Am 01.06.2010 06:29, schrieb Dave Blandston:

> radiosity {
>     count 300
>     error_bound .02
>     pretrace_start .08
>     pretrace_end .004
>     recursion_limit 2
>     normal on
> } //radiosity
>
>    Pass     Depth 0    Depth 1           Total
> ---------------------------------------------
>    2              1        330             331
>    3              6       1002            1008
>    4             48       7766            7814
>    5+           277      27392           27669
>    Final      61083    1057679         1118762
> ---------------------------------------------
>    Total      61415    1094169         1155584
>    Weight     0.092      0.047

Yes, with these statistics numbers it's no surprise that you get those 
blocky artifacts.

Your error_bound is unconventionally low, thus increasing the desired 
sample density; typically, a value of 0.5 will do. "normal on" increases 
desired density even more; unless you use radiosity as primary 
illumination you can typically go without.

As already mentioned, do use "always_sample off"; I'd also recommend 
"low_error_factor 0.5" to force the pretrace to go for a higher sample 
coverage than the final trace would demand for.

The pretrace_start and pretrace_end look quite ok to me at a first 
glance. If the other changes don't give you the desired final-vs-total 
ratio though, you might want to decrease pretrace_end.

The statistics also show that only very few samples are gathered during 
the first passes (and no samples at all in the very first one), so you 
may want to reduce pretrace_start; I'd suggest to start at what is now 
pass 3, i.e. "pretrace_start 0.02"

Post a reply to this message