Automatic MP4 muxing without external tools

[5p4k]

Hello! This is a minimal implementation of a MP4 muxer directly in python, so we do not need a second pass through ffmpeg or avconv. This introduces a dependency from pymp4 and, through that, from construct.

To use it, just record to a .mp4 file or use format='mp4'. I tested this in several ways in on a Raspberry Pi 2 model B (this started actually as a separate project), and so far I had no issue.

What is missing:

• Documentation
• Testcases
• Possibly specifying multiple chunks in the stsc atom

Regarding the last item, I think that, formally, every time a different SPS header is issued together with a keyframe, we should specify a new chunk in there. Also, although it seems that the SPS/PPS parameters are usually always the same, perhaps we should allocate a different sample descriptor for those chunks. Nevertheless, the current one-chunk approach was playable on all platforms and all players I could get my hands on.

[waveform80]

Fascinating! Many thanks for the PR - proper MP4 muxing has been something I've been dying to research for picamera for ages but it's always been too big a job to start on top of everything else. Unfortunately, the external dependency this adds (pymp4) complicates things as it's not debian/raspbian packaged which means this'd break the primary means by which picamera is distributed (i.e. as a .deb package in raspbian). However, this is sufficiently interesting that it might be worth adding as an optional dependency for pip-based installs. I'll have a deeper dig into the PR for 1.14; I can't promise I'll include it yet, but it's definitely very interesting.

Many thanks for the PR, and sorry it's taken so long to get around to looking at it!

[5p4k]

Sounds good! Let me know if there is something that I can contribute with for this merge. In the past year I’ve been using this extensively with no issues.

Having it as a pip optional dep would be nice; re-implementing all those atoms required by MP4 is quite some effort.