Page 1 of 1

BIDIRVMCPU

Posted: Wed Jun 06, 2018 2:54 pm
by jgrover110
Is there anything planned for the BIDIRVMCPU renderer, such as use with Blendluxcore or improvements to the engine, or will it stay as is?

Re: BIDIRVMCPU

Posted: Wed Jun 06, 2018 4:30 pm
by Dade
jgrover110 wrote: Wed Jun 06, 2018 2:54 pm Is there anything planned for the BIDIRVMCPU renderer, such as use with Blendluxcore or improvements to the engine, or will it stay as is?
There is a problem in using vertex merging and Metropolis sampler at the same time. The problem is between the multi-pass nature of VM and Metropolis. I would say the answer is no, nothing is planned until the problem is solved.

As far I remember BIDIRVCPUVM is not exposed by BlendLuxCore or cited in any documentation :?:

Re: BIDIRVMCPU

Posted: Wed Jun 06, 2018 4:44 pm
by jgrover110
There is very little info about the engine at all. Could you tell me what units the start radius scale are in (bidirvm.startradius.scale)? Would it be possible for it to be in screen space pixels?

Re: BIDIRVMCPU

Posted: Wed Jun 06, 2018 6:01 pm
by Dade
jgrover110 wrote: Wed Jun 06, 2018 4:44 pm There is very little info about the engine at all. Could you tell me what units the start radius scale are in (bidirvm.startradius.scale)? Would it be possible for it to be in screen space pixels?
As far as I remember is in world space unit.

Re: BIDIRVMCPU

Posted: Wed Oct 17, 2018 12:18 pm
by epilectrolytics
Dade wrote: Wed Jun 06, 2018 4:30 pmThere is a problem in using vertex merging and Metropolis sampler at the same time. The problem is between the multi-pass nature of VM and Metropolis. I would say the answer is no, nothing is planned until the problem is solved.
I understand the problem is still not solved?
Being stuck in my attempts to get caustics rendered with correct reflections, in BiDir metro they are awfully slow and in BiDirVMmetro caustics get too bright with time (especially when used with sunsky). BiDirVMrandom and Pathmetro get it right but again awfully slow so there is currently no way to do it. :(


bidirVM.gif

Re: BIDIRVMCPU

Posted: Fri Oct 19, 2018 9:46 am
by FarbigeWelt
Have you tried with clamping around 50‘000?
Or adjusting settings for metropolis? E.g. 5%, 128 or 8192, 0.5%
Tonenapper Reinhard: 1.0,1.0,2.5 to 4.0? (automatic swiched off on tab for linearity)

Re: BIDIRVMCPU

Posted: Fri Oct 19, 2018 12:35 pm
by epilectrolytics
FarbigeWelt wrote: Fri Oct 19, 2018 9:46 am Have you tried with clamping around 50‘000?
Or adjusting settings for metropolis? E.g. 5%, 128 or 8192, 0.5%
Tonenapper Reinhard: 1.0,1.0,2.5 to 4.0? (automatic swiched off on tab for linearity)
With Bidir+metro I tried all possible combinations (2%, 8192, 80% worked better) but never did the reflection catch up with the caustics.
While caustics are usable after 100 samples (5-15 min), the reflections need 10 000 or more (5+hours).
I installed again old LuxRender 1.6 on my main system (10.12.6), Bidir there renders caustics and SDS reflections equivalently (1000samples ~1h).

BidirVM would be the solution, it's fast and gets those pesky SDS paths, but with this energy conservation problem (getting too bright) it's still unusable.
Once the light balance is off, no tone mapping can adjust for it and clamping is not yet available for BidirVM.

I'm planning to get a PC with an rtx 2070 for xmas and than we'll see what path metro OCL will make of those 10 000+ samples ...