3.55 denoiser test with phoenixfd too

[jstrob]

I did this test with the denoiser. I can't get the denoiser to work with separate elements. It works only when choosing replace RGB with denoise result and single pass denoise.
When I try "store denoise result in separate element and denoise render element separately I get the first render: there is denoising at all in the render elements and the phoenixfd smoke becomes very bright.
The second one is when choosing "replace RGB with denoise result"and "single pass denoise".

[jstrob]

Other problem the denoiser does not work when I submit to my farm (RPM+deadline).

[vlado]

For the brightness issue, will try to reproduce and fix it, for RPM I'm not really sure - if it look at the V-Ray log, does it say anything useful? How are you saving out the render output - through the 3ds Max VFB, through the V-Ray raw image output section, or from the "Split render channels" option?

Best regards,
Vlado

[jstrob]

V-Ray log, you mena in C:\VRayLog.txt ? I just see info from october in it.

I save via the Render Output. I don't know if RPM does something else under the hood.

Here are some screenshots of my settings:

[svetlozar.draganov]

We managed to reproduce the issue whit the brighter smoke, this is already added into the tracker.
Regarding the log file, the default log file location is in windows temp directory, you can access it easily by typing %temp% in the address bar of windows explorer.

Regarding the non denoised result with RPM and Deadline would you please try to disable first RPM and then Deadline just to narrow down which application is involved.
You may also try to save the output from V-Ray Raw Image File or from Separate Render Channels.

[jstrob]

First thing to note is that even LOCALLY right in the VFB the denoiser does not do anything when I have these settings and outputing all V-Ray passes (I tried by saving via render output and V-Ray raw image output section with the "Split render channels" with same results):

1- Replace RGB element with denoised result/Denoise render elements separately (no denoising on any element and even the RGB pass).

2- Store denoised results in separate elements /Denoise render elements separatel (no denoising on any element and even the RGB pass).

The only way to see a denoised result locally is to choose these:
3-Replace RGB element with denoised result/Single pass denoise on RGB only.

When submitting on my farm, like I said, even 3 does not work via RPM/Deadline. But it works via Backburner with V-Ray raw image output and max render output. When I submit via RPM, it uses the max VFB so I guess that's why it does not denoise. I could make it work by unckecking everything in the RPM's deadline section in the pathing tab and checking separate render elements and manually assign the same path and filename as in rpm in there. But it involves many more steps for something that is normally taking car of by RPM. Also I end up with all files in the same folder instead of separate folders for each sequences. But I guess I have to contact Grant and thinkbox about this.

So for network rendering it's a RPM issue, but what about denoising the elements? Can you do it on your side?

[svetlozar.draganov]

I have tried the described setups but I didn't get any of the issues with the denoiser you are facing. Can you specify the exact V-Ray build that you are using?

Here what I've done.

1.I have added all render elements
2.I have used all possible ways to save the file on HDD, render output, raw image, separate render channels
3.This is the result of the RGB channel before and after denoising with your first approach:

4.Same comparison but this time between RGB and Denoise element (you second approach):


Would it be possible to send us also the scene you are working on, there might be something specific about it.

Looking forward to hearing from you.

[jstrob]

Hi Svetlozar,

1-I'm testing V-Ray 3.55.01. I reran my test again and now I see some very subtle denoising when I choose the strong settings (with separate elements). So the denoising is far stronger when I choose denoise RGB only. Is this normal?

2-Also I thought that with the denoise separate element and store denoise result in separate element, we could see the element being denoised. so if I look at the specular pass for example I could see it being noise free. But I guess it's not implemented yet?

here is my scene (with a phoenixFd fire to simulate and render frame 35 please):

[jstrob]

About Q # 2, I see in the beta 3.5 doc that denoising the render element is not implemented yet. So sorry about that I should have read again before asking.

About #1 I still wonder why I get very different result for both method. Maybe it could be related to my setting in the color mapping section. I saw this note in the doc:
"This mode can only work correctly if no color mapping is applied to the render, so that the mentioned render elements add up correctly to the RGB image."
I use use sub-pixel mapping and clamp output? I'll try to see if this can be this.