[Radiance-general] ar and aa

[Greg Ward]

This is just getting so confused (and confusing).  By itself, the -aa  
parameter AFFECTS the distance between ambient values, but it does not  
dictate them.  The spacing actually depends on how close nearby  
geometry is.  If there is nothing nearby, then ambient values may be  
spaced very far apart.  You really need to go back to my original  
paper and understand what is going on with the interpolation:

	http://radsite.lbl.gov/radiance/papers/sg88/paper.html

The -ar setting gets used with -aa and the overall scene dimensions to  
determine a minimum spacing between values.  This avoids having  
infinite calculation density at inside corners and other places where  
objects are right next to and "see" each other.  If you set -ar 0,  
then it guarantees accuracy everywhere, but at a potentially large  
expense, as you end up computing a new hemispherical sampling at each  
pixel for certain parts of the image.  Combined with a -ab setting  
greater than 1, you can end up with a very long calculation, indeed.

Don't worry about what the exact spacing value is.  It doesn't  
matter.  All that matters is the scale over which you maintain the  
accuracy set by the -aa parameter.  Once your objects are closer to  
each other than the maximum scene dimension divided by the -ar  
setting, you will gradually lose accuracy.  There is nothing more to  
say on the topic.  Read the code in src/rt/ambient.c if you want to  
understand exactly what is happening.

Cheers,
-Greg

> From: Jia Hu <hujia06 at gmail.com>
> Date: July 19, 2010 10:47:57 AM PDT
>
> Sorry for the typo, I meant to compare "-ar" and "-aa".   I can  
> understand your explanation as to -ar and -aa.
>
> The book and some online materials also discussed about the meaning  
> of -ar and -aa. And they seem to explain it from a different  
> perspective. For my understanding of those statements, the  
> interpolation will always happen when the distance between two  
> points is less than minimum spacing distance (sceneSize * aa /ar).  
> When the distance between two points is larger than the minmum  
> spacing distance, interpolation may happen if the point is within  
> the "radius of validity" of another point. In other words, the  
> minimum "radius of validity" is sceneSize *aa /ar?
>
> According to Greg's explanation, when the point falls within  
> SceneSize / ar, the accuracy starts to relax. Can I say that when  
> the error gets its maximum, -aa, when the point falls within the  
> minimum spacing distance (SceneSize *aa /ar)?   I know there must be  
> something wrong with my understanding about this issue. But I can  
> not find where the problem is.
> The book Rendering with Radiance says "  -ar parameter acts as a  
> limiting device and If you are already running up against the -ar  
> limit, increasing the setting will result in a higher density of  
> sampling. If the limit has not been reached, then increasing -ar  
> should have no effect." I also have difficulty in understanding the  
> term "limiting device".
>
> Thank you for help.
>
> Jia Hu

> On Mon, Jul 19, 2010 at 2:31 AM, Greg Ward <gregoryjward at gmail.com>  
> wrote:
>
>> I assume you meant to compare "-aa" and "-ar" in your first  
>> sentence.  These are not two variables that affect the same  
>> behavior -- not much point in that.  Rather, think of the -ar  
>> setting as determining a scene resolution below which the accuracy  
>> of the indirect calculation (as determined by the -aa parameter)  
>> will start to relax.  If you divide your global scene size (the  
>> fourth value reported by "getinfo -d octree") by the -ar setting,  
>> you will get the scene size below which the indirect calculation  
>> will begin to lose accuracy.
>>
>> Best,
>> -Greg