New OIDN 1.4 with new features and better detail preservation

Discussion related to the LuxCore functionality, implementations and API.
juangea
Donor
Donor
Posts: 332
Joined: Thu Jan 02, 2020 6:23 pm

Re: New OIDN 1.4 with new features and better detail preservation

Post by juangea »

Have you implemented also the option to assume that the Albedo is 100% clean? The one that does not prefilter the Albedo and uses it directly as fully clean
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: New OIDN 1.4 with new features and better detail preservation

Post by B.Y.O.B. »

To make the test more informative, it would be good to add a ground truth (high sample count without denoising), then we could compare which denoised result resembles the ground truth best.
juangea
Donor
Donor
Posts: 332
Joined: Thu Jan 02, 2020 6:23 pm

Re: New OIDN 1.4 with new features and better detail preservation

Post by juangea »

But in this case, that scene is not a noise heavy scene with a lot of Albedos, there are more transmissive and specular materials, and small dielectric with small albedos.

A scene with cloth, small fine cloth detail in a sofa cloth or a wooden floor with a lot of wood detail, that's the type of scene what will benefit from this, we are working in a project where we can test this, in fact I have a 360 visit where the effects of loosing detail are very clear, and with this, that could be improved a lot :)
CodeHD
Donor
Donor
Posts: 437
Joined: Tue Dec 11, 2018 12:38 pm
Location: Germany

Re: New OIDN 1.4 with new features and better detail preservation

Post by CodeHD »

I would also rate is as an improvement in most areas. On the plate, it looks like sometimes there can be inverse behavior between some regions.
Since the undenoised result is always available, I guess it should be possible to recalculate after rendering is finished. Thats probably on the BlendLuxCore side of implementations though?
User avatar
TAO
Developer
Developer
Posts: 851
Joined: Sun Mar 24, 2019 4:49 pm
Location: France
Contact:

Re: New OIDN 1.4 with new features and better detail preservation

Post by TAO »

I think Dade already implement OIDN 1.4 in LuxCore 2.6.
juangea
Donor
Donor
Posts: 332
Joined: Thu Jan 02, 2020 6:23 pm

Re: New OIDN 1.4 with new features and better detail preservation

Post by juangea »

Yep, I was asking about BlendLuxCore, the Blender Addon, it needs some touches for sure to be able to enable the pre-filter or the clean Albedo options :)
User avatar
Dade
Developer
Developer
Posts: 5672
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: New OIDN 1.4 with new features and better detail preservation

Post by Dade »

juangea wrote: Tue May 25, 2021 5:58 pm Yep, I was asking about BlendLuxCore, the Blender Addon, it needs some touches for sure to be able to enable the pre-filter or the clean Albedo options :)
Albedo pre-filtering is enabled by default and done on the fly by the new image pipeline OIDN plugin code (so I think there is no need to affect or of a duplicate Albedo AOV). It should work all out of the box.

I assume there is no need to not pre-filter Albedo (aka pre-filtering an already clear Albedo pass should have no effects or harm) :?:
Support LuxCoreRender project with salts and bounties
juangea
Donor
Donor
Posts: 332
Joined: Thu Jan 02, 2020 6:23 pm

Re: New OIDN 1.4 with new features and better detail preservation

Post by juangea »

No, there should not be harm, but there could be some benefit of using the non-filtered Albedo as a filtered one, specially because there may be parts of the Albedo that could be destroyed by the filter, or because a bit of noise in the Albedo could generate an interesting detail retention (grain-like) effect.
I’m not sure because I have not tested it, but maybe.

I’ll update BlendLuxCore later and test it in our current scene :)
juangea
Donor
Donor
Posts: 332
Joined: Thu Jan 02, 2020 6:23 pm

Re: New OIDN 1.4 with new features and better detail preservation

Post by juangea »

I did a quick test but weirdly enough I see no difference in our scene.

In fact I compared the Denoised pass result in 2.92 with 2.6 alpha vs the denoised result of the Denoise node and the result was exactly the same, I may have did something wrong doig the update, I'll check again and say something.
User avatar
Dez!
Posts: 368
Joined: Sun Apr 08, 2018 1:09 am
Location: Ekaterinburg
Contact:

Re: New OIDN 1.4 with new features and better detail preservation

Post by Dez! »

Image
32 sempl
Linux Plasma | Ryzen 5, 32Gb, SSD M2, GT 590 RX | BenQ 27 | Wacom One | Microsoft Ergo | Tie Guan Yin tea
http://dezigner.tilda.ws/
Post Reply