LuxCoreRender Forums

Hi ! just open this thread for testing and learning purposes about DLCS as said in the title. I wish to test here many diferent test scene where DLCS shine.

this is my first scene. it is yet unfinished but want to understand more before i add more stuff.

.Why DLCS lighting is so diferent from log power/power strategy ? it deliver so much light

. I also have some strange black artifact

EDIT

Scene file here :
https://drive.google.com/open?id=1237dq ... EQh9Sg5adH

. I also have some strange black artifact

Dade wrote: ↑Wed Jul 04, 2018 9:19 pm

B.Y.O.B. wrote: ↑Wed Jul 04, 2018 5:56 pm As far as I understand it, the light cache only adjusts the probabilities of a light being sampled and is still unbiased.

Correct.

B.Y.O.B. wrote: ↑Wed Jul 04, 2018 5:56 pm The phenomenon in the linked image might be startup bias or something else caused by not enough samples (but it is a bit hard to tell without a ground truth image).

It is noise, a (huge) different level of noise. DLSC fallback to standard log power light strategy when a valid cache entry is not avilable for the point I'm rendering. It is very noticeable. Look, this is a scene with a well populated cache (1 sample per pixel):


im1.jpg


and this is the same scene with not enough cache entries:


im2.jpg


The surfaces without a valid cache entry are very easy to spot due the HUGE difference in noise. It is still unbiased so it will converge to the same result but in practice they look as artifacts.

Have you tried a smaller entry radius?
(nice scene btw.)

I'm looking into the scene but before to start I'm checking a strange problem I have when exporting the scene in text format.

B.Y.O.B. wrote: ↑Thu Jul 05, 2018 7:52 am Have you tried a smaller entry radius?
(nice scene btw.)

thanks you glad you like it !
here is a render with entry 10cm and max passes to 10000 : As can see there is black artifacts on the ladder. It also lead to longer precomputation time. And falling to very low entry like 2cm take forever to compute and freeze blender.

I think,
- the smaller the radius, the more samples there are & the more computation is needed
- dark artifacts, from my observations happen only on shadow's edges under some special (?) conditions (most distinctive with shadow catcher used).

Another test both rendered in 2hrs and again light cache take a very very long time to compute when it's under around 10cm :

.DLCS :


.UNIFORM :

Sharlybg wrote: ↑Wed Jul 04, 2018 3:59 pm Why DLCS lighting is so diferent from log power/power strategy ? it deliver so much light

There was a bug in area light source power computation (when the using power/efficiency parameters) causing power light strategies to erroneously sample only the soft area light outside the windows.

I should have fixed this problem.

Dade wrote: ↑Sun Jul 08, 2018 9:02 am

Sharlybg wrote: ↑Wed Jul 04, 2018 3:59 pm Why DLCS lighting is so diferent from log power/power strategy ? it deliver so much light

There was a bug in area light source power computation (when the using power/efficiency parameters) causing power light strategies to erroneously sample only the soft area light outside the windows.

I should have fixed this problem.

Thanks for solving this. Also do you think the longer precomputation time is a normal behaviour ? And can we do valid test DLCS with the current development stage ?

Sharlybg wrote: ↑Thu Jul 26, 2018 9:34 pm Thanks for solving this. Also do you think the longer precomputation time is a normal behaviour ? And can we do valid test DLCS with the current development stage ?

I wonder too how complete/incomplete this feature is. In some cases even small entry radius gives me visible sample areas (2.1alpha3) where the edge of the shadow is just "blotchy" it doesn't matter if I use sun, sky, hdri or area lights.