[Radiance-general] BSDF xml into Radiance

[Lars O. Grobe]

Hi Greg,

thank you for the reply. I tried a bit since, so it is almost one week
later that I can report on what I experienced so far.

> > 2) rtcontrib and patch-based model. > > Advantages: faster for annual
> simulations. > > Disadvantages: noise, nice images require high (slow)
> -ad and cannot be > optimized using mkillum, limitations about specular
> non-planar > reflectors apply.
> 
> The accuracy of non-planar, specular reflectors is actually better than
> #1, but the results are somewhat noisy.  A new -c option to vwrays
> (coupled with the rtcontrib -c option) is a good way to reduce noise
> that is available in the latest HEAD.  This is a better way to reduce
> noise than increasing -ad, and less costly.

Classical Radiance scenes with both "light" and "glow" sources used to
require high settings of -ad to generate "noiseless" images. This led to
something like, say, -ab 4 -ad 1024 -as 512. This was typically
impossible to render, so mkillum was used to smooth out the ambient
transmission through fenestration and keep the sharp transmission
distribution only in the direct calculation. The ambient settings would
be released to e.g. -ab 2 -ad 256 -as 128 - fast and noise-reduced.

With rtcontrib, we typically have no "light" sources any more,
everything is "glow". At the same time, mkillum is not available as a
tool any more.

It is possible to reduce the amount of rays spawned at any ambient
bounce by not increasing -ad but -c and having the jitter smooth out the
distributions from different sky patches instead. This means that
additional rays are spawned only at the first level (direct) but not for
daughter rays. However it reduces only the direct noise - noise in e.g.
shadow patterns for the patches representing the direct sun. Noise for
areas lid only by indirect, diffuse reflected light e.g. in the back of
the room is not reduced.

My guess is that the -c option aims at the same strategy as what Mark
Stock found to be promising for scenes with a lot of geometry and many
ambient bounces:

http://markjstock.org/radmisc/aa0_ps1_test/final.html

What is the actual advantage of using -c 4 instead of doubling x- and
y-resolution?

Cheers, Lars.