[FFmpeg-devel] [PATCH] Fix empty G-VOP header decoding in MPEG-4

[Måns Rullgård]

Anatoly Nenashev <anatoly.nenashev at ovsoft.ru> writes:

> On 08.02.2011 22:00, M?ns Rullg?rd wrote:
>> Anatoly Nenashev<anatoly.nenashev at ovsoft.ru>  writes:
>>
>>    
>>> Hi all!
>>> There are some cameras which send mpeg-4 streams with empty G-VOP header.
>>> This part of stream looks like this:
>>> ... 00 00 01 B3 00 00 00 01 B6 ...
>>> Sample file uploaded in issue 2592.
>>> FFmpeg reports "header damaged" and ignores first I-frame in G-VOP.
>>> Attached patch fix this problem.
>>>
>>> Anatoly.
>>>      
>  [...]
>> Looking at this, I wonder why this is there at all.  The s->time_base
>> field is only used to incorrectly set the pts field of the decoded
>> frame (now I understand why those values always are wrong).  The
>> actual PTS is passed from the input packet to the pkt_pts field of the
>> output frame.
>>
>> IMO this nonsense should be removed.  The time_code field in the
>> bitstream has nothing to do with PTS.
>>
>>      
>
> Hmm... So, if I understand you clearly, we just can ignore GOP header
> as done in attached patch.

The value seems to also be used in a convoluted calculation of direct
mode MVs.  This calculation doesn't actually need the time_code value,
only the distance (in frames) between reference frames and the predicted
frame.  I'd be careful messing with it.

The part setting the PTS of the output frame should of course be
removed.  It is clearly wrong.

-- 
M?ns Rullg?rd
mans at mansr.com