|
Author Topic:   Triangle budget allocation
assen
Member
posted September 14, 1999 06:25 AM            
And what do you suggest for allocating the available triangle budget among the visible patches? If you build the triangle trees for each patch top-down for each frame, you should have a rough estimate for the affordable depth (i.e. number of leaf triangles) of each tree. You have a total triangle budget for the entire frame (or just for the terrain), but how do you derive from it the budget for each tile?
I'm thinking about something along the lines of an initial error metrics computation for the entire patches, similar, but not necessarily the same as the one used for the triangle split logic.

IP:

LDA Seumas
unregistered
posted September 15, 1999 03:57 PM           
I handle the triangle budget limiting in a somewhat brute force manner. Since my engine is frame incoherent, I can't just count the number of triangles present and lock all split/merge operations to that number. The single value that determines the "quality" of the terrain is a variance test value which is used in the split check, and stays constant for an entire frame. What I do is I remember the actual number of triangles created in the last frame, and the variance quality value that produced that triangle count, and then for this frame, I throw those two values plus the user's desired triangle count into a weird, hacky function which spits out a new variance quality value to try for this frame. After much trial and error, I was able to make it converge on the desired triangle count within just a few frames, usually without overshooting. (Undershooting is better than overshooting, which causes ping-ponging.)

------------------
-- Seumas McNally, Lead Programmer, Longbow Digital Arts

IP:

Chris C
Member
posted September 15, 1999 04:44 PM         
Ah! That explains something I meant to mention a while back - during our posting regarding tri stripping, you mentioned that you used a frame incoherent tactic which puzzled me for a bit as I'd noticed that if you press '0' and then '9' on the Tread Marks test version you can actually visibly see the landscape refining itself over a period of a a couple of a second or so. I guessed that you used some frame coherent method that had strict frame time limiting, but now that explains it. That's the only time you can see the refinement though, I didn't notice it at all during normal gameplay.

Chris

------------------
Chris Cookson
Computer Science Undergraduate, University of Warwick

IP: