mpeg2video decoder loses frames when progressive_sequence flag flips

[SianLiao]

Summary of the bug:

In mpeg2video streams where progressive_sequence flag flips, one-frame will be lost due to context reinitialized in mpeg_decode_postinit call

Without entering the initialization block the otherwise lost frame is decoded correctly.

(ffmpeg binary causes a duplicate frame in that position when decoding)

How to reproduce:

Decode an mpeg2 stream with progressive_sequence flipping

ffmpeg version 2.2.1 and prior

[Carl Eugen Hoyos]

Please test current FFmpeg git head and provide your failing command line (the command that doesn't produce the expected output) together with the complete, uncut console output and please upload a sample to make this a valid ticket.

[SianLiao]

The file (too big to attach, 13 MB) is uploaded to upload.ffmpeg.org under filename 3743_mpeg_progressive_seq_flip.avi

Running ffmpeg decoder will cause "1 dup!", failing to decode a frame and therefore the frames 19 and 20 will be identical.

The frame otherwise is decodable, if no context reinit happens in the code as described above.

./ffmpeg -i 3743_mpeg_progressive_seq_flip.avi -loglevel verbose -y -an -f image2 "imgs/img_%05d.jpg"
ffmpeg version N-64226-g7e8fdf0 Copyright (c) 2000-2014 the FFmpeg developers
  built on Jun 26 2014 19:28:40 with Apple LLVM version 5.1 (clang-503.0.40) (based on LLVM 3.4svn)
  configuration: --disable-optimizations --disable-stripping --enable-debug=3
  libavutil      52. 90.101 / 52. 90.101
  libavcodec     55. 68.100 / 55. 68.100
  libavformat    55. 44.100 / 55. 44.100
  libavdevice    55. 13.101 / 55. 13.101
  libavfilter     4.  9.100 /  4.  9.100
  libswscale      2.  6.100 /  2.  6.100
  libswresample   0. 19.100 /  0. 19.100
Input #0, avi, from '3743_mpeg_progressive_seq_flip.avi':
  Metadata:
    encoder         : Lavf55.33.100
  Duration: 00:00:02.21, start: 0.000000, bitrate: 50237 kb/s
    Stream #0:0: Video: mpeg2video (4:2:2) (mpg2 / 0x3267706D), yuv422p(tv, bt709), 1920x1080 [SAR 1:1 DAR 16:9], 51233 kb/s, 50 fps, 25 tbr, 50 tbn, 50 tbc
    Stream #0:1: Audio: mp3 (U[0][0][0] / 0x0055), 48000 Hz, mono, s16p, 64 kb/s
[graph 0 input from stream 0:0 @ 0x7fa7585003a0] w:1920 h:1080 pixfmt:yuv422p tb:1/50 fr:25/1 sar:1/1 sws_param:flags=2
[auto-inserted scaler 0 @ 0x7fa758500de0] w:iw h:ih flags:'0x4' interl:0
[format @ 0x7fa758500980] auto-inserting filter 'auto-inserted scaler 0' between the filter 'Parsed_null_0' and the filter 'format'
[swscaler @ 0x7fa759000600] deprecated pixel format used, make sure you did set range correctly
[auto-inserted scaler 0 @ 0x7fa758500de0] w:1920 h:1080 fmt:yuv422p sar:1/1 -> w:1920 h:1080 fmt:yuvj422p sar:1/1 flags:0x4
[image2 @ 0x7fa758825a00] Using AVStream.codec.time_base as a timebase hint to the muxer is deprecated. Set AVStream.time_base instead.
Output #0, image2, to 'imgs/img_%05d.jpg':
  Metadata:
    encoder         : Lavf55.44.100
    Stream #0:0: Video: mjpeg, yuvj422p, 1920x1080 [SAR 1:1 DAR 16:9], q=2-31, 200 kb/s, 25 fps, 25 tbn, 25 tbc
    Metadata:
      encoder         : Lavc55.68.100 mjpeg
Stream mapping:
  Stream #0:0 -> #0:0 (mpeg2video (native) -> mjpeg (native))
Press [q] to stop, [?] for help
*** 1 dup!
*** 1 dup!9 fps=0.0 q=18.0 size=N/A time=00:00:00.36 bitrate=N/A dup=1 drop=0    
No more output streams to write to, finishing.:02.16 bitrate=N/A dup=2 drop=0    
frame=   56 fps= 20 q=24.8 Lsize=N/A time=00:00:02.24 bitrate=N/A dup=2 drop=0    
video:4148kB audio:0kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: unknown
Input file #0 (3743_mpeg_progressive_seq_flip.avi):
  Input stream #0:0 (video): 55 packets read (13833108 bytes); 54 frames decoded; 
  Input stream #0:1 (audio): 31 packets read (5952 bytes); 
  Total: 86 packets (13839060 bytes) demuxed
Output file #0 (imgs/img_%05d.jpg):
  Output stream #0:0 (video): 56 frames encoded; 56 packets muxed (4247491 bytes); 
  Total: 56 packets (4247491 bytes) muxed

