V-Ray denoiser plugin for Nuke WIP

[sharktacos]

wait... is the Vray denoiser plugin for Nuke available now? Link please.

[flat]

The free version doesnt support plugins, so if there was the vray denoiser available you'd have to start paying for fusion.

Exactly like that.

Expect also in the OpenFX version.
It can also be used in Fusion Studio and Natron.

[flat]

wait... is the Vray denoiser plugin for Nuke available now? Link please.

It is here.
http://forums.chaosgroup.com/showthr...722#post720722

[stilgarna]

Can you tell me which version of the denoiser plugin you are using? Or the date of the DLL?

Best regards,
Vlado

November 9 2016

@stilgarna
I managed to reproduce the problem and it seems to happen when frame_blend is set (not 0) and rendering without GUI.
We'll see to cleaning up the warnings and fixing the issue.

Could you confirm if your setup works when frame_blend is 0?

Thanks!

Yes it works with frame_blend set to 0 !!!
Thanks !

This denoiser is just.... no word... game changer ! I hope I can post some test scene soon

In CPU mode : 3min20s - 4min00 (i7 970)
In GPU mode : 1min30s (very old GTX 470)
Crazy !

Very good job guys !!!

[sharktacos]

The denoiser apparently uses a custom render element called "noise level."
Looking at a multi-channel EXR in Nuke, I do not see this render element.
I do see the all others (world normals, etc) and the Nuke denoiser appears to be working fine.
I was just curious as to why I was not able to see the "noise level" render element in the EXR.

[vlado]

The denoiser apparently uses a custom render element called "noise level."
Looking at a multi-channel EXR in Nuke, I do not see this render element.
I do see the all others (world normals, etc) and the Nuke denoiser appears to be working fine.
I was just curious as to why I was not able to see the "noise level" render element in the EXR.

It's a single channel floating-point element, I think Nuke groups these channels in a slightly different location.

Best regards,
Vlado

[sharktacos]

Part of the reason I was asking is I recall you saying that in theory the Denoiser could be used for other renderers. It would be easy enough to create the other render elements it needs (world normals, world position, etc), but the "noise level" element would of course not be something other renders have. So I'm curious whether the Denoiser could work without it, or, if not, how you imagined that functioning?

[vlado]

Part of the reason I was asking is I recall you saying that in theory the Denoiser could be used for other renderers. It would be easy enough to create the other render elements it needs (world normals, world position, etc), but the "noise level" element would of course not be something other renders have. So I'm curious whether the Denoiser could work without it, or, if not, how you imagined that functioning?

The denoiser can work without this element. In this case, it assumes that the noise level is uniform across the image and you must specify that specific noise level in the denoiser settings manually through the "threshold" parameter. It seems to work fine generally.

Best regards,
Vlado

[georgi.dunchev]

It's a single channel floating-point element, I think Nuke groups these channels in a slightly different location.

Best regards,
Vlado

In Nuke, single channels are usually grouped in the "other" layer. So the noiseLevel is actually other.noiseLevel and the denoiser looks for it there.

When no noiseLevel is supplied a threshold control is unlocked in the denoiser so you can have some control.

[sharktacos]

In Nuke, single channels are usually grouped in the "other" layer. So the noiseLevel is actually other.noiseLevel and the denoiser looks for it there.

When no noiseLevel is supplied a threshold control is unlocked in the denoiser so you can have some control.

Thanks, I see it there.

[sharktacos]

Wanted to confirm that this is the correct workflow for denoising separate render elements:
1. shuffle the render element out from the multi-exr
2. denoise it
3. repeat for all other render elements
4. comp the render elements back together

This works fine, but all the denoiser nodes are of course pretty heavy in Nuke. Just wondering if there was a more efficient way of doing this?

[georgi.dunchev]

Instead of shuffling the render element, you can switch the channel assignment to manual and feed the desired layer to the RGB. A nice touch is to output the layer name set to rgb in the label of the node, so you can see how each denoiser is set in the DAG.

You can denoise some of the channels, and then use them inside a final VRayDenoiser instead of denoising every channel and then compositing the beauty. But you have to figure out what works best for the image.

However, for the moment you can't skip running multiple denoisers if you want to denoise multiple render elements.

[stilgarna]

@stilgarna
I managed to reproduce the problem and it seems to happen when frame_blend is set (not 0) and rendering without GUI.
We'll see to cleaning up the warnings and fixing the issue.

Could you confirm if your setup works when frame_blend is 0?

Thanks!

Hello

Do you think there will be (soon or not) an updated version of the Nuke plug' in ?

Thanks you !

Best regards.

[sharktacos]

Vlado and Chaos folks,

The Denoiser .dll does not work in the latest version of Nuke (Nuke10.5v1). It was working fine in Nuke10.0v3 but now that we upgraded, it does not load (i.e. the menu item appears, but you get an error message saying it cannot find the .dll). Would it be possible to get a new .dll of the denoiser that works with Nuke10.5v1 or can you suggest a fix?

thanks!

[pechart]

Vlado and Chaos folks,
Would it be possible to get a new .dll of the denoiser that works with Nuke10.5v1 or can you suggest a fix?
thanks!

I asked that in a seperate thread and they said they will come up with something.

Another question:
Are you using multichannel EXRs? We usually work with single channel EXRs, but in our vray build (from Jan, 30th) only multichannel EXRs are supported. I'd like to know, if this will change at some point or is already supported, maybe. Thanks!