POV-Ray : Newsgroups : povray.general : Some more Radiosity Voodoo Server Time
5 Nov 2024 09:25:31 EST (-0500)
  Some more Radiosity Voodoo (Message 1 to 1 of 1)  
From: clipka
Subject: Some more Radiosity Voodoo
Date: 30 Dec 2008 21:10:01
Message: <web.495ad39d2e2f6e7b6e13ad820@news.povray.org>
Well, not actually active black magic this time, but just some occult knowledge:

A while ago, there was a discussion here or someplace near about using the
radiosity save_file/load_file feature, wondering why a final render run
(loading samples and using them) would sometimes seem to take longer than a
pre-render run (to take samples and save them for the final run), and why the
final render would take lots of additional radiosity samples.

Maybe this is the reason why:

While digging through radiosity code, I found that POV-Ray 3.6 will only save
the "final" generation of samples.

That means that whenever a subsequent run loads this data to use it, but needs
to take more samples for whatever reason (focal blur for example), it needs to
take quite a lot of the "deeper-bounce" samples all over again.

This explains why a combined pre-render / final render run may often be
significantly faster than a pre-render run with an optimized scene, followed by
a full-detail final run - even if the difference is just the focal blur.


However, there is also an explanation why saving radiosity data, and then
running exactly the same scene with the loaded data again, may produce
significantly better results:

This indicates that the pre-render settings were actually insufficient, and
additional samples had to be taken during the (first) final render. This may
lead to "jumps" in brightness or color between areas rendered before and after
that additional sample is taken (because every sample typically affects all its
surrounding pixels as well). However, these additional samples will be included
in the saved data, and the second final render will have them available, so it
will be able to take them into account for all surrounding pixels, not just
those "after" the sample.

So seting "always_sample" to "off" is highly recommended (because it prevents
samples from being taken during final trace unless really absolutely inevitably
necessary) except for quick test renders where you want to save some pe-render
time.


Post a reply to this message

Copyright 2003-2023 Persistence of Vision Raytracer Pty. Ltd.