Page 2 of 5

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 11:42 am
by alpistinho
Things are looking better, but there are plenty of visible patterns indeed for now.
Screenshot from 2019-03-06 08-41-10.png

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 11:55 am
by alpistinho
I think it is converging to the right solution now. It also helps when you don't calculate the same constant every time =p
Screenshot from 2019-03-06 08-54-13.png

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 12:01 pm
by Dade
alpistinho wrote: Wed Mar 06, 2019 11:55 am I think it is converging to the right solution now. It also helps when you don't calculate the same constant every time =p
Now you have to establish if it is of any use :mrgreen:

Does it converge faster than Sobol sampler (i.e. less noise in the same amount of samples) ?

P.S. people will endless complain about patterns at the start but it may be still of some use if it is noticeably better than Sobol.

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 12:02 pm
by lacilaci
Is there a build to test this out?

PS: patterns only matter if it throws off denoiser (this could damage preview quality)

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 12:03 pm
by alpistinho
Dade wrote: Wed Mar 06, 2019 12:01 pm Now you have to establish if it is of any use :mrgreen:

Does it converge faster than Sobol sampler (i.e. less noise in the same amount of samples) ?

P.S. people will endless complain about patterns at the start but it may be still of some use if it is noticeably better than Sobol.
I still think there are some other stupid mistakes on the code, just not as severe.

I will try to do some comparisons to see if it is any better, but at least the samples/second metric is bit higher.

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 12:13 pm
by alpistinho
lacilaci wrote: Wed Mar 06, 2019 12:02 pm Is there a build to test this out?

PS: patterns only matter if it throws off denoiser (this could damage preview quality)
I can send you my build files if you really want, but I don't know if it is of any use

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 12:16 pm
by lacilaci
alpistinho wrote: Wed Mar 06, 2019 12:13 pm
lacilaci wrote: Wed Mar 06, 2019 12:02 pm Is there a build to test this out?

PS: patterns only matter if it throws off denoiser (this could damage preview quality)
I can send you my build files if you really want, but I don't know if it is of any use
Actually now that I think about it, I would need to have the sampler exposed in blender to properly test that.

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 12:24 pm
by B.Y.O.B.
lacilaci wrote: Wed Mar 06, 2019 12:16 pm Actually now that I think about it, I would need to have the sampler exposed in blender to properly test that.
You just have to edit this line in export/config.py: https://github.com/LuxCoreRender/BlendL ... fig.py#L40
Instead of the variable sampler, pass the string directly.
Alpistinho has called the new sampler "R2": https://github.com/LuxCoreRender/LuxCor ... 0fa71R1342

So you write:

Code: Select all

	definitions.update({
            "renderengine.type": luxcore_engine,
            "sampler.type": "R2",
            "film.width": width,
            "film.height": height,
            # ...
After this, you can test the new sampler in final renders only (F12).

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 12:30 pm
by lacilaci
Great, so if there is a build I'm willing to give it a go.

Re: Quasirandom sampler

Posted: Wed Mar 06, 2019 12:43 pm
by alpistinho
The pyluxcore.pyd file is not being built by default on my environment it seems. I will have a look at this when the sampler is working a little bit better.

Right now I believe all threads are producing the same sequence, so there is pretty low-hanging fruit to improve