Laci's wip

Post your tests, experiments and unfinished renderings here.
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: Laci's wip

Post by B.Y.O.B. »

lacilaci wrote: Fri Dec 07, 2018 10:10 pm It seems that both uniform and logpower not only redistribute samples but completely shift exposure/balance.
Does this also happen if you disable clamping (in a noise-free render)?
User avatar
lacilaci
Donor
Donor
Posts: 1969
Joined: Fri May 04, 2018 5:16 am

Re: Laci's wip

Post by lacilaci »

B.Y.O.B. wrote: Fri Dec 07, 2018 11:46 pm
lacilaci wrote: Fri Dec 07, 2018 10:10 pm It seems that both uniform and logpower not only redistribute samples but completely shift exposure/balance.
Does this also happen if you disable clamping (in a noise-free render)?
Haven't tried yet. Will test some more to find what's happening.
User avatar
lacilaci
Donor
Donor
Posts: 1969
Joined: Fri May 04, 2018 5:16 am

Re: Laci's wip

Post by lacilaci »

Ok so I made small testscene that reproduces the issue, you were right this is clamping related.

So this is what happens, with multiple lights and unclamped rendering every method (or it looks like it) gives similar and proper looking result.
The moment you enable clamping only power looks like it's properly working, dlcs, uniform and logpower are completely messed up.
Problem is visible in higher values of clamping too! But in the test scene to make it much more obvious I put very low value
So open the scene and testrender with power and then other lightstrategies.
Only power looks properly clamped, and I wouldn't be surprised if this issue is also exagerrating the cache splotches problem.

EDIT:Regarding this issue being related to dlcs problems: even with high clamping I'm seeing some lightcache splotches, but unclamped everything is clean as it gets. I'm sure these two issues are related.

EDIT2: If this could be fixed, and we'd get as reliable other light strategies with clamping as with power and also fix the splotches issue we could even have default clamping value that users might not need to touch in 90% cases!! Granted, sun needs to be redesigned as it's a horror to use it by default, terrible color controls and insane power that is not needed in 99% of cases(there should be a checkbox "use real sun power" and different default behavior) but that's a different topic for later..
Attachments
lstratest.blend
(1.53 MiB) Downloaded 160 times
User avatar
lacilaci
Donor
Donor
Posts: 1969
Joined: Fri May 04, 2018 5:16 am

Re: Laci's wip

Post by lacilaci »

So after some more testing I think it's all wrong...

Clamping with power light strategy kills weak lights completely, while clamping with logpower and uniform kills strong lights and with DLCS even introduces some heavy splotches in shadow borders...
This is really bad and I tried rendering regions until they're clean and it never helps.\
Wouldn't it help to clamp only indirect lighting? I mentioned this few times already I don't thing there is anything to gain clamping direct light contribution.
And why does light strategy even affects how clamping reacts?

EDIT: @Dade: you mentioned there is a problem with MIS when splotches appear with DLCS, isn't all this I'm observing here a problem with MIS in general? I can clearly recover killed highlights from sun(by clamping and logpower) simply making higher importance for sun... This is looking all kinds of wrong to me to be honest. It is pretty impossible to setup some proper lighting with this.
Attachments
clogpower.jpg
cpower.jpg
User avatar
lacilaci
Donor
Donor
Posts: 1969
Joined: Fri May 04, 2018 5:16 am

Re: Laci's wip

Post by lacilaci »

Here.. Now I'm using both logpower and clamping and I have weak and strong lights contributing as much as needed(this is just eyeballing it)
And I got this result by setting world importance to 4 and sun to 10!!! wtf, is this really how importance works? Cause I though it could make some lights more noisy/harder to resolve but this completely shifts lighting and I'm now 100% sure this is what is breaking your DLCS with all that splotches..
Attachments
clogpower_manualimportance.jpg
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: Laci's wip

Post by B.Y.O.B. »

It could just be caused by the behaviour of clamping that Dade explained here: viewtopic.php?f=4&t=700&p=7221#p7221
But I'm not entirely sure.
User avatar
lacilaci
Donor
Donor
Posts: 1969
Joined: Fri May 04, 2018 5:16 am

Re: Laci's wip

Post by lacilaci »

B.Y.O.B. wrote: Sat Dec 08, 2018 6:17 pm It could just be caused by the behaviour of clamping that Dade explained here: viewtopic.php?f=4&t=700&p=7221#p7221
But I'm not entirely sure.
I really don't know. But it doesn't make any sense to have to change light importance to correct lighting because clamping destroys it. This is as counter-intuitive as it gets imho.
User avatar
lacilaci
Donor
Donor
Posts: 1969
Joined: Fri May 04, 2018 5:16 am

Re: Laci's wip

Post by lacilaci »

toying around with different importance settings per lights... Can recover sun and performance is great. Still using very low clamping values but it helps to see what takes most damage from it...
Attachments
importancesettings.jpg
User avatar
lacilaci
Donor
Donor
Posts: 1969
Joined: Fri May 04, 2018 5:16 am

Re: Laci's wip

Post by lacilaci »

So I'm experimenting a bit with carpets/rugs.. and while it renders ok and memory is more or less ok with dense hair particles I have some issue copying/instancing... I can seem to be able to make an instance of object(carpet) with hair so that it don't eat more memory. I know there was an issue with instances, but wasn't it fixed? Is it not possible to create instance of an object with particles/hair?
Attachments
rugs_stresstest.jpg
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: Laci's wip

Post by B.Y.O.B. »

Thanks for the report.
See https://github.com/LuxCoreRender/BlendL ... issues/228, I will try to fix it after the port to 2.8.
Post Reply