[FFmpeg-devel] frames not parsed correctly for VOB file (mpeg2)

[Christian Cier-Zniewski]

Michael Niedermayer schrieb:
> On Wed, Feb 03, 2010 at 10:52:23AM +0100, Christian Cier-Zniewski wrote:
>   
>> Christian Cier-Zniewski schrieb:
>>
>>     
>>> I have analyzed the packet data a bit further and realized that it mostly 
>>> contains at least two frames, instead of a single frame which would I 
>>> expect to be correct.
>>>       
>
> they are called field pictures
>   
Just before reading your reply I realized that the material is 
interlaced and that it is correct to return two fields in one packet.
Thanks for clarifying that. That is what I want. My patch is indeed wrong.
> i suggest you look in svnlog we did had an alternative implementation
> of things that did not merge field pictures but i changed it to what we
> have now because it at that time appeared simpler. I had not considered
> apparently that one could store timestamps on only the wrong fields.
>   
Could you please explain what you mean by "on only the wrong fields"?
Since it is ok to have the two fields per packet I have no need to 
change the current implementation.
> Iam not opposed to replace the current code by what we had previously
> if someone (like you) provides a patch that does this. I guess this will
> involve some adapting of the old code from compute_pkt_fields() the
> rest might be simple apply the diffs
> And of course it would require some testing with mpeg-ps/ts and h264
The "only" problem I still have is that no timestamps at all are 
returned by av_read_frame().
The current implementation should return PTS/DTS info though, shouldn't it?
Could you please point me in the right direction where the problem might be?
av_read_packet() successfully returns PTS/DTS info if they are inside 
the PES header of this VOB file.
But somewhere inside av_read_frame_internal() they get lost which should 
not happen.

Christian