QTGMC, VSpipe, and FFmpeg Issues

[cscad]

Apologies if this is the wrong subforum.

Been trying to deinterlace a VHS-C capture recorded with ffmpeg using QTGMC, specifically the "havsfunc" version for Vapoursynth, on Fedora Linux 37. For reasons I can't figure out, QTGMC causes something in my pipeline to choke and fail.

Here is my script:

import vapoursynth as vs
import havsfunc as haf
from vapoursynth import core
video = core.ffms2.Source(source='test.mov')
video = haf.QTGMC(Input=video, Preset="Slow", TFF=True)
video = haf.Deblock_QED(video, quant1=60, quant2=60)
video.set_output()

I'm running it through ffmpeg with this command:
vspipe -c y4m test-deinterlace.vpy - | ffmpeg -i - -c:v ffvhuff test-deinterlaced.avi

Which gives me this error:
double free or corruption (!prev)
pipe:: Invalid data found when processing input

Here is the video in question:
Link

It does seem to be specifically QTGMC, because when I remove that line from the script, I no longer have this issue. I have also tested this on another file captured in the same manner, with the same problem.

If there's any additional relevant information I should share, please let me know, and thank you in advance for any help you can give.

[poisondeathray]

What error message does --info give for vspipe ?

vspipe --info test-deinterlace.vpy -

[Selur]

Sounds like there is an issue with one of the filters.
Using vspipe like poisondeathray suggested might show details of what is failing exactly.

[cscad]

What error message does --info give for vspipe ?

vspipe --info test-deinterlace.vpy -

double free or corruption (!prev)
Aborted (core dumped)

I also got a crash report from my system, which gives the "reason" as "vspipe killed by SIGABRT."

[_Al_]

To eliminate vspipe or ffmpeg, what if you just request frames, to know if it is really just some filter>
Make your file a *.py, not *.vpy and just run it.

import vapoursynth as vs
import havsfunc as haf
from vapoursynth import core
video = core.ffms2.Source(source='test.mov')
video = haf.QTGMC(Input=video, Preset="Slow", TFF=True)
video = haf.Deblock_QED(video, quant1=60, quant2=60)
for n, f in enumerate(video.frames()):
    print(n, f)

[cscad]

To eliminate vspipe or ffmpeg, what if you just request frames, to know if it is really just some filter>
Make your file a *.py, not *.vpy and just run it.

import vapoursynth as vs
import havsfunc as haf
from vapoursynth import core
video = core.ffms2.Source(source='test.mov')
video = haf.QTGMC(Input=video, Preset="Slow", TFF=True)
video = haf.Deblock_QED(video, quant1=60, quant2=60)
for n, f in enumerate(video.frames()):
    print(n, f)

Just tried it, same error unfortunately:

double free or corruption (!prev)
Aborted (core dumped)

[_Al_]

Did python specified a line for that error, if it was deblock or some of qtgmc library used?
It works for me here, but on Ubuntu, exactly the same script as you posted, using some wonderful vapoursynth package, but about 4-5 years old, I forgot name and where it was from, I remember that repo was not available anymore or updated. Not really helpful, I know. Looks like, as always, library version problem, if not a bug, that error is some library filter error.

In theory I could provide some libraries you need, if you want, but does Fedora uses same libraries as Ubuntu? And vapoursynth is R45, python using is 3.6.9

[cscad]

Did python specified a line for that error, if it was deblock or some of qtgmc library used?
It works for me here, but on Ubuntu, exactly the same script as you posted, using some wonderful vapoursynth package, but about 4-5 years old, I forgot name and where it was from, I remember that repo was not available anymore or updated. Not really helpful, I know. Looks like, as always, library version problem, if not a bug, that error is some library filter error.

In theory I could provide some libraries you need, if you want, but does Fedora uses same libraries as Ubuntu? And vapoursynth is R45, python using is 3.6.9

Far as I can tell, Python didn't get any more specific than that. Fedora is on Python 3.10, and I'm 99% sure the version of Vapoursynth I have is R61 (I compiled it from source, so I'm not using Fedora's package for it).

When I have a bit I might try spinning up a few VMs and trying this on different distros, this wouldn't be the first time I did that for something.

[Selur]

Have you tried different presets in QTGMC? (different presets have different dependencies, so maybe that helps to narrow down which library is causing the issue)

[cscad]

Have you tried different presets in QTGMC? (different presets have different dependencies, so maybe that helps to narrow down which library is causing the issue)

So, just went through and tried all the presets. Most of the presets still failed, however "Super Fast" and "Draft" rendered successfully, and "Ultra Fast" gives this error:

Traceback (most recent call last):
File "src/cython/vapoursynth.pyx", line 2886, in vapoursynth._vpy_evaluate
File "src/cython/vapoursynth.pyx", line 2887, in vapoursynth._vpy_evaluate
File "test-deinterlace.vpy", line 5, in <module>
video = haf.QTGMC(Input=video, Preset="Ultra Fast", TFF=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^
File "/usr/lib64/python3.11/site-packages/havsfunc.py", line 1942, in QTGMC
edi1 = QTGMC_Interpolate(
^^^^^^^^^^^^^^^^^^
File "/usr/lib64/python3.11/site-packages/havsfunc.py", line 2280, in QTGMC_Interpolate
interp = Input.bwdif.Bwdif(field=field)
^^^^^^^^^^^
File "src/cython/vapoursynth.pyx", line 1820, in vapoursynth.VideoNode.__getattr__
AttributeError: There is no attribute or namespace named bwdif

[Selur]

Okay, 'Ultra Fast' fails because BWDIF isn't installed/available:
https://github.com/HomeOfVapourSynthEvolution/VapourSynth-Bwdif

Cu Selur

[poisondeathray]

I've never seen "core dump" or similar for vapoursynth on windows.

It sounds like some Star Trek thing where they "dump the warp core"

[Selur]

No, on Windows you get a blue screen or some uninformative message.

[forart.it]

Just out of curiosity: do you think is possible to apply "VHS enhancements" by QTGMC on the fly by piping ffplay into it ?

[Selur]

Not sure what you are aiming for. If you want to reencode using ffmpeg as encoder and Avisynth/Vapoursynth for video processing does work.
If you are asking whether ont could in theory write a script, to:
a. create a Vapoursynth/Avisynth script which applies QTGMC to a source (and decodes audio to wav)
b. uses ffplay to playback that script?
Yes, I think so, assuming ffplay can play .avs/.vpy files. I doubt this will work if piping is involved anywhere.

Cu Selur