Vapoursynth

[Myrsloik]

Originally Posted by Selur

@Myrsloik: Would be nice if core.text.Text would also support RGBH in the future, atm. it gives:

vapoursynth.Error: Text: Input clip must be 8..16 bit integer or 32 bit float, passed RGBH

so I have to convert to RGS and back

It's on the todo list. The reason it hasn't happened yet is that only some compilers support half precision types natively (gcc/clang both have the _Float16 and __fp16 extensions at least) and visual studio doesn't.

Software fp16 support is pointless since then single precision would be faster for all processing anyway. At best it would become a performance trap. Internally x86 cpus need to convert everything to single precision anyway.

[Myrsloik]

Does anyone still use the Python 3.8 support? Are the windows 7 fanatics still around? It's now been quite a few years since my last check.

[_Al_]

I use 3.8 because of using old opencv before opencv introduced a bug, reading wrong pixel information for a pixel (it is3/4 quadrant of pixel off), and they do not care to fix it. So forced to use older numpy, older python etc.
But do not mind me. Just saying, you do not have to be win7 fanatic to use older version. :-)

[Selur]

Does anyone know how to do the same as Avisynth

overlay(o, t, mode="subtract", opacity=0.15)

in Vapoursynth?
The widely used Overlay function from havsfunc, does not work as the Avisynth function does.
see: https://forum.doom9.org/showthread.php?t=186264
Vapoursynth overlay uses

expr = f'x y -'

which seems the intuitive way, but that is not what Avisynth does.


Cu Selur

Ps.: is there an alternative port of Avisynths Overlay than the one in havsfunc, that I'm not aware of?

[Myrsloik]

https://github.com/AviSynth/AviSynth...OF_add.cpp#L83
With of_add=false
Start transcribing into expr

[ChaosKing]

I asked chagpt for fun and it "corrected" it to

expr = 'x y - 0 max'

Should read as max(x - y, 0), but this is only a simple version of subtract.


Chatgpt: If your parser accepts this, you could use the following:

exprY = 'x y maskY * maskMode * y * (1 - maskMode) + -'
exprY = 'x (maskMode * y * maskY + (1 - maskMode) * y) -'
exprY = 'x (maskMode ? y * maskY : y) -'

For UV

exprU = 'xU (maskMode ? ((half * (1 - maskU) + maskU * yU)) : yU) - + half'
exprV = 'xV (maskMode ? ((half * (1 - maskV) + maskV * yV)) : yV) - + half'

Superdark-correction:

exprU = 'Yneg = max(0, -Y); mult = min(Yneg, over32); U = ((U * (over32 - mult)) + (half * mult)) >> shift'

ps I don't know what I'm doing

[GeoffreyA]

VSPipe from R71 detected as malware in AVG. I reported it as a false positive.

[Z2697]

AVG is now owned by the "security giant" Gen Digital.
https://en.wikipedia.org/wiki/AVG_AntiVirus
https://en.wikipedia.org/wiki/Gen_Di...d_acquisitions

It's hard to imagine why they are taking such act. I mean look how many companies and softwares acquired! (and agreed to be acquired?)
Some stories say that those products are much worse after the merges or acquisitions.

It's almost like they are trying to dictate what's malware and what's not.

[GeoffreyA]

Originally Posted by Z2697

AVG is now owned by the "security giant" Gen Digital.
https://en.wikipedia.org/wiki/AVG_AntiVirus
https://en.wikipedia.org/wiki/Gen_Di...d_acquisitions

It's hard to imagine why they are taking such act. I mean look how many companies and softwares acquired! (and agreed to be acquired?)
Some stories say that those products are much worse after the merges or acquisitions.

It's almost like they are trying to dictate what's malware and what's not.

From Grisoft to Gen Digital! I find that AVG has always been a bit high on the false positives. When I compile FFmpeg occasionally, every second executable from the Media Autobuild Suite gets quarantined, likely owing to heuristics, so I've got to add the whole directory as an exception. With OCCT, the linpack executable gets blocked. Other than that, AVG Free has been all right for me. I've been using it since 2007.

[Myrsloik]

This time I have something slightly different for you to test. This version should be functionally identical to the R71 release but with one important difference: Python 3.12 and later support. As in 3.13 and 3.14 as well.

Try it out and see if it works. The portable install script will probably download the wrong file or fail in some other hilarious way so install the portable version manually.

https://github.com/vapoursynth/vapou...ited-api-test1

[Z2697]

Originally Posted by Myrsloik

This time I have something slightly different for you to test. This version should be functionally identical to the R71 release but with one important difference: Python 3.12 and later support. As in 3.13 and 3.14 as well.

Try it out and see if it works. The portable install script will probably download the wrong file or fail in some other hilarious way so install the portable version manually.

https://github.com/vapoursynth/vapou...ited-api-test1

It should also work for any version >= 3.8? (also combine the two separate wheels for 3.8 and newer?)

[Myrsloik]

Originally Posted by Z2697

It should also work for any version >= 3.8? (also combine the two separate wheels for 3.8 and newer?)

No. Anything older than 3.12 (technically 3.11 if compiled for it but then it's slower) doesn't have a complete enough python version independent api.

