[FFmpeg-devel] ffvhuff bug when using median prediction + adaptive tables
Sat Aug 4 19:40:39 CEST 2007
I am experiencing a bug in current svn that was introduced on May 19th.
When encoding certain streams to ffvhuff .avi files with median
prediction + adaptive tables enabled, there are errors when playing back
the finished encode.
I've tried to figure out how to enable median prediction + adaptive
tables with ffmpeg svn (today) but cannot for the life of me figure out
what the commandline is, so I can only show how to reproduce with
mencoder using libavcodec (also todays svn, of course):
mencoder -oac pcm -nosound -o huffyuv_table_sample.avi -ovc lavc
Then if you run "ffplay -an huffyuv_table_sample_context_pred.avi"
you'll see 2 errors near or after frame 170:
Error generating huffman table
This error leads to a duplication of the last correctly decoded frame
instead of the error frame, which results in bad video.
We've already narrowed this down to a SVN commit; there was a short
discussion on mplayer-dev-eng from June 25th:
> Everything before patch:
> r9069 | lorenm | 2007-05-19 04:32:59 +0200 (Sat, 19 May 2007) | 2 lines
> change brute force search to min-heap. 3.6x faster generate_len_table,
> 8% faster ffvhuff encoding.
> works fine (this is everything newest revision except for the huffyuv.c file).
> All patches from 9069 -> now has the huffman table error.
Everything is fine if i enable only adaptive tables, or only median
prediction, but it seems this bug is triggered when enabling both.
I've uploaded this sample to
upload.mplayerhq.hu/MPlayer/incoming/ffvhuffbug/ and it's also available
I've also attached a valgrind log but I'm not much of a coder so I can't
tell if it's any help.
I'd appreciate it if someone could shine some light on this.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 71033 bytes
Desc: not available
More information about the ffmpeg-devel