[Radiance-general] rpict: color channel separation

[Greg Ward]

Hello Stefan,

This is simply the round-off error associated with the Radiance RGBE  
format.  Since the picture format uses three mantissas with a common   
exponent, it is not possible to represent a non-zero value in one  
channel and a zero value in another channel.  Even though the actual  
byte stored for green and blue may be zero in your calculation, the  
decoder will interpret this as half a count out of 255 and give a  
corresponding value.

Since the main purpose of Radiance is lighting calculations for  
visual simulation, and the eye's color channels have a (more or less)  
linear response, this is an appropriate/adequate representations for  
the design application.  If you need the original floating point  
values, and don't mind files than are about 5 times bigger, you can  
use vwrays with rtrace and output floating point images directly.

I hope this helps.
-Greg

> From: "Stefan Schwarzer" <stefan.schwarzer at gmx.net>
> Date: February 18, 2008 2:56:05 AM PST
>
> Dear list members:
>
> While getting acquainted with radiance, I stumbled on a phenomenon
> that I do not understand.
>
> I keep the following scene description (of a red sphere lonely in  
> vast Euclidean space) in a file called sphere_0.rad
>
> void light SphereLight_0
> 0 0 3 0.000431188 0 0
>
> SphereLight_0 sphere sphere_0
> 0 0 4 0 500 0 100
>
> I use oconv and rpict to render the scene and check the output  
> using pvalue.
>
> oconv sphere_0.rad > o.conv
> rpict -x 128 -y 128 o.conv | pvalue -o -h -u
>
> The first few lines of output of these commands are
>
> -Y 128 +X 128
>      60      95       4.320e-04       9.537e-07       9.537e-07
>      62      95       0.000e+00       0.000e+00       0.000e+00
>      63      95       4.320e-04       9.537e-07       9.537e-07
>      65      95       0.000e+00       0.000e+00       0.000e+00
>      67      95       4.320e-04       9.537e-07       9.537e-07
>      68      95       0.000e+00       0.000e+00       0.000e+00
>      56      94       4.320e-04       9.537e-07       9.537e-07
>
> They demonstrate that am picking up some small power contributions in
> the g and b channel which seem suspicious to me. When I use rtrace  
> in an
> attempt to reproduce the phenomemon, I find 0 for g and b as I'd  
> expect.
>
> I'd be grateful if somebody could share her/his insight
>
> ps - random observations:
>
> - I have experimented with program options, but even though the  
> defaults
> of rtrace and rpict differ slightly, they seem to not be able to  
> explain
> rpict's behavior
>
> - A colleague at Fraunhofer ISE checked the pic-file and mentioned to
> me that the non-zero contributions are present in the rpict output and
> are not created later by pvalue
>
> -- 
> Stefan Schwarzer