Vapoursynth

[Myrsloik]

Originally Posted by Selur

Nice, hoping to view the wave front of an input file to find the audio delay of some files.
Thanks!

Cu Selur

Only trim, splice and raw pcm output from vspipe is planned for the first test version. There's however one little problem...

All existing audio sources suck so I'll probably have to improve FFMS2 first.

Feel free to start implementing audio plugins using the doodle1 branch. The API is conceptually stable (but tweaks will definitely be made).

[Jukus]

@Myrsloik
What about variable fps?

[Myrsloik]

Originally Posted by Jukus

@Myrsloik
What about variable fps?

What about you asking a complete question?

[Jukus]

Originally Posted by Myrsloik

What about you asking a complete question?

VS does not support variable FPS for output, will it be implemented?

[Lypheo]

Originally Posted by Jukus

VS does not support variable FPS for output, will it be implemented?

It does, though? Just have vspipe write the timestamp file with --timecodes and mux it in (or pass it to x264 or whatever).

Only trim, splice and raw pcm output from vspipe is planned for the first test version. There's however one little problem...

All existing audio sources suck so I'll probably have to improve FFMS2 first.

Feel free to start implementing audio plugins using the doodle1 branch. The API is conceptually stable (but tweaks will definitely be made).

Can we expect a test build soon or not until you’re done fixing ffms2?

[Myrsloik]

Originally Posted by Lypheo

It does, though? Just have vspipe write the timestamp file with --timecodes and mux it in (or pass it to x264 or whatever).



Can we expect a test build soon or not until you’re done fixing ffms2?

There needs to be at least one decent source filter for audio or there won't be much to test really...

[Cary Knoop]

I can't figure this out on Windows 10.

1. I installed Python using Anaconda (e:\anaconda3)
2. Installed Vapoursynth (all users) (e:\vapoursynth)

python myscript.vpy works fine!

import vapoursynth as vs
from vapoursynth import core

print(core.version())
clip = core.std.BlankClip(format=vs.RGB24, color=[0, 0, 0])
clip.set_output()

vspipe and vsedit don't work.

vspipe --info myscript.vpy -

Setting PythonPath in registry to "e:\anaconda3" gives:

Failed to initialize VapourSynth environment

Setting PythonPath to "" gives:

Fatal Python error: initfsencoding: unable to load the file system codec
ModuleNotFoundError: No module named 'encodings'

Current thread 0x00008fdc (most recent call first):

Registry entries:

Key Name:          HKEY_LOCAL_MACHINE\SOFTWARE\VapourSynth
Class Name:        <NO CLASS>
Last Write Time:   12/31/2019 - 4:26 PM
Value 0
  Name:            Version
  Type:            REG_SZ
  Data:            48

Value 1
  Name:            Path
  Type:            REG_SZ
  Data:            e:\vapoursynth

Value 2
  Name:            CorePlugins
  Type:            REG_SZ
  Data:            e:\vapoursynth\core\plugins

Value 3
  Name:            Plugins
  Type:            REG_SZ
  Data:            e:\vapoursynth\plugins

Value 4
  Name:            VapourSynthDLL
  Type:            REG_SZ
  Data:            e:\vapoursynth\core\vapoursynth.dll

Value 5
  Name:            VSScriptDLL
  Type:            REG_SZ
  Data:            e:\vapoursynth\core\vsscript.dll

Value 6
  Name:            VSPipeEXE
  Type:            REG_SZ
  Data:            e:\vapoursynth\core\vspipe.exe

Value 7
  Name:            VSRepoPY
  Type:            REG_SZ
  Data:            e:\vapoursynth\vsrepo\vsrepo.py

Value 8
  Name:            PythonPath
  Type:            REG_SZ
  Data:            e:\anaconda3

Created a batch file clearing all environment variables except for path, systemdrive and systemroot.
No change, same problem.

[feisty2]

I wrote some syntactic sugar stuff for the C++ API here: https://github.com/IFeelBloated/vapo...ster/sugar.hpp
C++2a support is required (I'm 100% sure C++17 aint cuttin' it), it compiles with "-std=c++2a" for GCC10

you can use

auto MakePlane(auto Pointer, auto Width, auto Height, auto PaddingPolicy)

to access the plane in a 2-dimensional style, out-of-bound access (automatic padding) is allowed for the source plane, but not allowed for the destination plane.

I'll give you a concrete example, to write a 3x3 box blur, you would do the following:

auto srcp = reinterpret_cast<const float *>(srcp8);
auto dstp = reinterpret_cast<float *>(dstp8);

auto padded_src = reinterpret_cast<float *>(malloc((width+2) * (height+2) * sizeof(float)));
for (auto y : Range{ height })
     std::memcpy(padded_src + (y+1) * (width+2) + 1, srcp + y * width, width * sizeof(float));

// lots of code here to deal with padding for "padded_src"

auto gc_addr = padded_src;
padded_src += (width + 2) + 1;

for (auto y : Range{ height })
     for (auto x : Range{ width }) {
             auto above = padded_src - (width + 2);
             auto below = padded_src + (width + 2);
             dstp[x] = (above[x-1] + above[x] + above[x+1] + padded_src[x-1] + padded_src[x] + padded_src[x+1] + below[x-1] + below[x] + below[x+1]) / 9;
             dstp += width;
             padded_src += width + 2;
     }

free(gc_addr);

with "sugar.hpp", that huge pile of nuisance is equivalent to simply

auto srcp = MakePlane<const float>(srcp8, width, height, Repeat);
auto dstp = MakePlane<float>(dstp8, width, height, Zero);

for (auto y : Range{ height })
     for (auto x : Range{ width })
             dstp[y][x] = (srcp[y-1][x-1] + srcp[y-1][x] + srcp[y-1][x+1] + srcp[y][x-1] + srcp[y][x] + srcp[y][x+1] + srcp[y+1][x-1] + srcp[y+1][x] + srcp[y+1][x+1]) / 9;
             // possible out-of-bound access like "srcp[y-1][x-1]" is automatically handled here with the given padding policy, you got nothing to worry about

