On Sun, May 29, 2011 at 10:40 AM, Yusuke Nakamura <[email protected]> wrote: > 2011/5/28 Gil Pedersen <[email protected]> > >> >> The current code assumes that the first sample has the earliest >> composition time. When using b-frames, a codec is not required to enforce >> this, and thus I make a quick search for the sample with the earliest >> composition time. ISO/IEC 14496-12 section 8.6.1.1 has an open GOP example >> that illustrates this. As far as I can tell this composition time is what >> needs to go into "media_time" for proper sync. > > Do you wish to start presentation from incomplete samples such as > undecodable leading samples, which are samples that have earlier composition > time than the reference I-frame has, and depend on the earlier GOP deleted > from the stream? > > I think incomplete samples shouldn't be in any edit, i.e. whole > presentation, and you should add more delay by sum of duration of these > incomplete samples and set composition time of the reference I-frame to > media_time of second edit entry. > > So, I think it is enough for you to only set track->cluster[0].cts to > segment_duration of the first edit with timescale conversion and to > media_time of the second edit. >
I tend to agree with Nakamura-san's analysis of this. Are B1 and B2 from the Open GOP example supposed to be decodable without reference pictures before I1? Thanks, Alex _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
