On Sat, 21 May 2005 12:18:11 +0200, Tim Nikias wrote:

> Yup, I think so.

I feared as much :)

> For one, this will have to be done for EVERY pixel in the
> image (unless the arealight is confined with a spotlight). Once you go into
> transparencies and reflections, this complicated lighting technique will
> have to be done for every refracted/reflected ray. I don't expect the
> analytical method to be faster than the method used right now, as some
> objects are quite complicated in themselves, solving them analytically to
> find boundaries will surely take longer than simply sampling a set of
> points.
> For example, add some media, and you're in for a difficult ride.

Good points. Of course there would be quite some room for
optimizations, like skipping every other pixel and/or using the result
from adjacent pixels, or starting with some bounding box projections
(simpler shapes) and then fine-tuning with true projections.

OTOH, to obtain satisfactory results one still needs to use a rather
large amount of points, especially when not using jitter.

Maybe another idea would be a different form of adaptive sampling:
start with no pre-fixed number of lights in the area light, but
proceed by halving until the shadow delta is "good enough" (where the
user chooses the "good enough" instead of the number of lights)?

-- 
Giuseppe "Oblomov" Bilotta

"I weep for our generation" -- Charlie Brown

Post a reply to this message