vs-vfi

Using video frame interpolation models with vapoursynth. You can pipe the output directly into mpv or render it.

This repo uses a lot of code from HolyWu/vs-rife.

Install vapoursynth

sudo apt install yasm python3.9 python3.9-venv python3.9-dev
git clone https://github.com/sekrit-twc/zimg.git
cd zimg
./autogen.sh
./configure
make -j4
sudo make install
cd ..
pip install Cython
git clone https://github.com/vapoursynth/vapoursynth.git
cd vapoursynth
./autogen.sh
./configure
make
sudo su
make install
exit
cd ..
sudo ldconfig
sudo ln -s /usr/local/lib/python3.9/site-packages/vapoursynth.so /usr/lib/python3.9/lib-dynload/vapoursynth.so
pip install vapoursynth
sudo apt install ffmsindex libffms2-4 libffms2-dev

Just in case your libffms2.so path is different from mine, find it with sudo find /home -name "libffms2.so" and add it's path to inference.py.

core.std.LoadPlugin(path='/usr/lib/x86_64-linux-gnu/libffms2.so')

Usage

Modify stuff like input filename, resize dimension, fp16 and model within ìnference.py.

Watching video with mpv: (requires sudo apt install mpv)

vspipe -c y4m inference.py - | mpv -

Rendering with x264: (requires sudo apt install x264)

vspipe -c y4m inference.py - | x264 - --demuxer y4m -o example.mkv

Rendering with ffmpeg:

vspipe -c y4m inference.py - | ffmpeg -i pipe: example.mkv