Terrain LOD
Author
Could anyone recommend some good [easy] articles on Terrain LOD, particularly on ROAM?
I reccomend you forget about LOD for now, instead, add a quad tree + frustum culling and if that TRUELY, doesnt offer the performance you want, THEN look into LOD, first of all ROAM is very outdated and on modern cards will most likely HURT performance, there are very few that will offer any sort of speed increase as the computation time to teselate meshes tends to be longer now than to just brute force it in many cases (never completely brute force though) Id love to see a benchmark of ROAM, geomipmaps, geoclipmaps, CLOD, rotteger's and brute force, and brute force + quadtree running on say, a radeon 9600, my moneys on brute force + quad tree or maybe mipmaps. Having said that, dont worry about LOD till you are either a) in need of a performance boost (in which case i reccomend you SAVE your old code so when you find it was faster, you can go back to it ;-) ) or b) bored
hope that helps
-Dan
hope that helps
-Dan
Author
Ah! The error of my ways.
Thanks -- but another question -- what about simplifying the mesh where additional polys arent needed (Ex: on a flat area of land). Would that be efficient on today's graphics cards? (This is just a curious question -- I think I'll just implement quadtree culling).
Thanks -- but another question -- what about simplifying the mesh where additional polys arent needed (Ex: on a flat area of land). Would that be efficient on today's graphics cards? (This is just a curious question -- I think I'll just implement quadtree culling).
http://www.vterrain.org/LOD/precomputed-tins.html
triangulated irragular networks, basically, if theres a few coplanar triangles, you merge them, heck i think most of those schemes merge ones that are "nearly" coplanar as well, but really, as far as i can see, these dont go very well with quad trees, as your quad tree might be chopping one of these huge triangles down the center, which can be resolved, but you could end up with more triangles in the end that way. What you could do, is triangulate INSIDE each quad tree node, that way you could avoid tesselating across a node border and then just end up chopping it in half anyways.
hope that helps
-Dan
triangulated irragular networks, basically, if theres a few coplanar triangles, you merge them, heck i think most of those schemes merge ones that are "nearly" coplanar as well, but really, as far as i can see, these dont go very well with quad trees, as your quad tree might be chopping one of these huge triangles down the center, which can be resolved, but you could end up with more triangles in the end that way. What you could do, is triangulate INSIDE each quad tree node, that way you could avoid tesselating across a node border and then just end up chopping it in half anyways.
hope that helps
-Dan
Quote:Original post by Ademan555
I reccomend you forget about LOD for now, instead, add a quad tree + frustum culling
What are you talking about? There are very few instances where LOD will not immensly help terrain (or any object at least). It's true, if you are satisfied with a medium-detail mesh with frustum culling, then so be it. But if you want any kind of detailed mesh, LOD is gonna be a huge help with this.
I recommend looking at Chunked LOD, it's based off of quadtrees and is pretty easy to implement. Of course you will also need a way to simplify the mesh, like you asked in your second post
No actually i would like to invite you to test it yourself, in most every case LOD tends to be slower on modern cards
I have. I have implemented Chunked LOD and it enables me to view huge stretches of land at a constant 60+ fps with great detail near the viewer. If I kept the same high detail everywhere, it easily slows down to between 1 and 5 fps simple because of the amount of vertices.
edit: I think you're right about ROAM though, but I'm talking generally about LOD.
edit: I think you're right about ROAM though, but I'm talking generally about LOD.
Really? Interesting, i had actually heard bad things about chunked LOD specifically, but maybe they were misguided, which papers/resources did you work off of? Oh, and also, what card do you have?
thanks
-Dan
thanks
-Dan
Yeah, I don't understand why you heard bad things about it. Can anyone else confirm this? You waste so many vertices in the background if you don't deal with it...
I think this was the main paper: http://tulrich.com/geekstuff/sig-notes.pdf
but I also combined a lot of things from other stuff I'd read before. I just wanted to get something simple done, because you are right, complex LOD schemes can end up taking more time to compute. But it's good to help the GPU a little bit.
Chunked LOD just involves traversing down a quadtree and testing a couple boxes in the frustum.
I have a Radeon 9600 Pro.
I think this was the main paper: http://tulrich.com/geekstuff/sig-notes.pdf
but I also combined a lot of things from other stuff I'd read before. I just wanted to get something simple done, because you are right, complex LOD schemes can end up taking more time to compute. But it's good to help the GPU a little bit.
Chunked LOD just involves traversing down a quadtree and testing a couple boxes in the frustum.
I have a Radeon 9600 Pro.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
