animation, random frames with faulty noise

[jrgby]

Hi guys,
I'm rendering an animation and am coming up with this bug fairly consistantly.

Most frames are fine but every now and then a random frame will not denoise properly, or rather it hasn't rendered to the same noise limit.
You can notice the diference in quality in the .cxr before denoise.
See attached images of consecutive frames, both after full denoising at 1.0

Here's what I'm doing:

max 2018 with corona 1.7.4
Noise level limit set to 4.0
Set to gather data for later
Frames auto save as .cxr
Rendering at 720x1280 frames are taking under 1 minute
Then running the batch script to denoise at 1.0, set post settings + save as .PNG

If I re-run the render for the problem frame it will fix it, but this can take a while, for example a previous test animation had 10 frames in 100 at random points that needed rendering again.

This feels like a bug but maybe I am missing something?

[TomG]

This is a bug, one that we are aware of - sometimes, the calculated noise limit is artificially low for one calculation (usually, every 5 passes at the default for Adaptivity), and this artificial drop may take it below the Noise Limit and so the render ends, even though the noise is actually higher than what was calculated in that round of Adaptivity. This one has some test results on Mantis (https://corona-renderer.com/bugs/view.php?id=2811)

As yet, the cause / solution is still being investigated. One solution at the moment is to render to set time or passes if doing an animated sequence.

[jrgby]

Thanks Tom, glad to know it's not something random with my system.
Will go with setting time/pass limit in future.

[maru]

(internal note: id=231941804)

[maru]

According to our tests there is an improvement in Corona 3.1 and 4 (newest daily) compared to Corona 1.7 and 2. The pass number and rendering time differences between different frames are minimal. It would be great if others could test Corona 3 with noise limit as well and let us know if everything is fine now or not.