Posts: 36
Threads: 5
Joined: Jan 2018
(26.10.2019, 13:02)Selur Wrote: Quote:My input file is an 8 bit x264 YUV, not RGB. I'm encoding it to 10 bit x265, is that what you meant by high bit depth?
The important question is what color space and bit depth the content is in the filter chain when fed to the filter.
Cu Selur
Oh how do I determine that, what the color space and bit depth are?
I just tried the KNLmeansCL using the last build you sent me - while it doesn't crash anymore, I don't think any denoising is being performed, when I set the channels to UV.
Also - similarity radius can't be increased beyond 4 in Hybrid, while the developer of KNLmeansCl had stated that its maximum value had been increased to 8 some time back.
Posts: 10.548
Threads: 57
Joined: May 2017
26.10.2019, 17:02
(This post was last modified: 26.10.2019, 17:14 by Selur.)
Quote:Oh how do I determine that, what the color space and bit depth are?
Look at the script, in case no color conversion is done the it's still the same as the input otherwise it's determined by the last color conversion.
Quote:I don't think any denoising is being performed, when I set the channels to UV.
Then your noise it probably is only in the Y channel (noise only being in one of the YUV channels isn't uncommon).
Quote:while the developer of KNLmeansCl had stated that its maximum value had been increased to 8 some time back.
Would have been helpful if the author would have also noted that somewhere in the documentation and not only in the changelog,...
Atm. w-radius is limited to 16 and s-radius is limited to 4 in Hybrid.
-> will adjust s-radius limit to 8 for the next release.
Wondering whether the high bit depth support not only works in Avisynth+ but also in Vapoursynth.
(only see it mentioned in the changelog for Avisynth+ => nope, feeding it yuv420p10 throws 'vapoursynth.Error: knlm.KNLMeansCL: only YUV444P10 and RGB30 are supported!')
Cu Selur
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
Posts: 10.548
Threads: 57
Joined: May 2017
looking at
https://forum.doom9.org/showthread.php?p...ost1888701 it seems like the differences with other wmodes then 0 are just really subtle which is why they are difficult to spot, but they are there.
Cu Selur
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.