"Christoph Hormann" <chr### [at] gmxde> schreef in bericht 
news:eeesuu$npn$1@chho.imagico.de...
> Thomas de Groot wrote:
>> This is almost 'perfect' :-)
>> Still a bit of artifacts visible, but not much.
>> Essentially, what I have changed since the preceding render are:
>> increasing nearest_count to 10
>> increasing error_bound at 0.5 in the first pass
>> increasing low_error_factor to 0.9
>> increasing count to 250
>> decreasing recursion_limit to 2
>> decreasing brightness to 0.8 in the second pass
>
> It looks much better (although difficult to judge at this small size) but 
> under the aspect of the argument i gave concerning the adaptive 
> error_bound using a higher one in the pretrace than in the final trace is 
> counterproductive.  Of course the adaptation will automatically increase 
> it in the final pass but this will make it slower than necessary.  If you 
> set the starting value of the adaptation to the same as in the pretrace 
> this will probably make it quite a bit faster and not much worse (although 
> it is difficult to say for sure). Since there have not been a lot of tests 
> made with the adaptive error_bound it might be interesting to see the 
> (non-jpeg-compressed) results of this in comparison.   If you could quote 
> the number of samples taken in pretrace and in the final pass that would 
> also help to say if these assumptions are correct.
>
Thanks Christoph. I much appreciate your comments as I feel that I still 
have a lot to learn about this matter. Forgive me if I seem obtuse, but I 
find this a difficult matter to understand, despite the excellent tutorials, 
by the way, of you and others.
Let's see if I get this correctly.
Concerning error_bound, what I should do is (for instance):
error_bound 0.5 //first pass
error_bound {0.5 adaptive 1.5, 20} //second pass
OK? The quality of the results might then dictate using 0.5 or lower (or 
higher, to see where acceptable boundaries are).
My initial assumption was that by making error_bound adaptive (at least in 
the final pass) megapov would determine automatically what the necessary 
values were, thus gaining time and quality.

I saved the message files fortunately. These are the number of samples taken 
during pretrace:
Radiosity samples calculated:           424116 (1.34 %)
Radiosity samples reused:             31208079
  Samples (final trace)                  59336
  Samples (recursion 1)                  78012
  Samples (recursion 2)                 346104

And these of the samples taken during the second pass:
Radiosity samples calculated:            65094 (1.16 %)
Radiosity samples reused:              5549135
  Samples (final trace)                  64420
  Samples (recursion 1)                   7061
  Samples (recursion 2)                  58033

With the changes to error_bound, I shall make a new render and post both 
images (the present one and the new one) together in png format.

> BTW using different brightness in two passes is handled by POV-Ray AFAIK 
> but there is not much point in doing this.
>
Oh? I confess I followed the teachings of  Tim Nikias in this matter.

Thomas

Post a reply to this message