OpenImageDenoise

Discussion related to the Engine functionality, implementations and API.
User avatar
rickyx
Posts: 14
Joined: Sun Apr 29, 2018 5:23 pm

Re: OpenImageDenoise

Post by rickyx » Mon Feb 11, 2019 1:11 am

I have downloaded the latest build: how to activate OIDN?

Thank you,
Rickyx

User avatar
B.Y.O.B.
Developer
Posts: 2111
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: OpenImageDenoise

Post by B.Y.O.B. » Mon Feb 11, 2019 8:50 am

In the render properties, find the panel named "LuxCore Denoiser" and enable it. Since OIDN is the new default denoiser, this is all.
(I assume that you have updated BlendLuxCore as well, as described in the daily builds thread).
Support LuxCoreRender project with salts and bounties

User avatar
rickyx
Posts: 14
Joined: Sun Apr 29, 2018 5:23 pm

Re: OpenImageDenoise

Post by rickyx » Mon Feb 11, 2019 9:31 pm

Thank you, results are really amazing.

Just another question: the Intel OpenImageDenoise is optimized for Intel processors.
Will it work also on other processors (ex. Amd) or it is like Cuda for Gpu?

User avatar
lacilaci
Donor
Posts: 782
Joined: Fri May 04, 2018 5:16 am

Re: OpenImageDenoise

Post by lacilaci » Mon Feb 11, 2019 9:39 pm

rickyx wrote:
Mon Feb 11, 2019 9:31 pm
Thank you, results are really amazing.

Just another question: the Intel OpenImageDenoise is optimized for Intel processors.
Will it work also on other processors (ex. Amd) or it is like Cuda for Gpu?
unless your cpu is like, from 2004 or so... it should work I guess. (check if your cpu supports sse4.1)

User avatar
epilectrolytics
Donor
Posts: 218
Joined: Thu Oct 04, 2018 6:06 am

Re: OpenImageDenoise

Post by epilectrolytics » Mon Feb 11, 2019 9:50 pm

rickyx wrote:
Mon Feb 11, 2019 9:31 pm
Just another question: the Intel OpenImageDenoise is optimized for Intel processors.
Will it work also on other processors (ex. Amd) or it is like Cuda for Gpu?
Works perfectly on my Ryzen 7, like 2s denoising time for full HD.
On some very recent Intel CPUs it would work even faster, less than a second, because of optimisation.
Last edited by epilectrolytics on Mon Feb 11, 2019 9:57 pm, edited 1 time in total.
MBPro 15" 16GB i7-4850HQ GT750M, MacOS 10.13.6 _________________Win10Pro PC Ryzen 2700X, RTX 2070, GTX 1060

User avatar
rickyx
Posts: 14
Joined: Sun Apr 29, 2018 5:23 pm

Re: OpenImageDenoise

Post by rickyx » Mon Feb 11, 2019 9:53 pm

Ok, got it ;)
I meant... I didn't know if it was related to specific intel hardware.
I'm lucky! I have an i7 ;)

User avatar
Dade
Developer
Posts: 1913
Joined: Mon Dec 04, 2017 8:36 pm

Re: OpenImageDenoise

Post by Dade » Mon Feb 11, 2019 10:36 pm

rickyx wrote:
Mon Feb 11, 2019 9:53 pm
Ok, got it ;)
I meant... I didn't know if it was related to specific intel hardware.
I'm lucky! I have an i7 ;)
Usually, Intel Libraries like Embree and Oidn have different code paths optimized for various level of SSE/AVX/AVX512 support. Indeed they will run very well on top Intel hardware (i.e . AVX512) but they will work well on modern AMD CPUs too.
Support LuxCoreRender project with salts and bounties

voxelium
Posts: 3
Joined: Tue Feb 05, 2019 10:50 pm

Re: OpenImageDenoise

Post by voxelium » Mon Feb 11, 2019 11:04 pm

We've updated the OIDN documentation with details regarding the albedo and normal images:
https://openimagedenoise.github.io/documentation.html

Yes, OIDN should work with any CPU supporting SSE4.1.

User avatar
B.Y.O.B.
Developer
Posts: 2111
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: OpenImageDenoise

Post by B.Y.O.B. » Mon Feb 11, 2019 11:25 pm

Thanks for the detailed documentation update!
https://openimagedenoise.github.io/documentation.html wrote:The albedo for dielectric surfaces (e.g. glass) should be either 1 or, if the surface is perfect specular (i.e. has a delta BSDF), the Fresnel blend of the reflected and transmitted albedos (as previously discussed). The latter usually works better but only if it does not introduce too much additional noise due to random sampling. Thus we recommend to split the path into a reflected and a transmitted path at the first hit, and perhaps fall back to an albedo of 1 for subsequent dielectric hits, to avoid noise.
@Dade: Could this be implemented in LuxCore? I found that the noise from randomly sampled reflections/refractions on glass really throws the denoiser off, sometimes not denoising the affected area at all. It would be much better if the albedo was completely noise-free after the first sample.
Support LuxCoreRender project with salts and bounties

User avatar
epilectrolytics
Donor
Posts: 218
Joined: Thu Oct 04, 2018 6:06 am

Re: OpenImageDenoise

Post by epilectrolytics » Tue Feb 12, 2019 7:37 am

voxelium wrote:
Mon Feb 11, 2019 11:04 pm
We've updated the OIDN documentation with details regarding the albedo and normal images
Thanks!

@Dade:
We have already bounced albedo integrated, the new docs propose bounced normal (for delta) as well!
If noise is accumulating after several bounces, setting max path depth = 3 or 4 for AOV could be considered.

The "average normal AOV" is still noisy in certain cases (like very small bump detail) and needs better antialiasing there.
In my comparison with Cycles denoiser OIDN is far superior in every respect except bump/normal reconstruction.
MBPro 15" 16GB i7-4850HQ GT750M, MacOS 10.13.6 _________________Win10Pro PC Ryzen 2700X, RTX 2070, GTX 1060

Post Reply