How to fix noise in Bokeh shapes?

[kosso_olli]

Hey,
I have a slight problem here. The rendering shown here is quite clean for a noise threshold of 0.01, but the noise in the bokeh shapes is very odd. I tried raising Min shading rate, but I couldn't get it cleaner. Higher Max AA also didn't help What can I do to fix this?



Regards

Oliver

[piotrus3333]

up min aa subdivs.

[Muhammed_Hamed]

Actually you need to use lower Min shading rate for this as your primary rays are responsible for cleaning this kind on noise. Use something like 3 for min shading rate..
Min subdivs 1
Max subdivs 50
noise threshold .008
GI,
Primary, BF
Secondary, LC default settings ,then change subdivs to 2000

[joconnell]

This seems like the type of area that comes up a lot that could use a bit of smart denoising or some other kind of less accurate cheat, especially for areas that bloom?

[Morbid Angel]

this kind of highlite might be coming from dome hdr, if it is you can't really fix it easily other then paint it out in the hdr just a thought tho.

[kosso_olli]

The highlight is coming from the interior lights on the rear view mirror. I excluded them now from this object, because otherwise I just couldn't get it clean. Tried all the settings mentioned here.

[kosso_olli]

This is still an issue for me. I have no idea how to avoid this. Happens in every scene with bright light sources and DOF. Which is basically all our jobs these days.

[Magnum_design]

Try different hdri´s. Try ticking "sub pixel mapping". Lower intensity of reflected lights. Change specular addition in option.

[^Lele^]

I think those paths are of the class of "high energy, low probability" (read: nasties.).
Meaning you'll likely have to resort to ancillary techniques (SPM,clamping of secondaries/primaries, soft roll-off of highlights, turning the paths to non-caustic ones, resorting to post effects on the offending lights, f.e.), as there is not going to be a set of "pure" sampling values which will clean them up properly.
This isn't an issue with us alone either: i'ts very much at the forefront of research in the rendering literature, worldwide.
Try Mitsuba, for example, and see how many different integrators and samplers it brings to bear to solve precisely this class of issues, and with which results (in speed and quality both.).

[piotrus3333]

any chance for a simple scene with the issue?
so far I can’t get bokeh so noisy.

edit:
ok, got it..

[piotrus3333]

these are getting cleaner with higher max AA (24vs64)

[kosso_olli]

these are getting cleaner with higher max AA (24vs64)

Nope, sadly not. This is with 2/100 for min and max AA, noise threshold is 0.005. Only solution I found so far is clamping the HDRI, which alters the lighting a lot. I'm not getting sharp shadows, of course.

[piotrus3333]

how about blasting the bokeh with fixed sampler and 100 subdivs?
if its all bright light sources (ibl or regular lights) than maybe you are using this old trick with using exponential color mapping for sampling but keeping the image in linear color? just a wild guess here.

[kosso_olli]

how about blasting the bokeh with fixed sampler and 100 subdivs?
if its all bright light sources (ibl or regular lights) than maybe you are using this old trick with using exponential color mapping for sampling but keeping the image in linear color? just a wild guess here.

For me there is too much trickery involved in that solution.

[joconnell]

Lele do you think that it'd be possible to group samples based on luminance / COC radius around the source point and feed those into a defocus / averaging technique to smooth out? We likely know the bokeh we're aiming for so is there any theory to blur / average while aiming for that known disc shape?