Selecting A Partitioning Plane In A BSP Tree

Well in kD-trees, the cost function (assuming surface area heuristic) between 'split plane candidates' placed on vertices of input polygons is constant. I'm not sure if this is the same for generic BSPs though.

That said, I don't think anyone tried to do this. In fact, normally you don't even consider all candidates, just a random subset. The result will be close to optimal.

eh?

http://www.theomader.com/public/Portfolio.html uses some extra crazy math.

Some use the polygon plane with the best score, some use only axis aligned planes...

Usually you start with the most axis-aligned partition planes and then go from there. You might be able to come up with some sort of cost function that determines the number of splits you'd make and then attempt to minimize them.