

• jpyutils [https://github.com/jwayne/jpyutils]
• tifffile (sudo easy_install tifffile)
• numpy
• g++ (to compile C++ files generated by Cython)
• optional-- Cython v0.20+ (to compile .pyx files from scratch)
• optional-- scipy (some outdated files)
• optional-- scikit-image (some outdated files)

Compile the Cython extensions. (Note that Cython is not needed The following command is assuming you want to use this package in its existing directory:

./setup.py build_ext --inplace

Run oversegment.py.

# Run `watershed_it` algorithm on a given boundary map, like that downloadable from SNEMI3D
./oversegment.py watershed_it -i boundary_map.tif -o labels.tif

# Run `watershed_it` algorithm on a given affinity graph, obtainable via scripts/bm2aff.py
./oversegment.py watershed_it -i affinity_graph.raw -o labels.tif

# Run `watershed_it` algorithm on a given affinity graph, like that used in aleks-watershed.
# There are some finicky restrictions on the filename here, though (filename must match exactly
# with aleks's benchmark affinity graph)
./oversegment.py watershed_it -i affinity_graph.raw -o labels.tif

# Help
./oversegment.py --help

Algorithms can be easily defined in oversegmenters/, and are run as an argument to oversegment.py by inputting their filename without the .py extension as the first argument. The algorithm in oversegmenters/watershed_it.py is used in the above examples.

Accepted data formats are

Other useful scripts can be found in scripts/.