RE: [Algorithms] adaptive radiosity and lightmaps

[Tom F. <tom...@ee...>]

We re-traced the lighting with the new sampling, but using multiple
(adaptive) samples over the area of the new texels. I was predicting =
cases
where the original sampling wasn't good enough to spot a feature that =
would
prevent shrinking, and then the lower density sampling blew this tiny
artefact up, and other pathological cases like that - but it never =
seemed to
happen in practice. Yay!

One of the things we did was make the initial sample res pretty small, =
and
then pieces would be enlarged as well as shrunk before being retraced. =
This
chopped a lot off the tracing time! Where a whole triangle was entirely =
lit
or entirely shadowed, it would shrink to a very small size and there =
would
be very few traces, which meant we could devote a lot more traces and
precision to the parts with shadows (especially sharp-edged shadows).

("trace" =3D raytrace to see if a light is visible from a texel or not - =
this
was a pure direct-lit lighting model - the artists preferred this to
radiosity-style indirect bounce stuff - any time they needed indirect
lighting they'd fake it with other lights, which meant the realtime =
lighting
matched better as well)

The shrink metric was done just by mipmapping that bit of the texture,
blowing it back up with a bilinear filter and asking what the greatest =
error
was. There's some fiddly stuff to ignore texels that aren't used by any
tris.

TomF.

>   What technique did you use for shrinking the triangles after you had
> generated the lightmap? Did you recompute lighting with the new
> parameterisation, or just resample the old bitmap?
>=20
> --
> Chris Butcher
> Networking & Simulation Lead
> Halo 2 | Bungie Studios
> bu...@bu...