DLCS and overlapping entries

Use this forum for general user support and related questions.
Post Reply
lacilaci
Posts: 342
Joined: Fri May 04, 2018 5:16 am

DLCS and overlapping entries

Post by lacilaci » Sun Nov 04, 2018 9:33 am

artifacting with multiple lights...

Not sure if that's the case, but a single light doesn't cause the issue, only if I have multiple lights overlapping.

I've used latest daily that fixes previous issues when rendering round objects etc /direct light cache works almost perfectly now...

Again this is a simplified scene showing issues I've seen in another scene, note that except the visible spots areas, there are also less visible spot areas that are just noisy and resolving very slowly. (tried changing warmup samples, but that doesn't seem to affect the issue)
Attachments
dlcs_issue2.jpg
luxcoreDLCS_issue2.blend
(3.59 MiB) Downloaded 1 time

lacilaci
Posts: 342
Joined: Fri May 04, 2018 5:16 am

Re: DLCS and overlapping entries

Post by lacilaci » Sun Nov 04, 2018 9:55 am

Ok, a small update.

It seems that it is a problem that is more visible with stronger clamping. It is almost gone (if not completely gone) with higher clamping values or if lights don't overlap.

Now it would be really good if this dlcs feature worked well with clamping. Although I still think direct and indirect light contribution should be decoupled for clamping.

EDIT: also high values for clamping show this issue too, but it is barely visible until rendering is almost clean. This could be a very big problem, because in a more complex scene, preview rendering would not show the issue and then the user would render for 1-2 hours and in a final rendering see some weird spots.

Other than this, I am blown away how well this seems to work :)

User avatar
Dade
Developer
Posts: 1457
Joined: Mon Dec 04, 2017 8:36 pm

Re: DLCS and overlapping entries

Post by Dade » Sun Nov 04, 2018 12:15 pm

The problem seems to be triggered by an interaction of different level of noise cause by different DLSC cache entries and a very very very small clamping value. It is a quite strange interaction but it looks like it is cause only by the variance clamping value you are using.
You are using clamp value of 100 while your light is in the range 5000-10000. The result is, as explained in another thread, that variance clamping will be very very slow to reach its working "window".
This is a rendering with 1000 clamping value (and colored lights for debugging):

ok.png

It looks correct to me (and 1000 is still a quite small value).

In the meantime, I have also fixed some incorrect computation in the relative weights of light sampling in DLSC so you may want to upgrade to a new daily build (in a couple of hours).
Support LuxCoreRender project with salts and bounties

lacilaci
Posts: 342
Joined: Fri May 04, 2018 5:16 am

Re: DLCS and overlapping entries

Post by lacilaci » Mon Nov 05, 2018 10:14 am

Dade wrote:
Sun Nov 04, 2018 12:15 pm
The problem seems to be triggered by an interaction of different level of noise cause by different DLSC cache entries and a very very very small clamping value. It is a quite strange interaction but it looks like it is cause only by the variance clamping value you are using.
You are using clamp value of 100 while your light is in the range 5000-10000. The result is, as explained in another thread, that variance clamping will be very very slow to reach its working "window".
This is a rendering with 1000 clamping value (and colored lights for debugging):


ok.png


It looks correct to me (and 1000 is still a quite small value).

In the meantime, I have also fixed some incorrect computation in the relative weights of light sampling in DLSC so you may want to upgrade to a new daily build (in a couple of hours).
yup, everything seems to work nicely now.

Post Reply