how to clean up the emissive aov?

   2519   5   0
User Avatar
Member
106 posts
Joined:
Offline
Hi again,

I've been doing some test renders with Karma lately, looking at all the aov's etc.. And, I've set up some emissive geometry in my scene in order to have an emissive pass for compositing. However, in my setup I am having troubles getting a clean emissive aov pass, regardless of whether I use cpu or xpu. After fiddling with a few things it still remains by far the most noisy. So:

Do you guys have some tips as to how to get a clean emissive pass?

Thanks,

Dag
User Avatar
Member
8177 posts
Joined: Sept. 2011
Offline
How is it getting noise in the first place? I would expect the emission aov to be 100% noise free even with 1 sample.
User Avatar
Member
106 posts
Joined:
Offline
I don't know, but here is a section of a frame, comparing the emissive pass (combined emissive) with another light pass and the glossy reflections. This render was done with xpu, set to 128 samples and denoiser on all passes. But, the result is the same or similar if I use the cpu render - the emissive pass is very noisy.

Cheers,
Dag

Attachments:
emission_noise_comparison.png (6.0 MB)

User Avatar
Member
8177 posts
Joined: Sept. 2011
Offline
Dougie0047
I don't know, but here is a section of a frame, comparing the emissive pass (combined emissive) with another light pass and the glossy reflections. This render was done with xpu, set to 128 samples and denoiser on all passes. But, the result is the same or similar if I use the cpu render - the emissive pass is very noisy.

Cheers,
Dag

Oh, the indirect emission. The combined emission combines the direct emission and indirect emission into one pass. If you separate them, the direct emission should be noise-free. Indirect emission is the emissive surfaces contribution of lighting to the scene, so I would expect it to be pretty noisy without cranking the number of samples into the thousands.

Keep in mind the render only calculates 'noiseness' based on the beauty. If the indirect emission's contribution is very little, then the adaptive sampling won't spend any extra samples trying to resolve it.
Edited by jsmack - Dec. 13, 2023 15:48:26
User Avatar
Member
9384 posts
Joined: July 2007
Offline
Also keep in mind that treating emissive surfaces as light should reduce noise as it will be possible to sample them directly rather than relying on random hits

Based on your other thread I know that you made conscious decision to not treat it as lights so I wonder why
Tomas Slancik
CG Supervisor
Framestore, NY
User Avatar
Member
106 posts
Joined:
Offline
Thanks for your replies jsmack and tamte!

@jsmack : is there some way of forcing karma to sample the indirect emission more, in relative terms? I mean, without having to krank up the samples everywhere?

@tamte: The setup I am rendering is meant to be a little test scene for CG comp setups in nuke. So I need to cover a wide range of surfaces, shaders and lights, in order to have something to play with and make sure that the comp setups work properly. I wanted to have an emissive pass, so that I can do some comparisons with other render engines and their standard passes. But, if I have to krank up the render samples so much that I have to wait for hours to get a final frame I will most likely just treat it as a light as you have suggested. Just trying everything else first

One more question for you guys: One of the shaders has coat enabled, but when I render with xpu, for some reason the coat aov does not appear but seems to be calculated with the combined glossy reflections. With cpu it does appear. Any idea as to why? Is this a limitation of xpu? Or have I most likely done something wrong in the setup of things?

Also, is there an in-depth schematic somewhere for all the additive passes that karma offers and how they should be combined in order to rebuild the beauty? I can figure this out myself of course, but it would be good with some official documentation on it. I admit I have not really searched much for it myself yet.

Thanks again,
Dag
  • Quick Links