Vray denoiser - squares and blotches are ruining my render.

[JD3D_CGI]

I've set the reflection to all channels - not that I think it matters for a still - but we really need to have a talk about these godawful blotches.

Admittedly they are reflections of light in a glossy chrome material with depth of field which in itself is amazing - but does the denoiser have to resolve a bright pixel into a honking great square that surrounds where that bright pixel would be? is there no better way?

And when they are denoised temporally, they dance around - it's not good!

[JD3D_CGI]

It feels like even the AI denoiser is handling it better. I may as well do a kamikaze animation render with that and see how it comes out.


edit: no, AI denoiser crashes on file output

[psanitra]

Well, been my experience as well trough the years. As soon as one start something with mid or high intesity/lights range, whole denoiser becomes pretty much useless because this squares behavior. Many times on forum, as proposed solution by devs was to clamp max rays or color. Uhm, dull renders, no thank you. I want my rich highlighs and bokehs and high specular values. Some users just got tired of asking and waiting for fix and are stamping the it photoshop. .

Because the nature of sampling,this firelfy behavior is more likely to appear on progressive or on gpu progressive. That`s why i stick to bucket for final renders and was looking forward for GPU bucket option. Render from vray for Houdini testing attached. Simple noise grid with reflection, toy, one light and dozens of squared pixels.

I hope dev can find and improved approach to locate individual biright pixels and threat them differently in denoising process. Vray denoise really is amazing, but few pixels can make a difference between using it and not using it at all.

[JD3D_CGI]

Damn, I didn't realise it was such a bad problem.

Yeah, I would love the devs to fix this - ASAP really as well...

Sadly this is part of an animation, so post effect blurring looks terrible too.

edit: PS, I love your hdris and your website dude - I use them all the time!

[Paul Oblomov]

It feels like gi caustics. Maybe the solution is to disable it OR maybe clamp intensity to smth like 4 I remember reported these blocks years ago and now they emerged again Nice to see good old fellas...

[JD3D_CGI]

No GI caustics... so you mean I should clamp the intensity in the colour mapping box?

[psanitra]

No GI caustic for me either, ray/pixel/sample intensity is all that matters. Of course you can avoid the issues just clamping the secondary, or clamp color. But that`s for another discussion about whole workflow.

My issue with denoiser is, that it`s not being able to deal with these types of scenarios. And i would love to see him to be able to. For example, there are lot of scenes where 99.99% pixel in the image could be rendered super fast and be clean with denoiser, in lets say under 1-2 minutes. But becasue of couple of bright pixels, we end up not using denoiser it at all, or rendering 10x longer.

And when inspecting the intensities on the pixels, it`s not like we have some crazy floats. You can have 0.3 on float, and 0.8 next to it, and denoiser will fail, and extrapolate the problem from 1 pixel to 5x5 or whatever the pattern/curve is you use to blend in. All i`m asking is to revise the approach of this specific problem, instead of proposing to clients to clamp their renders.

[JD3D_CGI]

Sorry mate, could to explain what you mean by "clamping the secondary"

[JD3D_CGI]

Bizarrely silent from Chaos on this!

[psanitra]

Sorry mate, could to explain what you mean by "clamping the secondary"

There is a setting called Max Ray Intensity which at default is set to 20, try something lower like 4 or even 2. That should help you. For more info check the docs for max https://docs.chaosgroup.com/display/...lobal+Switches

[vjaramillo]

I posted about this last month.

[JD3D_CGI]

I posted about this last month.

Yeah, I replied to your post and tried to bump it up - neither post has gotten a reply from Chaos yet... I sent a tech support request with my scene to them, so hopefully they might be able to shed light on this.

[matanov]

We are planning a fix for V-Ray Next: Update 2.

[JD3D_CGI]

We are planning a fix for V-Ray Next: Update 2.

That is amazing! I am so grateful... although, the timing kind of dictates whether I get this project done on time?

Is this being trailed at all in any of the nighties yet?

[vlado]

I'm tinkering with a de-speckle pass before denoising, but it's not quite where I want it to be yet. It's something that you can try on your own though - if you can take an OpenEXR file and de-speckle it in PhotoShop or something and then pass it back through the V-Ray denoiser, it might work better.

Best regards,
Vlado