Direct 8-bit Conversion Using QTGMC

[rgr]

This just throws up more general, not QTGMC related questions. *gig*
Why the AddGrainC?
Why ConvertToYUV444? Why for 'a' before and after QTGMC, but for 'b' you only convert to 444 after applying histogram.

AddGrain for a bigger challenge, not a simple test.
YUV444 for the same reason as the larger depth -- for greater accuracy.
The second YUV444 is actually unnecessary.

Because you are using a script that only uses a fraction of QTGMC.

Okay, so your statement is only true for specific settings? Maybe it's just one filter that benefits from 16 bit (or the YUV444 conversion). Do you know which parameter in QTGMC is causing the differences.

If it can be clearly shown, that QTGMC benefits from using 16bit and/or 444, maybe an option for this could be added to QTGMC.

Why? QTGMC works with the parameters you give it. It's your job to give them the best.

[Selur]

Settings always should be adjusted to the source.

Do you know which parameter in QTGMC is causing the differences.

If it can be clearly shown, that QTGMC benefits from using 16bit and/or 444, maybe an option for this could be added to QTGMC.

?

[rgr]

Settings always should be adjusted to the source.

Well, when dividing odd numbers (by 2, because you'll complain) I prefer not to.

[Selur]

Okay, so you have no clue, what happens. Thanks for clearing that up.

[rgr]

You too, you didn't even know how to get to it.
The important thing is that it works.

[_Al_]

I took real life footage, no colorbars etc., does not resize changes borders in them etc., of course, with higher chroma resolution, it will look better,
anyway,
I picked up a NTSC DV.avi footage which is interesting, has lines, edges, zigzaging, uniform areas prone to banding, and it is a pan.
Used this script using four formats to generate four clips with different format and QTGMC was applied:

import vapoursynth as vs
from vapoursynth import core
from havsfunc import QTGMC

# ffms2 Source loads it as YUV411P8
clip = core.ffms2.Source(r"D:\delorean_NTSC_DV.avi") 
clip = clip.std.SetFrameProps(_Matrix=6, _Transfer=6, _Primaries=6)

# there will be as many clips set for output as there are formats here:
formats = [vs.YUV420P8, vs.YUV422P10, vs.YUV444P10, vs.YUV444P16]
clips = []
for index, format in enumerate(formats, start=1):
    c = core.resize.Bicubic(clip, format=format)
    c = QTGMC(c, Preset='Slower', TFF=False)
    c = core.resize.Bicubic(c, format=vs.YUV420P8)
    clips.append(c)
    c.set_output(index)

if __name__ == '__main__':
    import view
    view.Preview(clips)

At the end all outputs are converted to YUV420P8 because we have to compare that delivery format.
I encoded four clips with proper naming, using crf0 (it used qp 0), so I think all should be lossless:
https://files.videohelp.com/u/198160/clip1_420P8.mp4
https://files.videohelp.com/u/198160/clip2_422P10.mp4
https://files.videohelp.com/u/198160/clip3_444P10.mp4
https://files.videohelp.com/u/198160/clip4_444P16.mp4
each clip has about 50-60MB
original NTSC DV (27 MB):
https://files.videohelp.com/u/198160/delorean_NTSC_DV.avi

[_Al_]

btw. I might include windows batch script, to post how easy to encode multiple outputs from a vapoursynth script:

@echo off
set "script=E:\vapoursynth_script.py"
set "vspipe=C:\Program Files\VapourSynth\core\vspipe.exe"
set "ffmpeg=E:\tools\ffmpeg.exe"
set "encoded_directory=E:\encodes"

set "index=0"
for %%f in (420P8, 422P10, 444P10, 444P16) do call :encode %%f
pause
exit

:encode <a string>
set /a index+=1
"%vspipe%" "%script%" --outputindex %index% --container y4m - | "%ffmpeg%" -y -f yuv4mpegpipe -i - -c:v libx264 -crf 0 "%encoded_directory%\clip%index%_%1.mp4"
goto :eof

[Selur]

@_AI_: Thanks
On a first glance, using:

clip420P8 = core.lsmas.LibavSMASHSource(source="C:/Users/Selur/Desktop/compare/clip1_420P8.mp4")
clip422P10 = core.lsmas.LibavSMASHSource(source="C:/Users/Selur/Desktop/compare/clip2_422P10.mp4")
clip444P10 = core.lsmas.LibavSMASHSource(source="C:/Users/Selur/Desktop/compare/clip3_444P10.mp4")
clip444P16 = core.lsmas.LibavSMASHSource(source="C:/Users/Selur/Desktop/compare/clip4_444P16.mp4")

clip = core.std.Interleave([clip420P8.text.Text("420P8"), clip422P10.text.Text("422P10"), clip444P10.text.Text("444P10"), clip444P16.text.Text("444P16")])

Sadly, I can't really see a difference. Can someone else? (Maybe I'll use subtract after work to try spotting differences)