Quote -
I'm also thinking about a selective subdivision, based on the normal of the polygon with respect to light sources. It would be pretty easy I think to decide which polygons are near a terminator and handle them differently. I'm not familiar with SubD algorithms, though, so odf would need to comment.

I don't know all that much about selective subdivision algorithms, except that they exist. :laugh:

That said, here are some thoughts on the matter:

1) I don't think looking for terminators is the way to go. First of all, they are just the most obvious places where the shading goes wrong because of 'large' polygons. But like stewer said, wherever you try to shade a flat polygon as if it were curved, you get a problem. So it seems to me that selective subdivision should be based on the angles between adjacent polygons. If the angle is too 'sharp' - i.e., too far away from 180 degree - then these polygons should be subdivided and the angle smoothed out.

Second of all, a selective subdivision that is independent of lighting would allow us to export the geometry once and then just play with the lighting without having to wait for the subdivision to complete each time. The meshes produced could also be exported to other software.

2. As I said, I don't really have a solid knowledge of these things, but I'm pretty sure Loop subdivision (what Lux uses internally) is better fitted to selective methods than Catmull-Clark (what I'm using in the exporter). I don't know the exact formulas for Loop subdivision, just the general scheme (one triangle turns into four smaller triangles). But it shouldn't be too hard to find references.

Anyway, interesting idea. I'd almost be inclined to do it just for the heck of it, because that's how I'm wired. :lol:

-- I'm not mad at you, just Westphalian.