Cycles - Subsurface Indirect render channel unable to resolve over-bright pixels from Branched Path Tracing #45014

New Issue

Scott K · 2015-06-10T03:10:15+02:00

Scott K commented

2015-06-10 03:10:15 +02:00

System Information
Win7-64, Intel Xeon CPU E5-2650, NVIDIA Quadro K4000, 32 GB system ram

Blender Version
Broken: 2.74 e1b8ed8

Short description of error
Cycles rendering with Branched Path Tracing, other channels resolve noise with higher sample numbers, but Subsurface Indirect seems to add more noise with more samples. Combined render view is very noisy only on the object which has subsurface scattering, (the white bowl) and that noise pattern matches the subsurface indirect channel.

When using Path Tracing (non-branched) a very similar error seems to move to the Glossy Indirect channel.

Exact steps for others to reproduce the error
test_subsurfaceIndirectNoise.blend
Render the attached .blend file using either Cam_CloseUp or Cam_XCU. Currently set at very low samples, but setting higher samples doesn't really help much.

**System Information** Win7-64, Intel Xeon CPU E5-2650, NVIDIA Quadro K4000, 32 GB system ram **Blender Version** Broken: 2.74 e1b8ed8 **Short description of error** Cycles rendering with Branched Path Tracing, other channels resolve noise with higher sample numbers, but Subsurface Indirect seems to add more noise with more samples. Combined render view is very noisy only on the object which has subsurface scattering, (the white bowl) and that noise pattern matches the subsurface indirect channel. When using Path Tracing (non-branched) a very similar error seems to move to the Glossy Indirect channel. **Exact steps for others to reproduce the error** [test_subsurfaceIndirectNoise.blend](https://archive.blender.org/developer/F188567/test_subsurfaceIndirectNoise.blend) Render the attached .blend file using either Cam_CloseUp or Cam_XCU. Currently set at very low samples, but setting higher samples doesn't really help much.

Scott K commented

2015-06-10 03:10:15 +02:00

Changed status to: 'Open'

Scott K commented

2015-06-10 03:10:15 +02:00

Added subscriber: @BaconCheeseburger

Thomas Dinges commented

2015-06-12 02:22:50 +02:00

Added subscribers: @Sergey, @ThomasDinges

Thomas Dinges commented

2015-06-12 02:22:50 +02:00

Rendering this kind of scene with only 4 AA samples will certainly not make it converge.
Try something like 64 AA Samples and Diffuse/Glossy/SSS samples set to 4. That will already look better.

@Sergey Not sure there is a bug here?

Rendering this kind of scene with only 4 AA samples will certainly not make it converge. Try something like 64 AA Samples and Diffuse/Glossy/SSS samples set to 4. That will already look better. @Sergey Not sure there is a bug here?

Scott K commented

2015-06-12 03:17:58 +02:00

I was rendering with 50 render samples, Diffuse and Glossy at 16, and Subsurface at 32, and it was still noisy.
It looks a little bit better at those settings than at 64, 4, 4, and 4.

The issue is that the noise comes strictly from the Subsurface Indirect channel, and doesn't seem to average out with higher numbers of render passes. All other channels average out the noise, but not Subsurface Indirect.

Funny thing is that the same issue moves to the Glossy Indirect channel (even with what I'll call higher numbers of samples) when using non-branched path tracing.

I was rendering with 50 render samples, Diffuse and Glossy at 16, and Subsurface at 32, and it was still noisy. It looks a little bit better at those settings than at 64, 4, 4, and 4. The issue is that the noise comes strictly from the Subsurface Indirect channel, and doesn't seem to average out with higher numbers of render passes. All other channels average out the noise, but not Subsurface Indirect. Funny thing is that the same issue moves to the Glossy Indirect channel (even with what I'll call higher numbers of samples) when using non-branched path tracing.

Scott K commented

2015-06-12 03:35:55 +02:00

This may not be a bug at all. I would ask you to look at how the Subsurface averages out the noise in this scene, but this also can be chalked up to my not understanding the difference between the AA samples and the Subsurface samples setting on the right side.

I had thought that setting the Samples higher for Subsurface would help, as if the settings on the right simply gave you more customization than the general AA Samples setting on the left. But I see that in order to remove the noise I need to balance more toward more AA Samples. Setting 128 AA samples and only 4 Subsurface samples produces a =much= better result. I guess the Subsurface increases the quality of each individual sample, but setting the AA Samples higher does more averaging of the results?

The noise seemed a bit heavy on the Subsurface Indirect, but this may have been strictly user error and not a bug.