So 3.8 support remains as it is for the windows 7 fans. In the future I probably won't drop support for python versions until they're end of life. So 3.12 is good until 2029 unless something unexpected happens.

[Z2697]

Originally Posted by Myrsloik

No. Anything older than 3.12 (technically 3.11 if compiled for it but then it's slower) doesn't have a complete enough python version independent api.

So 3.8 support remains as it is for the windows 7 fans. In the future I probably won't drop support for python versions until they're end of life. So 3.12 is good until 2029 unless something unexpected happens.

Do you mean that 3.11 is generally slower for VapourSynth, or the version independent api of it is slower?

[Myrsloik]

Originally Posted by Z2697

Do you mean that 3.11 is generally slower for VapourSynth, or the version independent api of it is slower?

It's python 3.11. I also think most people already have migrated to 3.12.

[Z2697]

Originally Posted by Myrsloik

It's python 3.11. I also think most people already have migrated to 3.12.

Somehow I'm still on 3.11 and think building new VS releases for 3.11 is easier than upgrading Python LOL
Better find some time to finally do it.

[Selur]

In the future I probably won't drop support for python versions until they're end of life.

That lessens the burden of trying to get pytorch&co working with Vapoursynth a lot.

Cu Selur

[Selur]

Using: https://github.com/Selur/Vapoursynth...icateFrames.py
with

FillDuplicateFrames(clip=clip, method="MV", thresh=0.030000).out

that calls:

  def interpolateWithMV(self, clip, n, start, end):   
    num = end - start
    sup = core.mv.Super(clip, pel=2, hpad=0, vpad=0)
    bvec = core.mv.Analyse(sup, blksize=16, isb=True, chroma=True, search=3, searchparam=1)
    fvec = core.mv.Analyse(sup, blksize=16, isb=False, chroma=True, search=3, searchparam=1)
    self.smooth = core.mv.FlowFPS(clip, sup, bvec, fvec, num=num, den=1, mask=2)
    self.smooth_start = start
    self.smooth_end   = end
    out = self.smooth[n-start]
    if self.debug:
      return out.text.Text(text="MV",alignment=9)
    return out

gives me tons of :

Explicitly instantiated a Cache. This is no longer possible and the original clip has been passed through instead.

and I'm wondering why?
Where am I explicitly instantiating a cache? How can I fix this?

Cu Selur

[Myrsloik]

You probably have an old version of some plugin that does it internally. Or that's my guess.

[Selur]

Since I this does not happen with other interpolation methods, it must be mvtools.
I checked, I'm using https://github.com/dubhater/vapoursy...leases/tag/v24 which seems to be the latest version.
Strangely using:

sup = core.mv.Super(clip, pel=2, hpad=0, vpad=0)
bvec = core.mv.Analyse(sup, blksize=16, isb=True, chroma=True, search=3, searchparam=1)
fvec = core.mv.Analyse(sup, blksize=16, isb=False, chroma=True, search=3, searchparam=1)
clip = core.mv.FlowFPS(clip, sup, bvec, fvec, num=50, den=1, mask=2)

instead of:

fdf = FillDuplicateFrames(clip=clip, method="MV")
clip = fdf.out

the problem does not occur.
So it must be something else in FillDuplicateFrames. Strange.
=> Thanks for the feedback, I'll try to figure out what is causing this.

Cu Selur

[Selur]

Correction this is caused by mvtools!
I simplified the script to:

# Imports
import vapoursynth as vs
# getting Vapoursynth core
import sys
import os
core = vs.core
# Import scripts folder
scriptPath = 'F:/Hybrid/64bit/vsscripts'
sys.path.insert(0, os.path.abspath(scriptPath))

# loading plugins
core.std.LoadPlugin(path="F:/Hybrid/64bit/vsfilters/Support/libmvtools.dll")
core.std.LoadPlugin(path="F:/Hybrid/64bit/vsfilters/SourceFilter/LSmashSource/LSMASHSource.dll")

clip = core.lsmas.LWLibavSource(source="G:/TestClips&Co/files/test.avi", format="YUV420P8", stream_index=0, cache=0, prefer_hw=0)

sup = core.mv.Super(clip, pel=2, hpad=0, vpad=0)
bvec = core.mv.Analyse(sup, blksize=16, isb=True, chroma=True, search=3, searchparam=1)
fvec = core.mv.Analyse(sup, blksize=16, isb=False, chroma=True, search=3, searchparam=1)
clip = core.mv.FlowFPS(clip, sup, bvec, fvec, num=50, den=1, mask=2)

# output
clip.set_output()

and calling:

VSPipe.exe c:\Users\Selur\Desktop\FillDuplicateFrames_MV_Cache.vpy NUL -c y4m

gave me:

Warning: Explicitly instantiated a Cache. This is no longer possible and the original clip has been passed through instead.
Warning: Explicitly instantiated a Cache. This is no longer possible and the original clip has been passed through instead.
Warning: Explicitly instantiated a Cache. This is no longer possible and the original clip has been passed through instead.
Warning: Explicitly instantiated a Cache. This is no longer possible and the original clip has been passed through instead.
Output 857 frames in 0.44 seconds (1926.36 fps)

=> will report to dubhater => https://github.com/dubhater/vapoursy...ools/issues/87