[REQ] remove aperture diffraction supersampling fireflies

[ior]

Can they be removed?

Were fireflies are persistent we can use supersampling to get rid of them, but when we use post process aperture diffraction the fireflies persists in appearing even if the render without diffraction does not have fireflies (traced diffraction is not an option because it only appears on the lights and not on the reflections or refractions or volumes).

If we use MLT, fireflies are not so many but in some situations the stripes of MLT do not disappear in acceptable time, so we have to use BiDir using supersamling. We can lose some caustics or sss but we can have a render without fireflies or at least with not so perceptible fireflies.

Here is a MLT render that do not converge at acceptable time:

aperture diffraction on

aperture diffraction off

The problem is that when we use aperture diffraction and high suppersamling, there are fireflies that appear on the render that are not there when we turn off diffraction, so, if we have a sort of super HDR image in indigo GUI when there is no aperture diffraction and not so apparent fireflies (maybe you can call it noise), we can now calculate the aperture diffraction based on the bright spots of the image without fireflies. Can we? Or can it be just a filter you add to diffraction and the fireflies are off?

Here an example of a render with aperture diffraction fireflies (supersampling 6, BiDir, aperture diffraction on):

Here is the same render without apreture diffraction:

There are not obvious fireflies.

So if we can have the bright spots info from the image without fireflies:

Can make aperture diffraction without fireflies from image info?

Here is another example of aperture diffraction fireflies:

I know that when the SDS problem gets solved or partially solved, maybe this kind of fireflies do not exist anymore, but for now we have them.
What do you think?

[Zom-B]

Having a noise filter kill Fireflies before applying PostPro AD would be a interesting approach.

But from my point of view I rather would invest development time to reduce fireflys directly,
that would be a win for everybody

In your case you could try using physical AD! From my experience it renders ~10% slower and causes some extra colored noise in the beginning, but avoids such AD firefly issues.

[CTZn]

(traced diffraction is not an option because it only appears on the lights and not on the reflections or refractions or volumes).

I tested about reflections however (opaque specular with very high IOR) and to be correct, traced aperture is happening altough at a muuuch slower rate than the direct source.

[ior]

Having a noise filter kill Fireflies before applying PostPro AD would be a interesting approach.

Yes it would be nice if it had a value for tweaking the filter.

From my experience it renders ~10% slower and causes some extra colored noise in the beginning, but avoids such AD firefly issues.

As CTZn said:

traced aperture is happening altough at a muuuch slower rate than the direct source.

in the reflections, etc. it is too slow. And post process AD works very well if there are no fireflies.

Of course the things I tried to render are not fast and traced AD, as well. So Ill make a test to see if traced AD is fast enough to follow the sss and in reflection caustics.

But from my point of view I rather would invest development time to reduce fireflys directly,
that would be a win for everybody

I hope not for long