Multi spectral rendering

Greg,

In your paper to the 2002 Eutographics Workshop on Rendering, you compare sRGB and Sharp RGB renderings against a multi-spectral rendering at 69 evenly-spaced spectral samples from 380 to 720 nm. I am curious how you made that multi-spectral rendering. Did you divide the 69 wavelengths into groups of three and run rpict(1) 23 times? Or did you run rpict(1) 69 times with proportional R, G, and B values? Or did you hack rpict(1) from a 3-band renderer to a 69-band one?

In the first case, wouldn�t there have been issues with gradient interpolation, which is calculated using the CIE definition of brightness from R, G, and B? Also adaptive shadow testing? I am currently wrestling with these issues as I have to perform multi-spectral renderings of many more than 69 bands from 380 nm to 14 microns, so the CIE definition of brightness won't do.

···

--
Andrew Bettison
Acacia Research Pty Ltd

Hi Andrew,

Since we needed to make exact comparisons between rendered images by pixel subtraction, I gave up on my efforts to run Radiance many times for spectral sampling, and we used my coauthor's (Elena's) Monte Carlo ray-tracer, instead. If regenerating the exact samples isn't an issue for you, you probably can run Radiance many times, however, and others have done this.

The best way to do it for non-human consumption is to defined many materials files, one for every three samples, and render from these in a for loop. The only place RGB is assumed in the code is when it's combining primaries to get brightness for a weight test or somesuch, and it shouldn't affect the result much for adjacent samples, which should be fairly similar.

Good luck!
-Greg

···

From: Andrew Bettison <[email protected]>
Date: Thu Jun 19, 2003 12:50:14 AM US/Pacific

Greg,

In your paper to the 2002 Eutographics Workshop on Rendering, you compare sRGB and Sharp RGB renderings against a multi-spectral rendering at 69 evenly-spaced spectral samples from 380 to 720 nm. I am curious how you made that multi-spectral rendering. Did you divide the 69 wavelengths into groups of three and run rpict(1) 23 times? Or did you run rpict(1) 69 times with proportional R, G, and B values? Or did you hack rpict(1) from a 3-band renderer to a 69-band one?

In the first case, wouldn�t there have been issues with gradient interpolation, which is calculated using the CIE definition of brightness from R, G, and B? Also adaptive shadow testing? I am currently wrestling with these issues as I have to perform multi-spectral renderings of many more than 69 bands from 380 nm to 14 microns, so the CIE definition of brightness won't do.

--
Andrew Bettison
Acacia Research Pty Ltd