Hi All,
We have been using Radiance in our lab for many years performing
lighting analyses mainly for exterior (on orbit) activities involving
the shuttle and the space station. We are now beginning to do more
interior lighting evaluations such as the space station laboratory
module.
The interiors we evaluate tend to have a lot of white, non specular
surfaces. So diffuse interreflections seem to be important. I've been
looking into the parameters that control the ambient calculation in
Radiance and I am baffled by the "-ab" ambient bounces parameter. I
thought, by keeping all other parameters the same and increasing the
ambient bounces, the results should be more accurate. That is not what I
see from our test case.
We built a box 17" x 17" x 36" with white walls. The wall material
properties were measured. A calibrated light source was placed in a hole
at one end of the box. Illuminance measurements were taken at various
points within the closed box. We then modeled the environment in
Radiance and took illuminance values at the same points. After some
experimentation I settled on the following ambient settings:
av: 0.0 0.0 0.0
aa: 0.08
ab: 3
ad: 1024
as: 512
ar: 64
The errors between Radiance and measured values were less than 10%.
But as I increased the ambient bounces beyond 3, the illuminance values
at the points kept increasing. By 5 bounces the errors were averaging
over 60%.
It appears that I am missing something fundamental. I'm hoping the group
can help me out.
Thanks,
rick
Rick Boettger
SF3/Habitability and Human Factors Branch
Lockheed Martin Space Operations
Graphics Research and Analysis Facility
Johnson Space Center
Rick,
I believe this is a known but rarely encountered condition. Radiance has various optimisations -- one of them is to assume that the overall environment reflectance is 0.5 (Greg, is that right?). That's fine for most realistic architecture. Your situation is more like an integrating sphere. From an earlier (private) exchange with Greg regarding a "dodgy" validation where this effect showed up:
Forgot about the -as value, and I also forgot there is another place where the AVGREFL (0.50) undermines the calculation. This value determines how many rays will be traced in deeper levels, which is the optimization I was referring to but didn't explain in my last e-mail. I thought by setting -aa 0 that I would override this limitation, and it does in one place, but not the more important place where the weight of additional interreflections is assigned. To get Radiance to really behave properly for this scene, one would have to redefine the AVGREFL macro to something closer to 0.95 and recompile -- not an honorable way to participate in a validation, however ill-conceived.
Sooo... where does that leave us? The only way you can get Radiance to send any rays at all at level 50 without recompiling is by setting -ad 1000000000000000 -ab 50, which is probably not a good strategy if you want to post a response during this century. It seems to me like this test was designed to thwart irradiance caching techniques in general, or Radiance in particular. Why not use a box with mirrored walls? I think Lightscape or whatever other radiosity software they might try wouldn't do as well in that circumstance. It's not like either one is a case anyone would encounter in the real world, and the optimizations in Radiance have been working for real world problems for over 15 years. Changing them to suit a special case as this would be silly, and would undermine the performance of the system in its intended domain. Given that Radiance is distributed in source form, there is nothing to prevent researchers with special needs redefining AVGREFL and recompiling the renderer to handle bizarre test cases such as this.
It sounds like you have one of those "bizarre test cases". Which is fair enough given the setting.
-John
PS. Would love to see some of the renderings.
···
-----------------------------------------------
Dr. John Mardaljevic
Senior Research Fellow
Institute of Energy and Sustainable Development
De Montfort University
The Gateway
Leicester
LE1 9BH, UK
+44 (0) 116 257 7972
+44 (0) 116 257 7981 (fax)
[email protected]
http://www.iesd.dmu.ac.uk/~jm
Hi Rick,
The AVGREFL macro notwithstanding, you shouldn't be converging to the wrong result when you increase the -ab setting under any circumstances. This sounds suspiciously like a bug, and it may be related to the sampling bias problem that Rick Mistrick and Younju Yoon ran across in their work at State College (Pennsylvania). It only shows up in very particular situations, but maybe yours is one of them.
If you don't mind, please send me your test scene privately, and the commands you used to generate your results.
Thanks,
-Greg
···
From: "Boettger, Richard G. \(JSC-SF\)[LM]" <[email protected]>
Date: April 6, 2006 8:55:18 AM PDT
Hi All,
We have been using Radiance in our lab for many years performing
lighting analyses mainly for exterior (on orbit) activities involving
the shuttle and the space station. We are now beginning to do more
interior lighting evaluations such as the space station laboratory
module.
The interiors we evaluate tend to have a lot of white, non specular
surfaces. So diffuse interreflections seem to be important. I've been
looking into the parameters that control the ambient calculation in
Radiance and I am baffled by the "-ab" ambient bounces parameter. I
thought, by keeping all other parameters the same and increasing the
ambient bounces, the results should be more accurate. That is not what I
see from our test case.
We built a box 17" x 17" x 36" with white walls. The wall material
properties were measured. A calibrated light source was placed in a hole
at one end of the box. Illuminance measurements were taken at various
points within the closed box. We then modeled the environment in
Radiance and took illuminance values at the same points. After some
experimentation I settled on the following ambient settings:
av: 0.0 0.0 0.0
aa: 0.08
ab: 3
ad: 1024
as: 512
ar: 64
The errors between Radiance and measured values were less than 10%.
But as I increased the ambient bounces beyond 3, the illuminance values
at the points kept increasing. By 5 bounces the errors were averaging
over 60%.
It appears that I am missing something fundamental. I'm hoping the group
can help me out.
Thanks,
rick