Thank you, and sorry for an errant bug report

Scott

This may not be a bug at all. I would ask you to look at how the Subsurface averages out the noise in this scene, but this also can be chalked up to my not understanding the difference between the AA samples and the Subsurface samples setting on the right side. I had thought that setting the Samples higher for Subsurface would help, as if the settings on the right simply gave you more customization than the general AA Samples setting on the left. But I see that in order to remove the noise I need to balance more toward more AA Samples. Setting 128 AA samples and only 4 Subsurface samples produces a =much= better result. I guess the Subsurface increases the quality of each individual sample, but setting the AA Samples higher does more averaging of the results? The noise seemed a bit heavy on the Subsurface Indirect, but this may have been strictly user error and not a bug. Thank you, and sorry for an errant bug report - Scott

Sergey Sharybin self-assigned this 2015-06-12 10:26:32 +02:00

Sergey Sharybin commented

2015-06-12 10:26:32 +02:00

@ThomasDinges, 4 AA samples is surely low value, but i wouldn't know why increasing AA samples to much higher value in such configuration would still give noisy results in particular passes without looking in details.

Mohamed Sakr commented

2015-06-12 14:22:57 +02:00

Added subscriber: @MohamedSakr

Mohamed Sakr commented

2015-06-12 14:22:57 +02:00

I think this is a normal behavior, indirect sampling won't work with branched PT, so if first hit -> it will sample all numbers in branched PT "direct" , other wise it will sample only 1 sample for indirect, which makes AA samples responsible for indirect noise.

I tested here with PT, cranked samples up to 16384, and it is noise free., took 8 minutes on GTX 780.

I think this is a normal behavior, indirect sampling won't work with branched PT, so if first hit -> it will sample all numbers in branched PT "direct" , other wise it will sample only 1 sample for indirect, which makes AA samples responsible for indirect noise. I tested here with PT, cranked samples up to 16384, and it is noise free., took 8 minutes on GTX 780.

Scott K commented

2015-06-12 18:48:03 +02:00

Thank you Mohamed, That makes sense.

I had Subsurface set to 32 samples, thinking I would get better results, but if the indirect doesn't work with branched PT, then that would explain why the indirect ss contributed so much noise, when all the other channels got so nice and smooth.

I'll keep my Samples settings low and lean heavily on the AA sampling.

Sorry to bother you guys, I consider this totally resolved and "not an issue"

Thank you Mohamed, That makes sense. I had Subsurface set to 32 samples, thinking I would get better results, but if the indirect doesn't work with branched PT, then that would explain why the indirect ss contributed so much noise, when all the other channels got so nice and smooth. I'll keep my Samples settings low and lean heavily on the AA sampling. Sorry to bother you guys, I consider this totally resolved and "not an issue"

Sergey Sharybin commented

2015-06-12 19:18:52 +02:00

With branched path tracing indirect lighting is getting sampled once for AA sample and once per each of SSS samples. So at this point i could understand why it might be noise in the combines pass, but increasing SSS samples should in theory reduce noise in SSS passes.

Sergey Sharybin commented

2015-08-27 02:07:34 +02:00

Changed status from 'Open' to: 'Archived'

Sergey Sharybin closed this issue

2015-08-27 02:07:34 +02:00

Sergey Sharybin commented

2015-08-27 02:07:34 +02:00

Finally had time to look into much more details.

So there are two sources of indirect light for SSS pass actually:

When ray just hits the surface indirect lighting for that point is evaluated. This light is being stored in the SSS indirect pass.
At each scatter event indirect light is also evaluated and stored in the SSS pass.

Increasing SSS samples will reduce noise caused by second case (when ray scatters and hits some bright light source i.e.). This does not affect noise caused by the first case and the only way to solve that noise is to increase AA samples.

Think some tricks are possible to avoid need to shoot full new AA sample to solve that noise, but that's not something i'll consider a bug.

Well, now it's at least clear what's going on for everyone :) So thanks for the report, but closing it now.

Finally had time to look into much more details. So there are two sources of indirect light for SSS pass actually: 1. When ray just hits the surface indirect lighting for that point is evaluated. This light is being stored in the SSS indirect pass. 2. At each scatter event indirect light is also evaluated and stored in the SSS pass. Increasing SSS samples will reduce noise caused by second case (when ray scatters and hits some bright light source i.e.). This does not affect noise caused by the first case and the only way to solve that noise is to increase AA samples. Think some tricks are possible to avoid need to shoot full new AA sample to solve that noise, but that's not something i'll consider a bug. Well, now it's at least clear what's going on for everyone :) So thanks for the report, but closing it now.

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Cycles - Subsurface Indirect render channel unable to resolve over-bright pixels from Branched Path Tracing #45014