[Carl Eugen Hoyos]

Regression since 0127b861
The attached sample outputs four frames for older FFmpeg versions, the third frame is missing since that change.

$ ffmpeg -vsync 0 -i 3743_mpeg_progressive_seq_flip.m2v -f framecrc -
ffmpeg version N-64228-g057d270 Copyright (c) 2000-2014 the FFmpeg developers
  built on Jun 26 2014 20:00:52 with gcc 4.7 (SUSE Linux)
  configuration: --enable-gpl
  libavutil      52. 90.101 / 52. 90.101
  libavcodec     55. 68.100 / 55. 68.100
  libavformat    55. 44.100 / 55. 44.100
  libavdevice    55. 13.101 / 55. 13.101
  libavfilter     4.  9.100 /  4.  9.100
  libswscale      2.  6.100 /  2.  6.100
  libswresample   0. 19.100 /  0. 19.100
  libpostproc    52.  3.100 / 52.  3.100
Input #0, mpegvideo, from '3743_mpeg_progressive_seq_flip.m2v':
  Duration: N/A, bitrate: N/A
    Stream #0:0: Video: mpeg2video (4:2:2), yuv422p(tv, bt709), 1920x1080 [SAR 1:1 DAR 16:9], max. 50000 kb/s, 25 fps, 25 tbr, 1200k tbn, 50 tbc
[framecrc @ 0x2430060] Using AVStream.codec.time_base as a timebase hint to the muxer is deprecated. Set AVStream.time_base instead.
#software: Lavf55.44.100
#tb 0: 1/25
Output #0, framecrc, to 'pipe:':
  Metadata:
    encoder         : Lavf55.44.100
    Stream #0:0: Video: rawvideo (Y42B / 0x42323459), yuv422p, 1920x1080 [SAR 1:1 DAR 16:9], q=2-31, 200 kb/s, 25 fps, 25 tbn, 25 tbc
    Metadata:
      encoder         : Lavc55.68.100 rawvideo
Stream mapping:
  Stream #0:0 -> #0:0 (mpeg2video (native) -> rawvideo (native))
Press [q] to stop, [?] for help
0,          3,          3,        1,  4147200, 0x4af40956
0,          4,          4,        1,  4147200, 0x468d0927
0,          6,          6,        1,  4147200, 0x32961317
frame=    3 fps=0.0 q=0.0 Lsize=       0kB time=00:00:00.28 bitrate=   6.0kbits/s
video:12150kB audio:0kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: unknown

[Michael Niedermayer]

the progressive sequence flag isnt allowed to change in mpeg, thus this is a bug in the used encoder.

Where is this file from ?

mpeg2:
In repeat sequence headers all of  the  data  elements  with  the  permitted
exception    of     those     defining     the     quantisation     matrices
(load_intra_quantiser_matrix,      load_non_intra_quantiser_matrix       and
optionally  intra_quantiser_matrix  and  non_intra_quantiser_matrix)   shall
have the same values as in  the  first  sequence  header.  The  quantisation
matrices may be redefined each time that a sequence  header  occurs  in  the
bitstream (Note  that  quantisation  matrices  may  also  be  updated  using
quant_matrix_extension()).
All of the data elements in the sequence_extension() that follows  a  repeat
sequence_header()  shall  have   the   same   values   as   in   the   first
sequence_extension().
If a sequence_scalable_extension() occurs after the first  sequence_header()
all    subsequent    sequence    headers    shall     be     followed     by
sequence_scalable_extension() in which all data elements are the same as  in
the    first    sequence_scalable_extension().

[Michael Niedermayer]

Fixed in 4d1fa38d984cba322284cfce416c4c969da58f51