LuxCoreRender Forums

B.Y.O.B. wrote: ↑Sun Jun 09, 2019 9:51 pm About the weird stripe: I can reproduce it even with just the volume cube and the sun. Maybe it's a bug, or maybe some other problem in the scene that I couldn't find.

I should have fixed the problem.

Looks like there is still a good reason to keep BCD on board of LuxCorerRender.
Scenes with scattered light, homogeneous volume or heterogeneous volume, lead on openCL to a strange distributed noise pattern even. These patterns do not disappear with 2500 samples. OIDN is currently not able to denoise these scattered patterns as shown in last post. OIDN creates organic very disturbing looking patterns.
But there are two solutions. One is to switch to BiDir, Metropolis,. Even with a noisy image OIDN makes a good result with the BiDir render.
Another one is to render openCL, what is much faster to achieve the same sample count, and use 'good old' BCD. Just switch denoiser type to BCD with default values and you get following results. Okay, BCD is slower. BCD plus openCL is still much faster than BiDir plus OIDN.

But isn't this result still wrong?
Look at the corners of the room - they are completely blurred.
Judging from the noisy images it looks to me like they should be sharp.

Try to use a texture on the back wall and see what the denoisers do to that.

B.Y.O.B. wrote: ↑Mon Jun 10, 2019 10:37 am Try to use a texture on the back wall and see what the denoisers do to that.

EDIT
Instead of a texture I placed a spot light for the projection of the test picture. The comparison of BCD and OIDN. BCD keeps details and flatness. OIDN creates wrinkled paper like faces and looses details.

lord Dade has blessed us once again!

Okay, the image is beside of the beam very noisy. BCD detects the correct patterns in the test picture. However, the result is a bit blurry but then with this amount of noise the result is well acceptable. Umpf. OIDN's algorithms got hick up by eating the openCL patterns of scattering volume. Fancy, but not very accurate.

EDIT
And this is the last image of the little investigation. I cannot see any wrong behavior of LuxCoreRender. There is obviously everything okay from material volume to camera volume including indirect cache.

I LOVE YOUR TESTS

madcat117 wrote: ↑Mon Jun 10, 2019 3:07 pm I LOVE YOUR TESTS

Thank you a lot. You are welcome.

Still appears to be something present...
RAW render CONVERGENCE mix with SAMPLECOUNT Scene

kintuX wrote: ↑Wed Jun 12, 2019 9:12 am CONVERGENCE mix with SAMPLECOUNT

Very nice drawing

Really need to play with those AOVs more...