Hey Michal, I'll see what I can do about SLICO, although I might need some help with

Cython.

My first experience with Cython was making scikit-image's SLIC 3D compatible, so it would perhaps be a great entrypoint for you as well! =) It now uses memoryviews instead of pointer arithmetic (thanks Stefan for pushing me to do this! ;), so I think it should be more accessible even! =) I also noticed that the scikit-image version doesn't do the seed

perturbation step in preprocessing, and doesn't seem to do the connected components step in post-processing. Was there any reason for omitting them?

Nope. =) A powerful combination of simplicity and laziness. =P In my own problems, SLIC with reasonable compactness and sigma parameters produced very good results, so I never felt compelled to port that code. Andy, on the other hand, got worse results with sigma > 0. I think it's because I'm looking at objects that are intrinsically fuzzier (cell nuclei). Anyway, adding the post-processing would also be appreciated! Juan.