Page 9 of 109

Re: PhotonGI cache

Posted: Fri Jan 25, 2019 11:30 am
by Dade
BTW, I forgot to write that my work on Metropolis is based on "Robust Adaptive Photon Tracing using Photon Path Visibility" by TOSHIYA HACHISUKA and HENRIK WANN JENSEN: https://www.ci.i.u-tokyo.ac.jp/~hachisuka/amcmcppm.pdf

Re: PhotonGI cache

Posted: Fri Jan 25, 2019 12:15 pm
by lacilaci
Dade wrote: Fri Jan 25, 2019 11:30 am BTW, I forgot to write that my work on Metropolis is based on "Robust Adaptive Photon Tracing using Photon Path Visibility" by TOSHIYA HACHISUKA and HENRIK WANN JENSEN: https://www.ci.i.u-tokyo.ac.jp/~hachisuka/amcmcppm.pdf
The question is now... will caustic rendering become "mainstream" in luxcore before corona? :D

Re: PhotonGI cache

Posted: Fri Jan 25, 2019 3:42 pm
by Dade
I added the support for building and using visibility information with Metropolis. This is a zoom out view of the cache for Psor's cube without using visibility information:

direct-novis.jpg

All the ground surface is covered with photons. The same cache using the visibility information:

direct-vis.jpg

99% of photons lands only on the seen rectangle of the ground (with the original camera position).

This is a quite hard rendering with a light source outside the room and light coming in only trough small rectangular holes, with Random sampler and no visibility information:

rnd.jpg

With Metropolis but no visibility information:

mnovis.jpg

And with Metropolis and visibility information:

mvis.jpg

Clearly plain Metropolis is better than Random and Metropolis+Visibility is better than plain Metropolis.

The Visibility information is currently build with a single thread, next step multi-thread support.

Re: PhotonGI cache

Posted: Fri Jan 25, 2019 4:27 pm
by epilectrolytics
:shock: Great progress, very exciting!

How will this work with displacement or objects with high instance count blowing cache size out of proportion, can those be excluded or some kind of averaging applied?

Re: PhotonGI cache

Posted: Fri Jan 25, 2019 4:33 pm
by Dade
epilectrolytics wrote: Fri Jan 25, 2019 4:27 pm How will this work with displacement or objects with high instance count blowing cache size out of proportion, can those be excluded or some kind of averaging applied?
You can disable the cache on a per material base in order to render nearly-specular glossy surfaces or surfaces with too many details:
viewtopic.php?f=5&t=840&start=30#p8885

With cache disable, normal path tracing will be applied, bouncing over the surfaces and probably hitting another surface where cache is enabled (so still indirectly benefiting of the cache speedup).

Re: PhotonGI cache

Posted: Fri Jan 25, 2019 4:39 pm
by lacilaci
does it work well with clamping?

Re: PhotonGI cache

Posted: Fri Jan 25, 2019 4:42 pm
by epilectrolytics
Dade wrote: Fri Jan 25, 2019 4:33 pm With cache disable, normal path tracing will be applied, bouncing over the surfaces and probably hitting another surface where cache is enabled (so still indirectly benefiting of the cache speedup).
:mrgreen: Ingenious!

Re: PhotonGI cache

Posted: Fri Jan 25, 2019 10:27 pm
by Sharlybg
I'm afraid all this is going too far. Can we stop now ?

Nooooooooooooo :twisted:

Re: PhotonGI cache

Posted: Fri Jan 25, 2019 11:22 pm
by Dade
lacilaci wrote: Fri Jan 25, 2019 4:39 pm does it work well with clamping?
As far as I have seen, yes, there no particular problems with clamping.

Re: PhotonGI cache

Posted: Sat Jan 26, 2019 8:44 am
by epilectrolytics
Dade wrote: Fri Jan 25, 2019 11:22 pm As far as I have seen, yes, there no particular problems with clamping.
Don't know if it is a clamping issue, but for the caustic cache clamp value = 1 is required and in my test scene caustics are muted and the light balance is very different when compared with a bidir metro render. :(
But for the first time we get fast caustics with correct SDS reflections, yay! :D
Also the innovative metropolis photon distribution works great, no problem with insane photon counts for smooth caustics.
.
photoncachetest.gif
.
cachetest_LuxCore.zip
(45.81 KiB) Downloaded 209 times