Author Topic:   getting started
posted October 31, 1999 07:26 PM            

well, i finished up my own "brute-force" landscape renderer, whose sole optimization is view frustum culling. this gave me something on the order of the Myth renderer (ie, your face jammed down into the dirt, no way to look at the horizon). pretty bad.

now i'm trying to get started implementing a landscape renderer roughly based on ROAM.

i seem to be getting stuck a little early on, so i'm hoping someone here can get me started - the later bits of the algorithm make sense, but the beginning confuses me.

"Let T = the base triangulation"

by base triangulation, i'm assuming they mean the coarsest level of detail -- so if my "world" is a square grid, the base triangulation is simply a diamond, or 2 triangles.

"For all triangles in T, put triangle in Q"

i understand this, assuming my understanding of the first step is correct. i would basically be putting the 2 triangles in my diamond into the queue.

"While T is too small or inaccurate { "

i understand the meaning of T being too small, ie, the triangulation is very coarse, i want to render more polys. is the "size" of T simply the size of the queue, however? i'm a bit more confused about the accuracy measurement, though. how do i get an error metric on the entire triangulation? is it the sum of the errors for every triangle in T?

thanks for any help!



LDA Seumas
posted November 13, 1999 06:57 AM           
Sorry about the delay...

I think what that means basically is that while you have less triangles than you would like, or while the largest error in any of the triangles is bigger than you would like, keep refining the landscape more. The "error" for T would probably be best expressed as the largest screen space error of any of the triangles in T, not as an additive quantity. If all of the errors are very low, it doesn't matter that there might not be many triangles used, for example.

-- Seumas McNally, Lead Programmer, Longbow Digital Arts