"Kari Kivisalo" <pro### [at] luxlabcom> wrote:
> Try this ...

Yeah ... I'm currently doing some tests with my own scene,
and I think somehow I understand the behaviour now. What I
could observe till now is, that samples is (of course) the
most limiting factor. Lowest priority seems to be the con-
fidence value, but I can get good results with conf.=0. It
also seems responsible for banding in some blur-gradients.
The important limit seems to be determined by the variance
because this is the parameter, that creates the edges that
are at the outer area of the blurring. But I have problems
to make it suitable for a nice image without setting it to
zero. But my current tests show that, even 10^-12 will not
be small enough to avoid the artefacts. But I think, that
I can also say, that normally 256 samples are enough, and
if one also sets variance to 0 then confidence seems to be
irrelevant, but to be sure I will still leave it on 1. And
I think now, that I will for all my coming focal blur pics
the variance at 0 and for "fast renders" only lower number
of samples ...

Post a reply to this message