Path Space Regularization (aka the solution to SDS paths)

[Dade]

I have written a "proof of concept" based on "Path Space Regularization for Holistic and Robust Light Transport" (https://cg.ivd.kit.edu/english/PSR.php). This should represent a solution to the problem of rendering SDS paths with Hybrid Back/Forward path tracing.

This is a rendering of my standard SDS/caustics test scene with normal path tracing:

No caustics, no SDS paths, just few fireflies here and there.

This is the same scene rendered with light tracing:

Nice caustics but light tracing is not able to render directly visible specular surfaces at all.

This is the same scene rendered with light tracing and a very first hacked code for "Path Space Regularization":

Being able to render a mirror (even if blurred) with a light tracer is quite remarkable result. Indeed, I was not looking for an alternative (blurred) way to render mirrors but to the ground caustic reflected on the mirror (i.e. the SDS paths). That piece of information will enrich Hybrid Back/Forward path with the capability to render SDS paths.

This feature is schedule for v2.3.

P.S. I will held hostage the code until the final release of v2.2

[lacilaci]

This feature is schedule for v2.3.

P.S. I will held hostage the code until the final release of v2.2

Aw yeaah...
So let's release 2.2 today evening and first 2.3 daily tomorrow morning, there you go problem solved

Ok, but seriously, this looks very promising!

[Dade]

So let's release 2.2 today evening and first 2.3 daily tomorrow morning, there you go problem solved

I will be out of the town for few days

P.S. v2.2 starts to look like a final release, we shouldn't be too far from the release date.

[epilectrolytics]

I have written a "proof of concept" based on "Path Space Regularization for Holistic and Robust Light Transport" (https://cg.ivd.kit.edu/english/PSR.php). This should represent a solution to the problem of rendering SDS paths with Hybrid Back/Forward path tracing.

Thanks for the teaser, looks very promising!
In the paper they mention it runs on GPU, Iooks like it will also be a fast solution

And the paper is from 2013, has this been implemented elsewhere or will LuxCore be first?

[Dade]

In the paper they mention it runs on GPU, Iooks like it will also be a fast solution

It will run on CPU in our case because the light tracing is done there in our Hybrid Back/Forward path tracing but it isn't a big deal because the CPU will really handle only caustics and SDS paths and nothing else.

And the paper is from 2013, has this been implemented elsewhere or will LuxCore be first?

It has been implemented in a couple of demo but I'm not aware of its usage in any big commercial render engine.

[provisory]

I really like this direction!

I've found an easy to understand comparison video between VCM and Path Space Regularization:
https://www.youtube.com/watch?v=Hc9zu5-O7Eo

So, if I understand correctly, the degree of the "blur" is adjustable, and it is the quality vs. speed variable here.

[Dade]

So, if I understand correctly, the degree of the "blur" is adjustable, and it is the quality vs. speed variable here.

Yes, more initial bias => more blurred =>"faster" but it is unbiased so it will converge to the right solution (first or later).

[provisory]

Very exciting, can't wait to test it!

[Sharlybg]

Nice progresse Dade. Thanks for sharing such nice info to us.

Btw why the psor glass look so dark in the new algo aka PSR ? I mean compared to regular path tracing render.

[Dade]

This is the first rendering with integration between hybrid backward/forward path tracing and path space regularization:

It is good but a bit too blurred/noised/slow to render SDS paths.