We are building an artificial sky for an architecture school where we teach, and before starting we prepare a Radiance model, to simulate the interior of the artificial sky. The idea is to help us to choose the right bulbs, the ceiling, etc.
Interesting approach. I would have used physical mock ups to decide on
materials (diffusers, layout and spacing) and measurements on a number
of lamps of the same type to check for manufacturing inconsistencies.
With Radiance you introduce the new complexity of modelling the
physical materials _correctly_ in a virtual world and in some cases
you have to deal with the optimisation of the algorithms.
To simulate the overcast sky (Moon and Spencer) distribution in Radiance we made walls of mirror material and ceiling is filled with fluorescent tubes (very close to one another) and covered with translucent material. Then we tried to render few images of interior, with several sets of rendering parameters, but final images still aren't good enough with a lot of spots.
Please see our results on http://docs.google.com/Doc?id=dfvbdbg3_13f66n99kx
At a first look I'd say your spotty images are the result of "-av 0 0 0" and a rather low
"-ab 4". Note that the trans material will swallow one bounce just to let light pass. That
leaves you 3 bounces to hit a light source which is not a lot.
As a first step I'd set "-av" to something realistic. You should be able to calculate a
reasonably good approximation of the ambient value from your physical set up.
Next: Your rendering times might benefit from an 'illum' calculation. Replace the
"trans" material ceiling with an "illum" to make it a direct light source. This will affect
the appearance of the ceiling but not the accuracy of the rendering. However, this
may go against your idea of simulating the physical set up.
In your last image you're disabling the ambient cache with "-aa 0". Now every
ray is calculated with your high settings. You can probably reduce "-ad" and "-as"
a lot in your uniform environment. I'd also set "-lr" back to 8 again. There is a
difference but it has not such a big visual impact.
On 22 Oct 2008, at 10:04, jeando wrote: