I have a scene with simple geometry, but complex lighting and materials.
The glass is modeled to resemble the IKEA Planera glass and the material is the Lead Crystal from the material database.
The floor is the red velvet material also from the material database. I changed that to violet color.
There are two light layers. One uses the Circus Arena HDRI from Plolyhaven . This one provides the "general lighting"
The other light layer is assigned to a mesh emitter in the upper right corner with an IES profile for a spotlight. This light source produces distinct caustics.
GPU path tracing is pretty useless for such a scene.
Using BIDIR MLT without any Contribution Clamping rendered to 12770 SPP
Using BIDIR MLT with contribution clamping at 500 rendered to 6800 SPP
Using BIDIR MLT with COntribution clamping at 50 rendered to 3490 SPP
All of the above images were rendered with a Supersampling factor of 4.
Clearly, Contribution Clamping does not only reduce noise and fireflies but also a lot of "lighting detail" for lack of better words.
Does contribution clamping introduce bias ?
- thesquirell
- Posts: 428
- Joined: Fri Aug 29, 2014 3:49 am
- Location: Novi Sad, Serbia
Re: Does contribution clamping introduce bias ?
It sure does:
Please note that, since it is clamping bright spots, it introduces bias into the render. Because of this it will be off by default. However it's a useful tool for artists, to remove fireflies from their images in a simple and efficient way.
- Indigo Manual, Render Settings
Please note that, since it is clamping bright spots, it introduces bias into the render. Because of this it will be off by default. However it's a useful tool for artists, to remove fireflies from their images in a simple and efficient way.
- Indigo Manual, Render Settings
Re: Does contribution clamping introduce bias ?
@thesquirell Thanks for pointing that out! I just looked through that section of the manual the other day and this totally escaped my attention. I had assumed it would create bias.
It's a pitty though as that very much limits my options for getting an accurate but also mostly noise free image.
I would need a lot of patience paired with a lot of CPU cores. I have neither
It's a pitty though as that very much limits my options for getting an accurate but also mostly noise free image.
I would need a lot of patience paired with a lot of CPU cores. I have neither
- thesquirell
- Posts: 428
- Joined: Fri Aug 29, 2014 3:49 am
- Location: Novi Sad, Serbia
Re: Does contribution clamping introduce bias ?
I know, right?!
Did you try GPU with bigger depth numbers? Man, I can't wait til they add BiDirMLT to GPU for those nice caustics...
Did you try GPU with bigger depth numbers? Man, I can't wait til they add BiDirMLT to GPU for those nice caustics...
Re: Does contribution clamping introduce bias ?
For all of the images above the "depth numbers" were set to 99. But as I understand it, that parameter only applies to GPU rendering But I've tried a depth of 32 there with no change.thesquirell wrote: ↑Wed Jan 05, 2022 3:11 pmI know, right?!
Did you try GPU with bigger depth numbers? Man, I can't wait til they add BiDirMLT to GPU for those nice caustics...
This scene has many small, high intensity light sources and I can see why simple path tracing would take ages to converge.
I would already be happy if BIDIT without MLT can be implemented on the GPU. Who knows ...
There is basically no interaction with the developers on his forum.
Thank you for the replies!
Re: Does contribution clamping introduce bias ?
More interesting results. I rendered the image with LuxCore render and Indigo with what I believe to be similar settings.
Both use BIDIR + MLT on the CPU. I am aware that this is isn't an apples to apples comparison, but the difference in results is still interesting.
In order to reduce memory consumption I merged the 2 light layers in Indigo. That allowed me to increase super sampling to 8. That significantly reduced the noise in the image. Both images were rendered for roughly the same time.
Both render engines allow contribution clamping but that was turned off.
The LuxCore image is much less noisy but also contains a lot less information.
Most if not all of the reflections from the myriad of little lights in the HDRI one would expect to see on the outside of the glass are completely missing.
The Indigo image on the other hand contains a lot more of the aforementioned reflections and those seem to be what is contributing to the noise. Although, the noise level seems to have gone down significantly after I merged the 2 light layers and the increased the super sampling to 8.
Both use BIDIR + MLT on the CPU. I am aware that this is isn't an apples to apples comparison, but the difference in results is still interesting.
In order to reduce memory consumption I merged the 2 light layers in Indigo. That allowed me to increase super sampling to 8. That significantly reduced the noise in the image. Both images were rendered for roughly the same time.
Both render engines allow contribution clamping but that was turned off.
The LuxCore image is much less noisy but also contains a lot less information.
Most if not all of the reflections from the myriad of little lights in the HDRI one would expect to see on the outside of the glass are completely missing.
The Indigo image on the other hand contains a lot more of the aforementioned reflections and those seem to be what is contributing to the noise. Although, the noise level seems to have gone down significantly after I merged the 2 light layers and the increased the super sampling to 8.
Who is online
Users browsing this forum: No registered users and 116 guests