Am 13.12.2016 um 22:15 schrieb [GDS|Entropy]:

> Any idea why this is, and how to fix it? Is the only choice to bound the media?
> I am guessing that this is because it is not reasonable to calculate media for
> an infinite space, but I could be wrong.

Yes, I have an idea now.

It's a flaw in media computation regarding the self-extinction effect of
media.

With sampling method 3, in essence POV-Ray divides the media interval
into a number of sub-intervals, and for each sub-interval computes (a)
the emissive(*) effect and (b) the extinction effect. It then multiplies
the emissive effect of that sub-interval by a factor to account for the
extinction effect of earlier sub-intervals as self-extinction within the
sub-interval itself.

It does a clean job at accounting for the extinction effect from earlier
sub-intervals; however, when it comes to self-extinction, POV-Ray simply
pretends that a sub-interval's emissive effect is precisely located at
the center of the sub-interval, while its extinction effect is
distributed evenly alongt the entire length of the sub-interval.

That's a good enough approximation when the length of a sub-interval is
reasonably small compared to the density of the medium. However, with
sufficiently dense media or sufficiently large distances involved, this
approximation will tend to underestimate the net emissive effect, to the
point where it can turn out pitch black.

This can be fixed, by using a more precise formula for the
sub-interval's self-extinction.


Media sampling methods 1 and 2 are different beasts, which do an even
poorer job at estimating the self-extinction effect, and it may not be
feasible to fix those.


(*From the perspective of the media computations code, scattering is
also an emissive effect, the only difference being that its brightness
depends on incoming light.)

Post a reply to this message