PPT Slide
Face: 256x256 data points (It took several days to get the N=243 model)
Computationally very slow. (Especially since a full optimization is run for each point in the list before selecting the one split.)
The new point chosen is restricted to starting off from an existing point.
Not usable in the case that there is no normal data (as you might be modeling the outside instead of the inside.)
The field functions do not have compact support so this method does not really allow for local control.