[Bf-cycles] Complex lighting situations issues_light intensity decreases > number lights increases

[Guido Medici]

Hi again to everyone,
I think that Stefan has nailed the point, the problem in that and, i think,
in many other scenes with a lot of light source should be the clamp. In
particular, the one that contributes the most at this issue is the "direct
clamp".
As Stefan says, it isn't the "0 Clamp" that produce fireflies, contrarily,
the scope of Clamp is to avoid it. But in this particular case, increasing
the number of light sources seems to influence the overall intensity of
each light.
However, the most fireflies seems to come from indirect bounces, so I
resolved the intensity issues in this scene setting only clamp direct to 0
and leaving the indirect clamp to 4.
So, I would like to ask, it is reccomended "play" only with indirect clamp,
to avoid to lose energy in direct light?
I'll continue to experiment where and when "direct clamp" is necessary, for
the moment I'd like to thanks everybody.
I'd also like to attach another suggestion posted on the
blender.stackexchange.com from Zauber Paracelsus yesterday:

"The default "Path Tracing" integrator is coded so that for each sample, it
will bounce light from a randomly-chosen lamp.

My suggestion is that you change the integrator to "Branched Path Trace",
which you can change from the Sampling sub-tab.

Using Branched PT will make cycles take bounces from every light for each
sample. This comes at the cost of increasing render times, but with the
added benefit of reducing noise.

More info here:
https://docs.blender.org/manual/en/dev/render/cycles/settings/integrator.html

EDIT: As a further note, Blender 2.79 will feature probablistic termination
of lights based on contribution. Basically, it will randomly skip sampling
from lights if their contribution to a sample would be below a certain
threshold, causing it to spend less time computing bounces for lights which
make little or no difference for a given sample. This provides speed gains
in complex lighting setups, particularly where some of the lights are
distant or occluded. More info here: https://developer.blender.org/rB26bf230
"

Guido


2017-02-13 10:58 GMT+01:00 Stefan Werner <stewreo at gmail.com>:

>
> > On 13 Feb 2017, at 02:02, F Tamas <tamasf.3d at gmail.com> wrote:
> >
> > But setting clamp to 0 makes fireflies, so rendertime will be far
> longer, am I right?
>
> Setting clamp to 0 does not *make* fireflies. Setting clamp to values
> other than 0 imposes a maximum brightness limit on individual samples. In
> many cases, limiting that brightness hides fireflies, but in cases like
> this, it can take away important contributions.
>
> A simple case to demonstrate this is if you take an object with a bright
> and sharp specular highlight and have that reflection be blurred either by
> depth of field or motion blur. Setting a clamp value on direct light very
> quickly dims what otherwise is a bright spot/streak.
>
> With many lights and path tracing, the following situation takes place:
> In standard path tracing, only one light is being sampled per pixel
> sample, all other lights are ignored. To compensate for that, the
> contribution of that light must be divided by the probability of that light
> being sampled, that is, divided by the number of lights. With 100 light
> sources, that means any given light source will be considered for only 1 in
> 100 pixel samples, but its contribution to that sample will be 100 times
> that of what it would be if it were just one light source. When that sample
> is now being clamped, this brightness gets limited and the appears darker
> than it should be.
>
> Without looking at the Guido's scene, I cannot tell if this is what’s the
> issue there, but this is my educated guess.
>
> -Stefan
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> https://lists.blender.org/mailman/listinfo/bf-cycles
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20170213/5e5b94ba/attachment.htm