broken frame after seek in ogg container

[j]

ogg demuxer does not seek to keyframe:
running

ffmpeg -ss 23 -i test.ogv -s 640x480 -f mjpeg -an -vframes 1 test.jpg

the output frame is not fully decoded. it did not seek to a keyframe and just dumps output.

if insted one seeks on the decoded stream. the frame looks ok.

ffmpeg -i test.ogv -ss 23 -s 640x480 -f mjpeg -an -vframes 1 test.jpg

ffmpeg's ogg demuxer should be able to seek to the previous keyframe in the first case.
and decode from there

[Carl Eugen Hoyos]

(Complete, uncut output and possibly a short sample missing.)

I wonder if you aren't simply describing how -ss works for streams that do not only contain keyframes...

[j]

here a sample: ​http://v2v.cc/~j/samples/seektest.ogv

ffmpeg  -ss 10 -i seektest.ogv  -vframes 1 seektest.ogv.jpg

ffmpeg version N-32369-gc0323b9, Copyright (c) 2000-2011 the FFmpeg developers
  built on Sep  6 2011 17:15:36 with gcc 4.6.1
  configuration: --enable-gpl --enable-postproc --enable-libvorbis --enable-pthreads --enahble-libtheora --enable-libx264 --enable-libmp3lame --enable-libschroedinger --enable-libxvid --enable-libspeex --enable-libgsm --enable-nonfree --enable-libfaac --enable-version3 --enable-libopencore-amrnb --enable-libopencore-amrwb --disable-shared --enable-libvpx --enable-x11grab --disable-encoder=vorbis
  libavutil    51. 16. 0 / 51. 16. 0
  libavcodec   53. 13. 0 / 53. 13. 0
  libavformat  53. 11. 0 / 53. 11. 0
  libavdevice  53.  3. 0 / 53.  3. 0
  libavfilter   2. 39. 0 /  2. 39. 0
  libswscale    2.  1. 0 /  2.  1. 0
  libpostproc  51.  2. 0 / 51.  2. 0
[ogg @ 0x1b39780] Unknown skeleton version 4.0
[ogg @ 0x1b39780] max_analyze_duration 5000000 reached at 5000000
Input #0, ogg, from 'seektest.ogv':
  Duration: 00:00:21.52, start: 0.000000, bitrate: 1016 kb/s
    Stream #0.0: Data: none
    Stream #0.1: Video: theora, yuv420p, 720x576 [SAR 16:15 DAR 4:3], 25 fps, 25 tbr, 25 tbn, 25 tbc
    Stream #0.2: Audio: vorbis, 48000 Hz, stereo, s16, 80 kb/s
    Metadata:
      ENCODER         : ffmpeg2theora-0.27
      SOURCE_OSHASH   : f8455fdcb9f10909
Incompatible pixel format 'yuv420p' for codec 'mjpeg', auto-selecting format 'yuvj420p'
[buffer @ 0x1b3bfe0] w:720 h:576 pixfmt:yuv420p tb:1/1000000 sar:16/15 sws_param:
[buffersink @ 0x1b7a580] auto-inserting filter 'auto-inserted scaler 0' between the filter 'src' and the filter 'out'
[scale @ 0x1b3e600] w:720 h:576 fmt:yuv420p -> w:720 h:576 fmt:yuvj420p flags:0x4
Output #0, image2, to 'seektest.ogv.jpg':
  Metadata:
    encoder         : Lavf53.11.0
    Stream #0.0: Video: mjpeg, yuvj420p, 720x576 [SAR 16:15 DAR 4:3], q=2-31, 200 kb/s, 90k tbn, 25 tbc
Stream mapping:
  Stream #0.1 -> #0.0 (theora -> mjpeg)
Press [q] to stop, [?] for help
[theora @ 0x1b3da20] vp3: first frame not a keyframe
frame=    1 fps=  0 q=4.4 Lsize=      -0kB time=00:00:00.04 bitrate=  -4.4kbits/s dup=0 drop=1    
video:17kB audio:0kB global headers:0kB muxing overhead -100.129184%

results ​http://v2v.cc/~j/samples/seektest.ogv.jpg

ffmpeg -i seektest.ogv -acodec copy -vcodec copy seektest.mkv
ffmpeg  -ss 10 -i seektest.mkv  -vframes 1 seektest.mkv.jpg

