SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
- Odilkhan Yakubov
- Posts: 208
- Joined: Fri Jan 26, 2018 10:07 pm
- Location: Tashkent, Uzbekistan
SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
Hi there. When I surfing on the internet I found a brend-new denoising algorithm which is called "Sample-based Monte Carlo Denoising using a Kernel-Splatting Network" aka SBMCD.
Original info from the authors:
Denoising has proven to be useful to efficiently generate high-quality Monte Carlo renderings. Traditional pixel-based denoisers exploit summary statistics of a pixel's sample distributions, which discards much of the samples' information and limits their denoising power. On the other hand, sample-based techniques tend to be slow and have difficulties handling general transport scenarios. We present the first convolutional network that can learn to denoise Monte Carlo renderings directly from the samples. Learning the mapping between samples and images creates new challenges for the network architecture design: the order of the samples is arbitrary, and they should be treated in a permutation invariant manner. To address these challenges, we develop a novel kernel-predicting architecture that splats individual samples onto nearby pixels. Splatting is a natural solution to situations such as motion blur, depth-of-field and many light transport paths, where it is easier to predict which pixels a sample contributes to, rather than a gather approach that needs to figure out, for each pixel, which samples (or nearby pixels) are relevant. Compared to previous state-of-the-art methods, ours is robust to the severe noise of low-sample count images (e.g. 8 samples per pixel) and yields higher-quality results both visually and numerically. Our approach retains the generality and efficiency of pixel-space methods while enjoying the expressiveness and accuracy of the more complex sample-based approaches.
DOWNLOAD:https://groups.csail.mit.edu/graphics/r ... oising.pdf
Note: I already asked Dade and Simon on working on this algorithm to implementation into LuxCore.
Original info from the authors:
Denoising has proven to be useful to efficiently generate high-quality Monte Carlo renderings. Traditional pixel-based denoisers exploit summary statistics of a pixel's sample distributions, which discards much of the samples' information and limits their denoising power. On the other hand, sample-based techniques tend to be slow and have difficulties handling general transport scenarios. We present the first convolutional network that can learn to denoise Monte Carlo renderings directly from the samples. Learning the mapping between samples and images creates new challenges for the network architecture design: the order of the samples is arbitrary, and they should be treated in a permutation invariant manner. To address these challenges, we develop a novel kernel-predicting architecture that splats individual samples onto nearby pixels. Splatting is a natural solution to situations such as motion blur, depth-of-field and many light transport paths, where it is easier to predict which pixels a sample contributes to, rather than a gather approach that needs to figure out, for each pixel, which samples (or nearby pixels) are relevant. Compared to previous state-of-the-art methods, ours is robust to the severe noise of low-sample count images (e.g. 8 samples per pixel) and yields higher-quality results both visually and numerically. Our approach retains the generality and efficiency of pixel-space methods while enjoying the expressiveness and accuracy of the more complex sample-based approaches.
DOWNLOAD:https://groups.csail.mit.edu/graphics/r ... oising.pdf
Note: I already asked Dade and Simon on working on this algorithm to implementation into LuxCore.
___________________________________________________________________________
LuxCoreRender Developer for Blender
___________________________________________________________________________
LuxCoreRender Developer for Blender
___________________________________________________________________________
Re: SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
Thanks, it looks interesting however Intel Oidn works well and it is one less problem to solve for us so, given our lack of resources, it is hard to consider anything outside an already available solution that works well and it is easy to integrate like Oidn. It is the same with Intel Embree.
If someone else want to work on this topic, sure, otherwise I will hardly find the time to look into it.
If someone else want to work on this topic, sure, otherwise I will hardly find the time to look into it.
- Odilkhan Yakubov
- Posts: 208
- Joined: Fri Jan 26, 2018 10:07 pm
- Location: Tashkent, Uzbekistan
Re: SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
Thank you Dade. I'm considering into it. It will be cool if someone implement this feature to the engine.
___________________________________________________________________________
LuxCoreRender Developer for Blender
___________________________________________________________________________
LuxCoreRender Developer for Blender
___________________________________________________________________________
Re: SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
Is there any benefit to this agains intel's OIDN? Results aren't that impressive
- Odilkhan Yakubov
- Posts: 208
- Joined: Fri Jan 26, 2018 10:07 pm
- Location: Tashkent, Uzbekistan
Re: SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
Why? Is it so unuseful thing, are you think so?
___________________________________________________________________________
LuxCoreRender Developer for Blender
___________________________________________________________________________
LuxCoreRender Developer for Blender
___________________________________________________________________________
- Odilkhan Yakubov
- Posts: 208
- Joined: Fri Jan 26, 2018 10:07 pm
- Location: Tashkent, Uzbekistan
Re: SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
@Dade, can you try to explain to us this algo in the example?
___________________________________________________________________________
LuxCoreRender Developer for Blender
___________________________________________________________________________
LuxCoreRender Developer for Blender
___________________________________________________________________________
Re: SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
it isn't that people think it as a bad feature or a super good feature. It is simply that we all need a reference point.Why? Is it so unuseful thing, are you think so?
How it compare against current working solution.past month the lux team change from BCD denoiser To intel OIDN because of better performance (memory / speed / quality).And before that we also tried to introduce Nvidia AI denoiser.
This community is far from being closed as Dade said human ressources here is the main issue.Dev have lot of work on the hands.
We are asking just to have better idea of the feature and improvement SBMCD bring on the table. just that.
please don't be offended
- FarbigeWelt
- Donor
- Posts: 1046
- Joined: Sun Jul 01, 2018 12:07 pm
- Location: Switzerland
- Contact:
Re: SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
Are you sure? Did you look at the pictures on p. 125:11 of the paper linked on main post?
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
MacBook Air with M1
- FarbigeWelt
- Donor
- Posts: 1046
- Joined: Sun Jul 01, 2018 12:07 pm
- Location: Switzerland
- Contact:
Re: SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
What does SBMCD mean? What is its purpose?
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
MacBook Air with M1
Re: SBMCD (Sample-based Monte Carlo Denoising using a Kernel-Splatting Network)
I only skimmed through the paper quickly but one interesting feature I noticed is that this denoiser doesn't require a clean albedo AOV.