I tried lowering nearest_count to 1 and recursion_limit to 2, and increasing
low_error_factor to 0.5. This results in a very blotchy image. I can reduce
that some by lowering the final pretrace to 0.005 and increasing the count
but it takes very long to render and is still a bit patchy.

Stumped, I tried the cornell.pov scene. It renders in 30 seconds in the beta
and 1m 31s in v3.6, so for that scene at least there is a major decrease in
render time and the result looks pretty much identical. So I copied and
pasted the radiosity settings from cornell.pov into my scene and reran the
scene file. In the beta is takes 2m 59s and in v3.6 it takes 3m 9s. There
are some very slight differences but for the most part the images look
almost identical, even the brightness of the image. However there is no
detail evident, just some patchy shadows and color bleeding.

I might see if I can find some off Gilles Tran's old radiosity test files.

"clipka" <nomail@nomail> wrote in message 
news:web.496fcbd296903692dea10790@news.povray.org...
> "Mike Hough" <nos### [at] nospamcom> wrote:
>> If anyone has any suggestions for how I might alter the settings to 
>> improve
>> the render I would be happy to rerun this test.
>
> - Reduce the nearest_count to 1. There is no point in using any other 
> value
> anyway that could not be achieved by modifying the low_error_factor just 
> as
> well. Although little known, this was true for 3.6 likewise.
>
> - Increase low_error_factor; having this set to a low value may have 
> helped in
> 3.6 to reduce artifact "strength", but should no longer be required in
> 3.7.0.beta.30-rad1. A value of 0.5 should be perfectly fine now in most 
> cases.
>
> Note that with a nearest_count of 10 and a low_error_factor of 0.2, your
> pretrace will attempt to achieve a 250-fold (!) higher sample density than
> needed for the final trace. (It will never ever achieve this coverage 
> though,
> because it will not do enough passes.) Setting nearest_count to 1 and
> low_error_factor to 0.5 will cause the pretrace to still aim for 4-fold 
> sample
> density.
>
> - Try reducing the recursion_limit, unless you really need three bounces 
> (which
> is very rarely the case). The bugs and flaws in 3.6 basically caused 
> deeper
> radiosity "bounces" to be limited to a single trace level, i.e. every 
> mirror
> and window would look pitch-black to them. The beta.30-rad1 is not so 
> drastic,
> with the effect that deep-recursion pretraces take longer than before even 
> with
> the same number of samples gathered.
>
>

Post a reply to this message