[Libav-user] SEI picture timing message and timestamp in decoded frame, h264 es

[Carl Eugen Hoyos]

2019-02-13 18:53 GMT+01:00, Michal Rybinski <michal.rybinski+ffmpeg at gmail.com>:
> śr., 13 lut 2019, 16:28: Carl Eugen Hoyos <ceffmpeg at gmail.com> napisał(a):
>
>> 2019-02-13 16:15 GMT+01:00, Michal Rybinski <
>> michal.rybinski+ffmpeg at gmail.com>:
>>
>> > Is FFMPEG taking into account SEI picture timing message
>> > contents when establishing timestamp (pts or best effort
>> > timestamp) when my app reads decoded AVFrame (basically
>> > looking into pts and best_effort_timestamp fields)?
>>
>> I don't know the answer but I can confirm that there are known
>> (and apparently difficult to fix) issues with timestamps in raw
>> h.264 streams.

> If it does any difference, I might check things also with GPL version of
> ffmpeg that includes x264 decoder.

There is no x264 decoder.

> I get timestamping in raw  h264 is troublesome. If it wasn't, ffmpeg
> wouldn't be employed in the proces :-) Synchronization is a case I'm
> looking at currently.
>
> Nevertheless (sorry, being persistent because I want to understand) I
> noticed various kinds of SEI are parsed in ffmpeg:
>
> https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/h264_sei.c
>
> So given raw h264 elementary stream, it seems that info should be sitting
> (processed or not) and maybe used somewhere in ffmpeg structures (or my
> understanding what happens where in ffmpeg is totally wrong). Would you
> care to show me the right direction in ffmpeg sources to do my own digging?

You could start with the option "-debug_ts" that should show you timestamps
from different components of FFmpeg so you can see if they are
read by the decoder and if they are changed later.

Carl Eugen