The header includes 3 pre-defined padding policies

Zero
Repeat (or simply called "pad" on this forum)
Reflect

however, you can also define your own padding policy by completing the following function and pass it to "MakePlane()"

auto PaddingFunction = [](auto Canvas, auto Width, auto Height, auto y, auto x) {
      ...
};

note that the "PaddingPolicy" argument has no effect on "dstp" as no padding is applied here.

you are welcome to leave a comment if you have any suggestions or ideas to further improve this.

[feisty2]

simple live demonstration: https://godbolt.org/z/MHebt8

[Jukus]

How to properly process the video where 3 progressive frames, 2 interlaced and it turns out that every 5 frame is a duplicate?

[Boulder]

Originally Posted by Jukus

How to properly process the video where 3 progressive frames, 2 interlaced and it turns out that every 5 frame is a duplicate?

Standard IVTC? VFM to match fields followed by VDecimate to drop the dupes.

[PlazzTT]

I'm getting this error when compiling Vapoursynth on Linux (Mint 19.3)

I installed cython through pip.

"which cython" gives "/home/me/.local/bin/cython"

...
CXXLD libremovegrain.la
CC src/filters/vinverse/libvinverse_la-vinverse.lo
CCLD libvinverse.la
CC src/filters/vivtc/libvivtc_la-vivtc.lo
CCLD libvivtc.la
CYTHON src/cython/vapoursynth.c
/bin/bash: cython: command not found
Makefile:2217: recipe for target 'src/cython/vapoursynth.c' failed
make: *** [src/cython/vapoursynth.c] Error 127

Any ideas how to fix this?

[Jukus]

By the way, to the page:
http://www.vapoursynth.com/doc/insta...-from-packages
May be added that there is a ready-made solution for Debian:
https://www.deb-multimedia.org/

And, apparently, remove the link to packages for Ubuntu.

[outhud]

Where should Vapoursynth plugins (.so) be placed by default on Linux (Ubuntu) so that they are auto-loaded?

Is there a way to print the autoload folder?

I've tried /usr/local/lib/ and /usr/lib/x86_64-linux-gnu/vapoursynth/ but it seems the modules are not being found.

[Richard1485]

Originally Posted by PlazzTT

I installed cython through pip.

Did you install it with pip3?

pip3 install cython

[Are_]

Originally Posted by outhud

Where should Vapoursynth plugins (.so) be placed by default on Linux (Ubuntu) so that they are auto-loaded?

Is there a way to print the autoload folder?

I've tried /usr/local/lib/ and /usr/lib/x86_64-linux-gnu/vapoursynth/ but it seems the modules are not being found.

http://www.vapoursynth.com/doc/autoloading.html

Most probably in /usr/local/lib/vapoursynth in your case.

[Myrsloik]

Originally Posted by Myrsloik

Only trim, splice and raw pcm output from vspipe is planned for the first test version. There's however one little problem...

All existing audio sources suck so I'll probably have to improve FFMS2 first.

Feel free to start implementing audio plugins using the doodle1 branch. The API is conceptually stable (but tweaks will definitely be made).

Audio update:
A a best but slow audio source has been created so that's no longer a major problem (will still get some usefulness improvements over time).

Testing and debugging everything else is however taking longer than expected but audio will be fully supported by AVFS and VFW as well from the start and vspipe will be able to output wave64 headers.

[Richard1485]

^ Excellent news! Down the line, I hope that it might be possible to have an equivalent to AviSynth's Dissolve(), which affects video and audio.

[stax76]

I'm not sure what's the focus of this thread because my question is about vs host application development.

There is a weird exception happening here:

    case WM_DESTROY:
        DiscardGraphicsResources();
        SafeRelease(&g_D2D_Factory);

        g_vsapi->freeNode(g_vsnode);
        vsscript_freeScript(g_vsscript);
        vsscript_finalize();

        PostQuitMessage(0);
        return 0;

Exception thrown at 0x00007FFAFC5B448D (vapoursynth.cp37-win_amd64.pyd) in VapourSynth Direct2D Rendering.exe: 0xC0000005: Access violation writing location 0x0000000000000000.

Weird thing is it crashes only inside WM_DESTROY and not when this code is executed before WM_DESTROY !

How can this issue be debugged?

[Myrsloik]

Originally Posted by stax76

I'm not sure what's the focus of this thread because my question is about vs host application development.

There is a weird exception happening here:

    case WM_DESTROY:
        DiscardGraphicsResources();
        SafeRelease(&g_D2D_Factory);

        g_vsapi->freeNode(g_vsnode);
        vsscript_freeScript(g_vsscript);
        vsscript_finalize();

        PostQuitMessage(0);
        return 0;

Exception thrown at 0x00007FFAFC5B448D (vapoursynth.cp37-win_amd64.pyd) in VapourSynth Direct2D Rendering.exe: 0xC0000005: Access violation writing location 0x0000000000000000.

Weird thing is it crashes only inside WM_DESTROY and not when this code is executed before WM_DESTROY !

How can this issue be debugged?

It's with a trivial script (BlankClip only)?

Are there any restrictions on what you're allowed to do when handling a WM_DESTROY message? If not then I'd simply see what the call stack is in a debugger. Maybe there's a hint in there.