Path Space Regularization (aka the solution to SDS paths)

[lacilaci]

to get some caustics going in my main scene I have to start with some incredible values for scale, like 10000 so I guess those parameters are very scene dependant...?

well.. you can't see what is going on in the pool but it got close to proper brightness, even pinacollada is no longer black(though still wrong)

[Dade]

to get some caustics going in my main scene I have to start with some incredible values for scale, like 10000 so I guess those parameters are very scene dependant...?

Distance from the camera is one of the factors so it matters in a scene of 1m Vs 1Km

[Dade]

clamping seem to skew the results even though it shouldn't affect lighttracing

But are you using GPU rendering (no one has yet written the support for PSR) ? Because clamping is disabled for light tracing on CPU (https://github.com/LuxCoreRender/LuxCor ... d.cpp#L151).

Mmm, I have the feeling you are using a clamping value extremely smaller of what should be and it is not the light tracing to "explode" but it is eye tracing to go boom: try to do a rendering disabling PSR and Hybrid, does still happen ?

[lacilaci]

clamping seem to skew the results even though it shouldn't affect lighttracing

But are you using GPU rendering (no one has yet written the support for PSR) ? Because clamping is disabled for light tracing on CPU (https://github.com/LuxCoreRender/LuxCor ... d.cpp#L151).

Mmm, I have the feeling you are using a clamping value extremely smaller of what should be and it is not the light tracing to "explode" but it is eye tracing to go boom: try to do a rendering disabling PSR and Hybrid, does still happen ?

I did use opencl yes... I can't do more testing now ,but I will play around with cpu and different settings in the morning and post results here.

[Fox]

I'm using the hetero volume type water in pool test scene, and there is huge speed increase when i roll back this commit:
https://github.com/LuxCoreRender/LuxCor ... aed3bf5cfb

Code: Select all

Water Material

Hetero
Multi
Step 50cm
Steps 32

abs
0.1882334886 0.0864110145 0.005475
inverted to T for Blender exporter: 0.8117665114 0.9135889855 0.994525

sss
0.18804175 0.08702118 0.14913043

g
0.929 0.910 0.945

cauchy A 1.326203018904
cauchy B 0.002152008167

[lacilaci]

ok so this is cpu only (looks same with clamped or unclamped rendering)

so the forever darkening is present always...

But I wouldn't be sure if that's the PSR fault. I think it's either volumes or glass rendering that's causing this.

Look at this example:
PSR

without PSR

the sharp outer ring only appears when using PSR(or bidir) but it never disappears or get weaker, why would psr behave differently here?

[lacilaci]

hm.. tried over 2000 samples in a mirror scene to see if sds will disappear and they do. (CPU)

[Dade]

So it is not related to clamping ? I assume your speed parameter is too low (i.e. speed is too high) but I can not say without a test scene.

[B.Y.O.B.]

Could it be that the problem is related to bump mapping? In Dade's pool, the mesh surface is displaced, while lacilaci's water surface seems to make use of bump mapping.

[Dade]

Oh, fu@k, I had an uninitialized variable in Metropolis code used by PSR (i.e. it could start with random values). This explain the very odd "0.99999997 Vs 0.99999998" problem, now smaller values works fine too (i.e. for faster sharpening).