The Fib25 soma surface point is not precise.

[JackieZhai]

Dear @bmatejek,

For the Fib25 example you provided, the soma surface point in "volumetric_somae/surfaces" ([1563, 1799, 2555] [z, y, x]) is not precise for 26-connected neighbors in the algorithm. And this leads directly to unsuccessful refinement, which generates the wrong acyclic skeleton in "connectomes".

Changing the point to [1569, 1790, 2559] [z, y, x] is one possible solution.

BTW, this open source repo helps me a lot, thanks :)