2002/11/11 19:09:44 PST by zorbathut [0/4] Awards: 2 from Dev |
3Der: *bad* polyreduction can be done in most 3d programs, including MAX :) I don't know any program that uses this algorithm, however, it looks beautiful. The artists had done an 8000 tri model and a 3000 tri model - running the polyreduction on the 8000 tri model brought it down to about 2200 tri, and it looked the same as the 3000 tri model. It's quite possible to write a program to do better polyreduction than a human can (basic explanation: humans require models that are still maintainable, and have logical triangle placement. Computers are under no such constraint.)
And there are a lot of companies doing the polyreduction thing - advantages include the fact that you can customize the model format for whatever hardware you like, *and* you can generate a beautiful bumpmap at the same time.
So actually, making a 50,000 poly model and reducing it to 2,000 polys might produce *better* results than just making a 2,000 poly model. And, in fact, making a million-poly model might be even better.
(I'm told that if you want to write a fantastic 10-page paper, you should write a 20-page paper, then reduce it to 10 pages. I think this is the same principle :P)
temporal: dynamic LOD's neat, but there are some good arguments for pregenerating the meshes. Basically, you can throw a lot more CPU power at it. If you're doing it in runtime, you need a fast algorithm - if you're doing it beforehand, you can have your computer spend an hour on each model without a problem.
Note that the game Messiah was trying to do beautiful dynamic LOD from enormous models. It looks junky and didn't run well - they had to use a quick-and-dirty reduction algorithm and it *still* wasn't fast enough.