Every day of trying to improve the radiosity code brings more quirks to the
surface...

Forget everything you learned at school about the radiosity adc_bailout.

The first thing to know is that adc_bailout handling in POV 3.6 radiosity code
is an ugly mess. It still is in 3.7.0.beta.29. It's even not all sorted out yet
in the code I submitted to the POV team.

I will spare you details here, because I'd have to explain half the inner
workings of POV-Ray. But there's a thing you should know:


The effective bailout value used for radiosity sampling rays varies with the
sample ray count!


So as you increase your sample ray count in hope to increase your sample
quality, your quality gain is offset by a quality loss in the individual sample
rays - maybe to a level where your realistically modeled glass window all of a
sudden turns opaque for radiosity, cutting off your only light source. Whoops!

With the existing versions, this "shutting off" of reflections and refractions
will not happen all at once, but depend on the surface for which radiosity is
being sampled (and actually some more factors), so it may look just like some
more artifacts. With the modifications I made, I expect the shut-off to occur
simultaneously throughout the whole scene.


So a few pieces of advice:

- Whenever you increase your sample count, make sure to check and possibly
reduce your radiosity adc_bailout

- Forget the formula from the online help; it is right in that your adc_bailout
value should somehow be reciprocal to the brightness of your brightest object
in the scene, but that's about it.

- Keep your radiosity adc_bailout lower than you'd mathematically expect;

- Try some toying around with the radiosity adc_bailout if you encounter
radiosity artifacts, especially when they appear after you increased the sample
ray count.

Post a reply to this message