In playing around with a recent build of 2.5, I was happy to discover that there's now a caustics pass option (I had been extracting it myself for separate denoising by subtracting all other passes from the main rendered image). Unfortunately, this has revealed some problems with 2.5...
Firstly, for some inexplicable reason, the caustics pass is barely affected by the Denoise node. Even after doing a manual extraction as before, there is hardly any effect. All other passes seem to be affected as normal. I switched back over to 2.4 and everything works as expected.
Second, there seems to be a lingering problem with the Indirect Specular pass. I've got some orange glass, and in the Indirect Spec pass it's got waaay too much blue going on for some reason. This problem also occurs in 2.4.
Not sure if either of these problems are known yet, but I figured I'd say something here in case I'm just being an idiot...lol (Edit: I'm using Blender 2.91)
I'm including a test file:
Indirect Specular pass problem
Forum rules
Please upload a testscene that allows developers to reproduce the problem, and attach some images.
Please upload a testscene that allows developers to reproduce the problem, and attach some images.
-
- Posts: 11
- Joined: Sat Apr 04, 2020 5:21 pm
Re: Indirect Specular pass problem
Has there been any progress with these issues?
Re: Indirect Specular pass problem
You can denoise the caustics pass by turning off the Pixel Filter (setting it to None). Find it under Sampling > Advanced > Filter. It's probably set to Blackman-Harris at 1.50 pixels by default. That does not work well with noise from the Metropolis sampler
AMD Ryzen Threadripper PRO 5995WX 64-Cores | 2x Gigabyte RTX 4090 Gaming OC
ASUS Pro WS WRX80E-SAGE SE WIFI | 256GB Kingston Server Premier ECC Unbuffered DDR4
ASUS Pro WS WRX80E-SAGE SE WIFI | 256GB Kingston Server Premier ECC Unbuffered DDR4
-
- Posts: 11
- Joined: Sat Apr 04, 2020 5:21 pm
Re: Indirect Specular pass problem
Ahhh, very interesting! Switching the filter to "None" did the trick, and I can now use the Denoise node in the compositor on the caustics pass...thanks so much!! Still very odd though, that it worked just fine in version 2.4 and not in 2.5...is it supposed to work as it is now, or is that a problem?
I don't suppose you have any insight into why the Specular Indirect pass has much more blue than it should...?
I don't suppose you have any insight into why the Specular Indirect pass has much more blue than it should...?
Re: Indirect Specular pass problem
This: https://github.com/LuxCoreRender/LuxCore/issues/413steveernst117 wrote: ↑Thu Feb 04, 2021 1:15 am Still very odd though, that it worked just fine in version 2.4 and not in 2.5...is it supposed to work as it is now, or is that a problem?
Re: Indirect Specular pass problem
Maybe pixel filtering should now be disabled in BlendLuxCore by default?
Re: Indirect Specular pass problem
Any advantage of having it enable ?
Re: Indirect Specular pass problem
This can already be done by setting the pixel filter to "None".
But yes, could be made more standard with a checkbox "use filtering" and then all filter options aside from "None" in the dropdown.
Re: Indirect Specular pass problem
I was assuming it was already if OIDN was used