2

I am performing chunked encodes of longer video files, where I've split the original file into individual sequences that I have encoded separately. These sequences are files of different length, depending on where the scene cuts appear—they may be between 2 and 5 seconds long. They all start with an I-frame and are standalone.

My encoded sequences are all MP4s, e.g.:

test_0000.mp4
test_0001.mp4
test_0002.mp4
test_0003.mp4
test_0004.mp4

They all have common properties:

$ mp4info test_0000.mp4

File:
  major brand:      isom
  minor version:    200
  compatible brand: isom
  compatible brand: iso2
  compatible brand: mp41
  fast start:       no

Movie:
  duration:   2016 ms
  time scale: 1000
  fragments:  no

...

Now, in order to play those with a DASH player, I have to create an initialization segment and individual fragmented MP4s.

I could generate the fragmented MP4s via mp4fragment which I run on each standalone MP4 file:

$ mp4info test_0000.m4s
File:
  major brand:      isom
  minor version:    200
  compatible brand: isom
  compatible brand: iso2
  compatible brand: mp41
  compatible brand: iso5
  fast start:       yes

Movie:
  duration:   2016 ms
  time scale: 1000
  fragments:  yes

...

But obviously, these are now not according to spec, and all contain a moov atom:

What I'd need is individual media segments with only one moof and mdat box, which then require an initialization segment with only a moov box.

How can I generate that from the existing, already encoded segments?

I know this appears like an XY problem. In principle, I could already segment my original file directly after encoding, and run those encodes at the same time, e.g. using ffmpeg's dash muxer, or MP4Box, however:

  • There is almost no control over the resulting segment sizes, with respect to minimum and maximum duration
  • This approach does not parallelize

I have also checked Bento4; it does not seem to offer this functionality. Neither does FFmpeg. MP4Box behaves similarly. They all assume you have one long file to start with.

I see I could splice off the ftyp and moov boxes from these “fake fragments” in order to create an initialization segment. But I would end up with segments containing multiple moof and mdat boxes, which is not according to the specification – it only allows one fragment and media data box:

4. Media Segments

[…] one optional Segment Type Box (styp) followed by a single Movie Fragment Box (moof) followed by one or more Media Data Boxes (mdat).

I guess I can live with this the styp not being present.

slhck
  • 36,575
  • 28
  • 148
  • 201
  • I don’t of of any tool to do this. But you just need to strip the ftyp and moov from every file, and save a copy for the init. I think a tool like dd could do it. – szatmary Oct 31 '18 at 15:06
  • Thanks – I'm just wondering why nothing is out there yet, as all tools assume you want to go from one big file into equal-sized segments. I was thinking about something like what you suggested, but then I'd still have segments containing multiple `moof` and `mdat` boxes (see question update), which I don't think would be valid? – slhck Oct 31 '18 at 16:08
  • The multiple moof/mdat is very common these days. It was added as part of cmaf to enable low latency live streaming. – szatmary Oct 31 '18 at 16:10
  • Sounds good, I'll investigate. Thanks for the pointers! – slhck Oct 31 '18 at 16:54

1 Answers1

0

I think the simplest way would be using MP4Box, see Split fragmented MP4 into multiple MP4 files for details.

Daniel
  • 1,208
  • 9
  • 14
  • Thanks for your answer, but could you please elaborate which command exactly would generate the initialization segment from existing fragmented segments? The command you've linked to only produces fragmented segments from a *long* sequence, but I already have the individual chunks. – slhck Nov 21 '18 at 07:56
  • I think the init segment should be created at the same time as the individual segments. However, you could try to use mp4box on a segment as it seems they contain the relevant information in your case. – Daniel Jan 02 '19 at 14:43