Run through the existing algorithm, then get the winding order of the border vertices and choose a likely starting point (orthogonally aligned, symmetry or closest radius or edge, or pre-calc and measure least distortion somehow). Then iterate in winding order, and rotate each vertex about the centre so the angle with the…
Looking good. I've been working on something similiar, and like yourself using a similar algorithm, i.e using the avg centre and re-positioning the vert along a vector from this centre a distance of average radius away and intersecting a normalized plane. Just wondering, have you looked at forcing the initial border…