[FFmpeg-devel] [PATCH] Fix start_time in MKV Demuxer

[Alex Sukhanov]

On Sat, Dec 21, 2013 at 2:51 AM, Michael Niedermayer <michaelni at gmx.at>wrote:

> On Wed, Dec 18, 2013 at 06:47:10PM -0800, Alex Sukhanov wrote:
> > Problem:
> > MKV demuxer explicitly set previously start_time = 0. So in case if
> first timestamp of video file > 0, update_initial_timestamps() located
> inlibavformat/utils.c, didn't update start time.
> >
> > How to reproduce:
> > $ffprobe chunk_00180.flv
> > $ffprobe chunk_00180.mkv
> >
> > It's the same video but packed into FLV and MKV containers
> >
> > FLV: Duration: 00:02:30.54, start: 150.000000, bitrate: 50 kb/s
> > MKV: Duration: 00:02:30.54, start: 0.000000, bitrate: 50 kb/s
> >
> > Fix:
> > drop line in MKV demuxer which sets start_time = 0
> >
> > TESTED:
> > FATE passed (I had to modify expectations of some tests)
> [...]
> > diff --git a/tests/ref/fate/vp9-parallelmode-akiyo
> b/tests/ref/fate/vp9-parallelmode-akiyo
> > index 9668c54..0591e2c 100644
> > --- a/tests/ref/fate/vp9-parallelmode-akiyo
> > +++ b/tests/ref/fate/vp9-parallelmode-akiyo
> > @@ -10,21 +10,21 @@
> >  0,          4,          4,        1,   152064,
> 6719f3a6c22f05dc53dd3906e4154bd7
> >  0,          5,          5,        1,   152064,
> 8cd9a12761e35f67c278949cd3aee88f
> >  0,          6,          6,        1,   152064,
> 8cd9a12761e35f67c278949cd3aee88f
> > -0,          7,          7,        1,   152064,
> 0160dec415234d39f148e91f72d264ab
> > -0,          8,          8,        1,   152064,
> 9f90d96d67d9e9b3716abe2a3faa854e
> > -0,          9,          9,        1,   152064,
> 1edb312f9d0be7835b964a3ffa014759
> > -0,         10,         10,        1,   152064,
> 7614fd674609afccacd355aa2f714c75
> > -0,         11,         11,        1,   152064,
> cb46868706dd246878bebf354aff66f4
> > -0,         12,         12,        1,   152064,
> da36fe96cb4956036f890bb2f6d05b98
> > -0,         13,         13,        1,   152064,
> af0a178c68b719b369c8fa8537d38e65
> > -0,         14,         14,        1,   152064,
> ff03dbc436376fc60ac240cd6c4fc518
> > -0,         15,         15,        1,   152064,
> b0bf25e139556bd9067616db7e4f47b5
> > -0,         16,         16,        1,   152064,
> e70d5480c1f82fc877bbe1a8093f807a
> > -0,         17,         17,        1,   152064,
> 622fb43e6ff63834f0f680a68b49f6e6
> > -0,         18,         18,        1,   152064,
> c331ebba15f2290f174533dbffb3c27b
> > -0,         19,         19,        1,   152064,
> 15cb153425c55f7065fb36606c48972e
> > -0,         20,         20,        1,   152064,
> b95c7699639c51b08b3615ef7fa7046c
> > -0,         21,         21,        1,   152064,
> b4774148c71c9c184bda5a18294e459c
> > -0,         22,         22,        1,   152064,
> 795b7ce4c5e0dc343bd8f80ad6c1a454
> > -0,         23,         23,        1,   152064,
> 19163601b7b6138e2940cf28f6df6c7f
> > -0,         24,         24,        1,   152064,
> b9b388e0892c52df0680a30bfa954506
> > +0,          7,          7,        1,   152064,
> 32841e8b6916f1daea7ef1b980da1710
> > +0,          8,          8,        1,   152064,
> d959ebaa0f45cdf93caac47cb22621ff
> > +0,          9,          9,        1,   152064,
> 66f0368b106e9c32cca7fa328cca833f
> > +0,         10,         10,        1,   152064,
> 092ec763ee4fd5e160fd66891baa9f0f
> > +0,         11,         11,        1,   152064,
> 1b1f70815e3554088da4797fd7e044f2
> > +0,         12,         12,        1,   152064,
> 0754cf3a6eb98b585e302be006288064
> > +0,         13,         13,        1,   152064,
> 5692f89e7a8f9b03311a664c0dc212a8
> > +0,         14,         14,        1,   152064,
> 589de0dc2fdef4914f53e4be5900d2a8
> > +0,         15,         15,        1,   152064,
> 992476c60d387ce7717a627ce8d40edd
> > +0,         16,         16,        1,   152064,
> 72e4135ac776ed19ce2f07183bba34cb
> > +0,         17,         17,        1,   152064,
> c93d2e298b597f0d0e91f8960c1c4588
> > +0,         18,         18,        1,   152064,
> ff5bfc1fc0248d9f22fac3b2c3339f02
> > +0,         19,         19,        1,   152064,
> 2c2e80f377b58a5ae1887297830a0c50
> > +0,         20,         20,        1,   152064,
> fe7579b65f02835d12b7a75c7b8d9346
> > +0,         21,         21,        1,   152064,
> 68e21ff75484658a0728a61ea0379672
> > +0,         22,         22,        1,   152064,
> adf6224c4a110ce86f5d87d661bb86dd
> > +0,         23,         23,        1,   152064,
> 91d5ed4e2aa61c0ac2ff7989712459be
> > +0,         24,         24,        1,   152064,
> 37faefac606e73aea42f30a933e036de
>
> what is causing this change ?
> has this been checked to be ok ? / correct ? / benign ?
>
> [...]
> --
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
>
> It is dangerous to be right in matters on which the established authorities
> are wrong. -- Voltaire
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
>

Hi Michael,


Let me briefly explain how we process input videos.
We split input video on chunks, transcode them in parallel and then merge
transcoded chunks back in result video.

As long as we keep original timestamps in these chunks, start time must be
greater than zero for all chunks except first one. That's true for FLV, but
not for MKV currently.
In transcoder we use "-vsync 1" filter, which adds a lot of frames at the
beginning in case of MKV, because  start_time = 0

You can reproduce problem by running following command line:

$ ffmpeg -i chunk_00180.flv -vsync 1 -vcodec libx264 -an -y flv.flv
$ ffmpeg -i chunk_00180.mkv -vsync 1 -vcodec libx264 -an -y mkv.flv

you will see that mkv.flv file has a lot of frames at the beginning.


I probably could fix this problem on our side: disable vsync 1 filter, or
reset timestamps in splitter, but that would require a lot verification
testing and I think that making FFmpeg consistent across different
containers is good idea.

> has this been checked to be ok ? / correct ? / benign ?

I ran FATE, and played manually with some videos - they look good.
Let me know please if you would like me to run any specific/additional
tests.

Thanks for review.