Adaptive sampling never resolves noise in some scenes

[gktb56x]

Greetings!

After upgrading to MtoA 3.3.0 I have noticed that progressive rendering and adaptive sampling behave differently, causing issues for me in some scenes, mostly involving sss material, and I have encountered this issue in all later versions.

This is my attempt to illustrate the issue: test_scene.zip

You may notice that with MtoA 3.2.2 (and earlier) standard render, progressive, and adaptive look very close to each other (as they should), but with MtoA 3.3.0 (and later) there is some extra noise on progressive and adaptive. This difference may seem insignificant, but it gets bigger in complex scenes, making adaptive sampling unusable, since noise never resolves, no matter how many max camera samples are (although render times increase proportionally).

The only thing that solves this issue for me is rolling back to MtoA 3.2.2, so I would like to know is there some other ways to solve it.

(I am using Maya 2019)

Thanks!

[lee_griggs]

Thank you for sharing your results. How many Camera (AA) samples are you using?

Have you tried it with more complex scenes? ie GPU, interiors, motion blur, etc? From the docs:

"Adaptive sampling is good for situations where small areas of the image have high levels of noise that require an impractical number of samples to clear up. For example scenes with bright, motion-blurred speculars, DOF, buzzing rim lights, or scenes with the hair shader. "

[gktb56x]

Yes, in many scenes adaptive sampling allows significantly reduce render times, it's working great for me in MtoA 3.2.2 and earlier.

This test scene is bad use case for adaptive sampling, it was only intended to demonstrate extra noise that appears in the newer versions of MtoA in some scenarios (involving adaptive sampling or progressive render). In this case it doesn't matter how many max AA samples (100, 1000) and how low noise threshold is, noise never resolves (although render times increase).

[thiago.ize]

The just released Arnold 6.1 (https://www.arnoldrenderer.com/arnold/download/) contains an adaptive and progressive sampling fix for this problem.

[T0M0X]

Little bit offtopic, but thank you very much for implementing nested dielectrics in Arnold.

Thanks !