This is the scenario: - unaligned PES and NAL encoding - the first NAL of the access unit begins at the very end of a ts packet, sometimes only the 0x00 of the trailing byte (unfortunately, this is still conformant to the standards!) - the video frame is so small (ex: typically still picture) it fits in a ts packet and a new PES is immediately started
Two sample files can be found here: a) https://0x0.st/HDwD.ts b) https://0x0.st/HDwd.ts For sample a, the first NAL (AUD) is splited this way: 0x00 / 0x00 0x00 0x01 0x09 And for sample b: 0x00 0x00 0x00 / 0x01 0x09 ffmpeg -i input.ts -f null /dev/null => Application provided invalid, non monotonically increasing dts... The parser can usually deal with unaligned packets thanks to the parser state, but here a new PES starts just right after the split and fetch_timestamp() does not know that the start of the PES was on the previous frame. An alternative straightforward fix directly in the mpegts demuxer is possible but really ugly, see: https://pastebin.com/J286CXDr I hope this proposal is looking better. It consists in two patches to get an identical output. The first patch fixes the fetch_timestamp mechanism. The second patch is to make parse_packet duplicate the side_data when spliting packets. It is not clear to me if this is required nor correct in a general manner? Nicolas Gaullier (2): avcodec/parser: fix fetch_timestamp in a scenario with unaligned packets lavf/demux: duplicate side_data in parse_packet() libavcodec/parser.c | 6 +- libavformat/demux.c | 23 ++- tests/ref/fate/concat-demuxer-simple2-lavf-ts | 164 +++++++++--------- tests/ref/fate/ts-demux | 8 +- 4 files changed, 107 insertions(+), 94 deletions(-) -- 2.30.2 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
