|
Author Topic:   Priority Computation in ROAM
Oiler
Member
posted September 29, 1999 08:00 PM            
Hello,

I have my ROAM implementation almost complete but i'm stuck on the priority computation part. My implementation is
straight from the paper except I support
paging of bin_trees.

The paper has a formula that goes dist(v)=| |s(v)-sT(v)| |. Do I use this formula for distortion? I assume not as it doesn't use the wedgie thicknesses i computed. Then there is the second large formula that has the square root in it. But I thought that one was for computing an upper bound.

I'm confused can anyone help me clarify this section.

Thanks in advance
Oiler

IP:

LDA Seumas
unregistered
posted October 02, 1999 01:57 AM           
Personally I use a very simple formula for checking whether each tri needs to be split. Basically I just take the pre-computed Variance for the tri, multiply by the distance from the camera, and test whether that is larger or smaller than a pre-computed Variance Limit (which has been pre mangled to be more in line with the result of variance * distance).

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

IP:

Oiler
Member
posted October 04, 1999 12:03 AM            
Ok I understand what your saying. And from looking at your test it works pretty well

But would you have any idea how the actual papers distortion formula works. I just emailed the author so I'm hoping for a reply soon (He helped me out once before).

Thanks
Oiler

IP:

LDA Seumas
unregistered
posted October 04, 1999 06:54 AM           
Not exactly... I've never tried to implement ROAM 100% as specified in the paper. All the computational exactness felt like overkill for a mostly horizontally viewed game egine.

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

IP:

Oiler
Member
posted October 05, 1999 01:25 AM            
Yeah I got the reply from the author today and geez does he go on....and on....and on. I think I may rewrite his doc when I'm done so someone can actually understand it.

Anyway he said that he only does the expensive computation shown in the paper on tris that are close. Otherwise he does what he calls a "first-order" approximation.

I'm kinda liking the way you handle it now though as it sounds ALOT cheaper.

Thanks
Oiler

IP: