RE: [Algorithms] object-space bump mapping
Brought to you by:
vexxed72
From: Jon W. <hp...@mi...> - 2002-03-26 23:53:52
|
> (I sent this to Jon Watte instead of the list initially. Grumble, reply And here's the reply I sent to Brian instead of the list :-) > Ignoring platform specifics for now, but why not use a hybrid scheme in > the case of skinned characters? > There are often large sections of the mesh that are fully rigid. These > are good candidates for object space mapping. The smooth skinned > sections between the rigid sections could be handled using a local space > approach, binormals and tangents which can be skinned. Sure, you'd need > large sections or state changes would kill the idea - but for high > detail skinned characters it could be worth a try. You can bunch ALL the tangent space stuff in one buffer, and ALL the model space stuff in another. I'm assuming you're not suggesting breaking the model in one component per bone, which would cause you to go buffer submission bound. So it might not be that bad, but what would the bump map look like? You'd have parts that are in tangent space, and other parts that are in object space. You'd pay for both object space and tangent space in the implementation and art pipe. Seems complicated :-/ My thoughts have been to transform the L into tangent space, which is a no-cost switch compared to transforming the N into world space which you do with per-vertex lighting. At least I think it is, unless I'm missing something. > Another idea that might be worth trying is sending in a 'skinned' L. Ah, great minds think alike! :-) Cheers, / h+ |