[Libav-user] Help in understanding seeking behaviour of FFmpeg API on MP4 video with hint tracks

Anton Shekhovtsov shekh.anton at gmail.com
Sun Apr 19 12:41:32 CEST 2015

2015-04-19 12:55 GMT+03:00 declan harrison <harrison.declan at gmail.com>:

> I need help understanding the internals of FFmpeg API seek call with
> certain files.
> Im using the ffmpeg libav* API in an appliaction that reduces the bitrate
> of MP4 files.  The output MP4 file has 1 H264 video and 1 AAC audio track.
> The application supports seeking on the input file to a specifc video
> frame using
> the seek API call.
> I added some diagnostics to FFmpeg to help me see what happens during seek
> calls.
> So generally seeking works fine for me. I make seek API call on the video
> track to a PTS and the MP4 demuxer
> aligns its internal sample index pointers for each track, so the next time
> demuxer reads for the next frame, it will
> seek to  position the file pointer into the required offset to get the
> required frame. The behaviour is predicatble
> for input files that have audio and video tracks. So ffmpeg may 2 two
> phyical file/stream seeks if it need to
> locate the correct frame for audio and video.
> Posted this originally to ffmpeg-devel, which is the wrong mailing list
> for this query so re-posting here. Apoligies for this.
> However I have an issue when i Have an input flle with 6 tracks. 1 H264
> Video track, 1 AAC audio
> track, 2 Hint tracks (one for vidoe and audio), an OD track (has 1 sample)
> and a Scenes track (which has 1 sample).
> the layout of this file is ftyp-moov-mdat-free-free
> When I perform the seek API call this time on this file, the MP4 demuxer
> as expected aligns its internal sample index pointer for each track, so the
> next time we request a frame from demuxer it will seek to the correct
> postiion in the file to get the correct frame.
> This works as expected, we seek into the correct  location in the file,
> and align correctly for each of the other tracks.
> Only 4 of the tracks have samples at the desired time, audio, video and
> both hint tracks.
> So application is able to start demuxing the frames from this seek time
> (e.g. 1 second) for
> all 4 tracks.
> However after reading a certain amount of data another 2 seek requests are
> made
> internally by FFmpeg to a position in bothj of the Hint tracks at a time
> unrelated
> to the oriignal seek request e.g. 3 seconds.
> Im at a loss to explain, why FFmpeg internally needs to perform these 2
> addiitonal
> seek requests into each of the hint tracks, when no other seek API call is
> made.
> Surely 1 seek request, results in at maximun 1 seek request per track.
> However in this
> case I have 2 seeks on the hint tracks.   The 'Scene'  and 'OD' track dont
> have any sample
> for this seek location.
> Im hoping someone can explain, why this happens and how predictable it is?
> Thanks
> _______________________________________________
> Libav-user mailing list
> Libav-user at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/libav-user
I failed to fully understand your scenario, but you may want to set discard
= AVDISCARD_ALL for streams you dont care about.
Otherwise, if you have stream with only one sample, this effectively means
you always seek right to the beginning of entire file, so seek is not
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://ffmpeg.org/pipermail/libav-user/attachments/20150419/59b5fae0/attachment.html>

More information about the Libav-user mailing list