[Libav-user] Fwd: Why is av_seek_frame() inaccurate in some cases ?

[Lucas Soltic]

Début du message réexpédié :

> De : Lucas Soltic <soltic.lucas at gmail.com>
> Objet : Réexp : [Libav-user] Why is av_seek_frame() inaccurate in some cases ?
> Date : 16 avril 2012 12:07:55 HAEC
> À : "libavdevice and libavfilter. This list is about using libavcodec libavformat libavutil" <libav-user at ffmpeg.org>
> 
> 
> Début du message réexpédié :
> 
>> De : Lucas Soltic <soltic.lucas at gmail.com>
>> Objet : Rép : [Libav-user] Why is av_seek_frame() inaccurate in some cases ?
>> Date : 14 avril 2012 19:24:01 HAEC
>> À : "This list is about using libavcodec, libavformat, libavutil, libavdevice and libavfilter." <libav-user at ffmpeg.org>
>> 
>> 
>> Le 14 avr. 2012 à 18:49, Michael Bradshaw a écrit :
>> 
>>> On Sat, Apr 14, 2012 at 10:40 AM, Lucas Soltic <soltic.lucas at gmail.com> wrote:
>>> So I don't think I did anything fancy, but I still can't find out why the video seeking is inaccurate. I would be really grateful if someone had ideas about this issue.
>>>  
>>> There's two reasons I can see why you would be getting this result: 1) It could be seeking to the closest keyframe in the video, and 2) some demuxers seek by DTS and not PTS.
>>> 
>>> You say it works fine for H.264 videos though, so I'm not entirely sure what could be causing it.
>> 
>> Hello!
>> 
>> I tried to use AVSEEK_FLAG_ANY for the av_seek_frame() parameter, so that it does not only seek to keyframes, but it still gives important gaps.
>> As for the DTS, in the movie I read, they are equal to the PTS (for any of the ~10 seeking I did). These results concern the MPEG4v2 video.
>> I'm going to investigate further with other codecs but I wanted to tell these points.
>> 
>> Lucas SOLTIC
> 
> 
> Actually I noticed than when seeking (even with AVSEEK_FLAG_ANY), the real sought position is 0 to 10s before the position I wanted. Then, if the video PTS is 9.90s before the position I wanted, and I seek a little bit further (say 0.1s), the video PTS becomes the exact position I wanted. So it looks like it's just seeking to keyframes, whatever seek flag I use.
> 
> Considering this, are there more accurate ways to directly reach the expected position or should I just do avcodec_decode_video2() until the PTS becomes the one I wanted ? Decoding 10s of video could take some time on high resolution movies...
> 
> Regards,
> Lucas SOLTIC

Sorry for the spam, I found another interesting point.

Actually if I disable the audio stream seeking (ie. do not call av_seek_frame() for the audio stream), the AVSEEK_FLAG_* are correctly handled. And if I use AVSEEK_FLAG_ANY for the video stream, I exactly get the PTS I wanted (with artifacts in the image, but it's normal as I didn't seek to a keyframe), AND the audio PTS is also correct, although I didn't do any seeking for the audio stream. THUS I'm wondering why both streams are bound that way? Aren't they supposed to be independent ?

Now, considering that I don't want to get artifacts, I suppose I still HAVE to *don't* use AVSEEK_FLAG_ANY and do avcodec_decode_video2() for up to 10 seconds of video in order to get full images, wherever I seek to. Am I right ?

One more question: what is the advantage of avformat_seek_file() over av_seek_frame()? In the documentation it's written that the function isn't "stable" yet and should therefore not be used right now. But in when I've searched for solutions about my av_seek_frame() issue, I saw a lot of people using this "unstable" function. Why ?

Regards,
Lucas SOLTIC
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://ffmpeg.org/pipermail/libav-user/attachments/20120416/aae2a236/attachment.html>