Lights + Glass = Noise

[RobSteady]

Is there any trick to render night scenes with many lights behind glass?
Octane handles many light sources pretty good but in combination with glass (especially behind windows) noiselevel and rendetimes rise sky high.

I've tested all kernels; Direct Light gives almost same speed as PT with lots of hotpixels (also Ambient Occlusion). PMC has a very even noise distribution but is so damn slow...
I've used little cubes rotated by 45° with a light material applied. I testet both regular spherical Octane lights and light materials; looked pretty much the same and rendered identical.
Rendertime was 25 min for each.


This is the scene without any glass, only lights inside PT Diffuse 6 / Glossy 6, 6 samp/s, 1800 samples
Noiselevel looks ok

This is with specular glass material (geometry has thickness, fake shadows on) PT Diffuse 11 / Glossy 11, 3,2 samp/s, 1000 samples
Heavy noise where light gets reflected, higher rendertimes and could use even more depth (window behind railing glass gets very dark)

Fake glass (glossy chrome material with low oppacity) PT Diffuse 3 / Glossy 3, 6,5 samp/s, 1800 samples
Also heavy noise in refelctions but better renderspeed because you can lower the depth.

[oguzbir]

Nice work btw.
Can I just say Lights = Noise

[RobSteady]

I know it's simple but it's only for testing
I would not say Lights = Noise; as you can see on the first image Octane handles 35 lights pretty good and fast as along as there is not any glass.

[Bendbox]

Hi RobSteady,

I've battled this myself . . .

Using PT, you can turn the Caustics to 1 and the GI Sampling to somewhere between 1-10 and it should help with the noise. I'm sure you're using Coherent ratio but if not try around 0.5 or so.

If all else fails, you can render region the areas until the noise disappears . . . I know . . . it's a pain. I had to do that on the night image in this post and it worked just fine you can't tell: viewtopic.php?f=5&t=45206&p=224924#p224924

Hope that helps some.

[RobSteady]

Thanks for the feedback.
GI Clamp and Caustic Blur are already set to 1. Coherent was on 0.4.
Then I guess it's just the way it is, there's nothing to do...

[mbetke]

I took the challenge and tried it by myself. After getting even more grey hair with the strange plugin bug I solved the puzzle and went on with yours:

Cooked it until 3000 samples. 6000 looked okay. You could refine it with a denoise software or integrate the noise to give it a realistic feeling.

The lights are just planes (0,2 x 0,2 m) and a backbody emission, you can use any other light. But it is important to set a sampling_rate so more light will be distributed there.
Used the settings from attached screenshot. The Glass-setup is pretty much straight forward. Double glazed windows in the frame with a specular material and slight bump to get imperfections. if you go with just a plane as window it will be faster i guess.
My curtains have a diffuse material with transluency.

All could be tweaked further. Was just a quick setup.

[RobSteady]

Thanks for giving this a try.
But I don't get what's different with your approach; you just let it render longer?

Regarding light samples: what I've read from the manual this makes only sense if you have more than one instance of a light and should only be used to balance sample rates between these. So it should make no difference if you have 35 instances of only one light source with a sampling rate of 1 or 1000000.
This is from the manual:

Sampling Rate
Emission nodes have a parameter called Sampling Rate, to control how much weight is given to the emitter when picking an emitter to sample. The sampling rate allows users to choose which light sources will receive more samples. For example, in a room with a big TV (which emits some light) and a couple of small bright spotlights, the TV would have been sampled most of the time, leading to a lot of noise in the light from the spotlights. In this case, users can increase the rate for the spotlights, so they get sampled adequately. Adjusting the light source sampling rates of the emissions in the scene will lead to a better balance between light sources. The sampling rate can also be set to 0 (zero), which means the emitter will be excluded from the direct light calculation

[mbetke]

I read it as a priorization of light. More samples mean more light and faster clearing. The more samples you put on the small lights the better they clear but on the other hand the remaining lights like daylight gets darker.

You also need more samples for an setup with low light. 1800 is not enough. For outdoors and daylight I usually got them clear around 800-1200 samples.
Night-images need more samples to clear up or you work with low-level supporting lights like a photographer would do.
The smaller your light, the more samples they need. I once had to light some LEDs and it was a nightmare...

Last but not least you could render the image 30% larger, use less samples but resize it so it looks more clear.

[RobSteady]

The more samples you put on the small lights the better they clear but on the other hand the remaining lights like daylight gets darker.

Just rendered the same scene with 1000000 samples for the light material; the image looks completely identical. Noiselevel and light/environment brightness.
Seems logical since you can't adjust any samples for daylights and environments (hdri).

So, again, if you have a scene lit by a sun/hdri + one light (or multiple instances of it) it doesn't matter if the sampling rate is set to 1 or 1000000 because there is no other relation.
If you have two lights and set sampling rates for both to 1 or 1000000 it shouldn't make any difference.
Maybe they should call it weight or something because samples seems confusing...

(Anyone from Otoy, correct me if I'm wrong)

[mbetke]

They are coding nerds we are artists. They are supposed to confuse us.