[Ffmpeg-devel] H.264 decoding bug

[Loren Merritt]

On Sun, 7 Jan 2007, Diego Biurrun wrote:
> On Fri, Jan 05, 2007 at 05:27:56PM -0800, Nicholas Schell wrote:
>> http://lists.mplayerhq.hu/pipermail/ffmpeg-user/2007-January/006054.html
>> I previously responded to ffmpeg-user with the above linked report,
>> but did not know much about the problem at the time.
>>
>> However now I have much more information which may be of help in
>> solving the bug. Thanks to nk215 asking about my thread on the mplayer
>> IRC channel, I know at least that other individuals on linux can
>> confirm the blocking in the samples with current ffmpeg builds. Just
>> as well he also found exactly which ffmpeg revision seems to cause the
>> problem, r6578.
>>
>> I compiled ffplay myself with mingw to confirm his results. Revision
>> 6577 shows none of the blocking, but 6578 does. It appears exactly the
>> same as the blocking in current in builds. Also if there were some
>> qualms about whether the files could in fact be the problem because of
>> an error on part of the H.264 encoder used, I tested each file with
>> the JVT reference H.264 decoder. All three files were decoded
>> flawlessly with no blocking at all. Not that it would matter but
>> CoreAVC also has no issues with these files.
>>
>> The sample files are now located here
>> http://samples.mplayerhq.hu/V-codecs/H264/
>
> I can confirm the blocking artifacts.

Calling it "blocking artifacts" is misleading. It's a bitstream parsing 
error, as expected since that the revision in question modified cabac. 
Given that the codec failed to parse the bitstream, the blocking is 
intentional error concealment.

--Loren Merritt