> On 4/7/2011 3:35 PM, Alain wrote:

>>
>>>> You can try 33x33, 65x65 or even 129x129. Those values match beter with
>>>> the adaptive algorythm than 32x32.
>>>
>>> Ah, that's good to know.
>>
>> Good dimentions for adaptive are 5, 9, 17, 33, 65, 129, 257, 513,
>> 1025,...
>>
>> You take the powers of 2 and add 1 to the result.
>
> Thanks Alain! Is that in the docs? I can't seem to find it. I wonder if
> it should be mentioned somewhere...

Not clearly in the documentations, anywhere. It should becomes obvious 
when you carefully read the explanation of the adaptive procedure. The 
illustrations do help.

For adaptive 0, you start with the corner elements. A 2x2 array: 2^0 +1 
= 2. If the elements are all shadowed or all visible, you assume that 
you are totaly in a shadow or totaly illuminated. All the points between 
the conserned points are assumed to have the same visibility of the light.

If not, you take the middle points between the elements that don't have 
the same status.
Now, a 3x3 aray: 2^1 +1.

On the next step, it becomes a 5x5: 2^2 +1, then 9x9: 2^3 +1, then 
17x17: 2^4 +1,...(2^n +1) aray.
For most of those array's elements, you don't need to make any test, 
thus the speed up.

The documentation also tell that with adaptive 1, you start with a 3x3 
aray, and a 5x5 one for adaptive 2, limited by the actual dimention of 
the defined aray. adaptive 2 have no effect on a 4x4 aray.

If the dimentions are not in the list, the subdivision becomes 
asymetrical and you get penumbrae part that no longer have the same 
number of samples for the same area.



Alain

Post a reply to this message