ffmpeg upscale videos with the sr filter using a docker container

ffmpeg's sr filter applies super-resolution methods based on convolutional neural networks to upscale videos.

The sr filter uses Libtensorflow and Cuda but the version of Libtensorflow must match your version of Cuda which must in turn match your NVIDIA driver and ffmpeg must be built with the tensorflow option.

This can make it tricky to set up, but luckily someone has built a docker image with ffmpeg built with all requirements preinstalled

In this video i show you how to install the ffmpeg-tensorflow docker image and use it with docker or podman, install the nvidia-container-toolkit which is used to upscale video with ffmpeg inside the docker container using your gpu

I also cover a workaround needed for systems running selinux like Fedora to allow the container access to the filesystem

ffmpeg sr filter https://ffmpeg.org/ffmpeg-filters.html#sr-1

super-resolution-podman. https://github.com/NapoleonWils0n/cerberus/blob/master/ffmpeg/super-resolution-podman.org

ff-super-resolution script https://raw.githubusercontent.com/NapoleonWils0n/fedora-bin/master/ff-super-resolution

ffmpeg-tensorflow https://github.com/MIR-MU/ffmpeg-tensorflow

video stack exchange https://video.stackexchange.com/questions/29337/how-do-the-super-resolution-filters-in-ffmpeg-work

docker https://hub.docker.com/r/miratmu/ffmpeg-tensorflow ... https://www.youtube.com/watch?v=mK8MOFte23o