[x264-devel] Re: using B-frames

[Måns Rullgård]

Alex Izvorski said:
> On Tue, 2007-03-13 at 10:32 -0700, Axel Gunter wrote:
>> I tried posting this at the ffmpeg-dev list but got no responses, trying
>> here since it pertains to x264 as well. Apologies for duplication.
>> -----------
>> Hi Experts,
>>
>> Do we know of any known problem using h264 video codec and having
>> B-frames? I am using mpegts as the transport.
>>
>> when using the command below the output video looks jerky with frames
>> being rendered out of order (some problem with timestamps using B-frames?)
>> ./ffmpeg -i source.ts -vcodec h264 -s 320x240 -refs 4 -bf 2 -y -f mpegts
>> -acodec copy output.ts
>>
>> However, when I disable B-frames using the command below, the video is
>> better (VLC says dropping late frames, but that might be a decoder
>> problem). At least the time stamps seem to be ok.
>> ./ffmpeg -i source.ts -vcodec h264 -s 320x240 -refs 4 -bf 0 -y -f mpegts
>> -acodec copy output.ts
>>
>>
>> Has anyone seen this behavior before? Any help is appreciated.
>>
>> -Axel
>
> Hello Axel,
>
> x264 essentially does not care about timestamps - it gets frames in what
> it assumes is the display order, and returns them in the correct
> decoding/transmission order based on that assumption.   There is a field
> for a timestamp (the PTS, in MPEG2 terms) in the input frames which is
> copied to the correct output frames, but otherwise completely ignored by
> x264 ;)  So the question is, does the ffmpeg muxer screw it up?  and the
> answer is, probably yes.

The FFmpeg TS muxer isn't very good.

> There *is* a way to set explicit frame timestamps in H.264: I believe it
> requires sending a picture timing SEI message before each frame
> (payloadType==1, defined in Annex D.1.2 in the standard).  However (a)
> x264 doesn't do this and (b) this is probably something better done at
> the container level and (c) most playback programs/devices probably just
> ignore it anyway and use the container timestamps instead.

You're referring to the SMPTE timecodes here.  They should not be used in
place of PTS.

-- 
Måns Rullgård
mans at mansr.com

-- 
This is the x264-devel mailing-list
To unsubscribe, go to: http://developers.videolan.org/lists.html