h264/README.md

# H264 transcoder with RPi4 hardware acceleration

This is a shell script to transcode video files to h.264.
It handles audio and subtitles gracefully.

It uses the Raspberry Pi's hardware acceleration
(`h264_v4l2m2m` codec)
to encode the h.264 files.

I don't know if the ffmpeg code is immature,
or if I just don't understand ffmpeg,
but this codec demands some special treatment,
which this shell script provides.

# `h264_v4l2m2m` Special Treatment

Here are the `h264_v4l2m2m` quirks I've identified
(February 2023):

* The codec will not write into a `.mkv` container,
  but it will write into a `.mp4` or `.nut` container.
* The codec does not play well with subrip subtitles,
  making it look like you don't have enough capture buffers.
  You will never have enough capture buffers:
  you have to stop trying to copy over subtitles.
* `-crf` value, if set, is ignored:
  you must use `-b:v`.

The solution this script uses is to
transcode video and audio
into an intermediate `.nut` file.
Then it combines the `.nut` file,
and any subtitles from the original file,
into a new `.mkv` file.
Since the "copy" video codec is used in the second step,
everything works properly.

Doing things in two steps adds a little bit of time,
but in my experience,
the hardware transcoder is so much faster,
the second step time isn't a problem.
I see ffmpeg reporting running at `speed=84x` and higher
on the second step.

# Thanks

Thank you to Will Usher,
who figured out how to get the codec working at all.
Without Will's
[helpful blog post](https://www.willusher.io/general/2020/11/15/hw-accel-encoding-rpi4)
I would still be fighting with ffmpeg trying to make it use hardware acceleration.
Will also wrote the
[FBED](https://github.com/Twinklebear/fbed) batch transcoder,
very similar to this.
You might try it out if you don't like this script.

I created this simpler shell script mainly
to avoid Python package dependencies:
FBED needed an ffmpeg library that I had to patch
before it would run.
I wound up having to fiddle with ffmpeg so much
dealing with the codec quirks,
that working in Bourne shell
made things much simpler
as I chased down the subtitle problems.