ffmpeg version N-32369-gc0323b9, Copyright (c) 2000-2011 the FFmpeg developers
  built on Sep  6 2011 17:15:36 with gcc 4.6.1
  configuration: --enable-gpl --enable-postproc --enable-libvorbis --enable-pthreads --enable-libtheora --enable-libx264 --enable-libmp3lame --enable-libschroedinger --enable-libxvid --enable-libspeex --enable-libgsm --enable-nonfree --enable-libfaac --enable-version3 --enable-libopencore-amrnb --enable-libopencore-amrwb --disable-shared --enable-libvpx --enable-x11grab --disable-encoder=vorbis
  libavutil    51. 16. 0 / 51. 16. 0
  libavcodec   53. 13. 0 / 53. 13. 0
  libavformat  53. 11. 0 / 53. 11. 0
  libavdevice  53.  3. 0 / 53.  3. 0
  libavfilter   2. 39. 0 /  2. 39. 0
  libswscale    2.  1. 0 /  2.  1. 0
  libpostproc  51.  2. 0 / 51.  2. 0
[matroska,webm @ 0x1cf9780] parser not found for codec vorbis, packets or times may be invalid.
[matroska,webm @ 0x1cf9780] Estimating duration from bitrate, this may be inaccurate
Input #0, matroska,webm, from 'seektest.mkv':
  Metadata:
    ENCODER         : Lavf53.11.0
  Duration: 00:00:21.51, start: 0.000000, bitrate: N/A
    Stream #0.0: Video: theora, yuv420p, 720x576 [SAR 16:15 DAR 4:3], 25 fps, 25 tbr, 1k tbn, 25 tbc (default)
    Stream #0.1: Audio: vorbis, 48000 Hz, stereo, s16 (default)
    Metadata:
      ENCODER         : ffmpeg2theora-0.27
      SOURCE_OSHASH   : f8455fdcb9f10909
Incompatible pixel format 'yuv420p' for codec 'mjpeg', auto-selecting format 'yuvj420p'
[buffer @ 0x1d01640] w:720 h:576 pixfmt:yuv420p tb:1/1000000 sar:16/15 sws_param:
[buffersink @ 0x1e839e0] auto-inserting filter 'auto-inserted scaler 0' between the filter 'src' and the filter 'out'
[scale @ 0x1e829a0] w:720 h:576 fmt:yuv420p -> w:720 h:576 fmt:yuvj420p flags:0x4
Output #0, image2, to 'seektest.mkv.jpg':
  Metadata:
    encoder         : Lavf53.11.0
    Stream #0.0: Video: mjpeg, yuvj420p, 720x576 [SAR 16:15 DAR 4:3], q=2-31, 200 kb/s, 90k tbn, 25 tbc (default)
Stream mapping:
  Stream #0.0 -> #0.0 (theora -> mjpeg)
Press [q] to stop, [?] for help
frame=    1 fps=  0 q=6.2 Lsize=      -0kB time=00:00:00.04 bitrate=  -4.4kbits/s dup=0 drop=46    
video:31kB audio:0kB global headers:0kB muxing overhead -100.070312%

result ​http://v2v.cc/~j/samples/seektest.mkv.jpg

i would expect the same result for ogv and mkv.

[j]

looks better with this, not sure its the right thing to do though

[reimar]

I incorrectly mentioned this issue in commit d7b542ae, but that one only fixes issue #941.
This issue can be fixed in a much more obvious way, but it risks breaking seeking completely for some files, and fixing it really correctly might be a bit tricky:

--- a/libavformat/oggdec.c
+++ b/libavformat/oggdec.c
@@ -670,8 +670,8 @@ static int ogg_read_seek(AVFormatContext *s, int stream_inde
 
     ret = ff_seek_frame_binary(s, stream_index, timestamp, flags);
     os = ogg->streams + stream_index;
-    if (ret < 0)
-        os->keyframe_seek = 0;
+//    if (ret < 0)
+//        os->keyframe_seek = 0;
     return ret;
 }
 

[reimar]

Forget that patch, it just breaks seeking completely.
For some reason the seek code does not find any keyframes.

[reimar]

More correct patch sent to ffmpeg-devel:
oggdec: fix keyframe seeking when granule_is_start is 0.

[reimar]

Patch applied, output looks fine enough to me.