More here.On the preprocess structure:
I use simple arrays for a 1024x1024 example--
int lmax; // finest level of resolution
float **h; // vertex height
float **e0,**e1; // max error in triangle 0/1
Note that there are half as many vertices as triangles.
I just sub-sample the h's, which is by far the simplest thing when you
to the dynamic triangulations (else what would the hieght be when eight
*different* size triangles meet at a vertex?). The wedgie thicknesses
can actually be tighter than mentioned in the paper, but then you can
get splits causing *more* error on occasion and also the view-frustum
cull can be flakey because the bounds are not nested. I use the tightest
possible wedgie bounds computed by exhaustive sampling and just deal
with the flakeyness issues (hacks and kluges, mostly).
Updated Dec 22, 1999