Page 7 of 8

Re: CGI tech news box

Posted: Tue Oct 23, 2018 12:36 am
by B.Y.O.B.

Re: CGI tech news box

Posted: Mon Oct 29, 2018 10:04 am
by lacilaci
Maybe this could be better and faster than current solution? https://www.disneyresearch.com/publicat ... denoising/

Re: CGI tech news box

Posted: Wed Nov 07, 2018 5:20 pm
by kintuX

Re: CGI tech news box

Posted: Wed Nov 07, 2018 7:45 pm
by Sharlybg
Could it be... the next step for light caching? 8-)
Already seen it before but decide to look closely at it today and it is really really mind blowing. Also notice that the more probes you have better will be the quality and rendering are always under 1scs. so we can match Brute Path tracing in just minute easily or with Beefy CpU or Opencl version on GPU in just few sec.

Re: CGI tech news box

Posted: Thu Nov 08, 2018 10:24 am
by lacilaci
Looks like a good solution for games, maybe if 2.8-eevee would have this it could be pretty nice.
I'm not sure if this is a solution for caching that would work in tandem with pathtracing. Looks more like a completely different approach for GI than something you could mix with pt. it was impressive how it worked with glossy reflections though

Seeing how well recently DLCS works I think if that could be used as basis for ILCS(which I guess is how it's gonna be) it will be best solution. But I wonder if it also can help with rough glossy reflections similar to vray's lightcache.

Re: CGI tech news box

Posted: Thu Nov 08, 2018 10:33 am
by Sharlybg
But I wonder if it also can help with rough glossy reflections similar to vray's lightcache.
Normally upcoming ILCS is more robuste than Vray light cache (it is suposed to be unbiased) so i expect less limitation.

Re: CGI tech news box

Posted: Sun Nov 11, 2018 4:26 am
by oldladyfear
"A Simple and Robust Mutation Strategy for the Metropolis
Light Transport Algorithm"

http://sirkan.iit.bme.hu/~szirmay/paper ... tronic.pdf

looks like it could be a simple way to increase the performance with little overhead.

Re: CGI tech news box

Posted: Sun Nov 11, 2018 3:48 pm
by kintuX
Image
Multiple importance sampling seems to work with reconstruction. Three sampling strategies fused by MIS: central pixel BRDF sampling, neighbor-stealing BRDF sampling, light sampling. Here in glorious 15fps on my GeForce 940M :P It's all specular btw, even the diffuse-looking walls
https://twitter.com/h3r2tic/status/1061259372884443138

Re: CGI tech news box

Posted: Sun Nov 11, 2018 8:08 pm
by patrickawalz
oldladyfear wrote:
Sun Nov 11, 2018 4:26 am
"A Simple and Robust Mutation Strategy for the Metropolis
Light Transport Algorithm"

http://sirkan.iit.bme.hu/~szirmay/paper ... tronic.pdf

looks like it could be a simple way to increase the performance with little overhead.
im pretty sure this idea the kelemen mlt paper. a variation of this is already implemented at the heart of luxcore and is available as the metropolis sampler

Re: CGI tech news box

Posted: Mon Nov 12, 2018 4:27 pm
by lacilaci
http://drz.disneyresearch.com/~jnovak/p ... index.html

obviously ,i don't know much about this but If I knew how, I'd be willing to spend free time training the denoiser for luxcore.