[MPlayer-dev-eng] [RFC] h264 over rtsp without AVParser

[Carl Eugen Hoyos]

Hi Uoti!

On 2007-03-09 12:07, Uoti Urpala wrote:
> On Fri, 2007-03-09 at 04:13 +0100, Carl Eugen Hoyos wrote:
> > Possibly fix a bug in -correct-pts (Uoti?) -
> > If (framerate<10 && b_frames) correct-pts does not work!
>
> Exactly what does not work and how does it fail? I can't tell whether
> there is a problem, much less fix one, based on this description.

Sorry for the bad description - I shouldn't write reports like this at 4am!

I tried to play h264 streams that contain (only) NALs that are not
complete frames. These NALs have identical timestamps as long as they
belong to one frame. Unfortunately, mplayer sends those NALs to
libavcodec as if they would each contain a whole frame, which leads to
massive A/V desync: If two NALs correspond to one frame, video plays at
half speed (the more NALs, the worse the desync). This can be fixed by
the user by supplying a double (or higher) framerate.

If, instead, -correct-pts is activated, the streams are played correctly
(it parses the timestamps, if I understand that correctly).

But for one stream [1], 8 fps, which is the lowest rate I found, I get
lots of "pts value <= previous" messages, and the video stutters.

I thought (which might be completely wrong, I did not RTFS), that the
(negative) timestamp difference for b-frames gets so high for (only) 8
fps that -correct-pts can not recognize them as normal negative b-frame
pts difference.

Today, I tried to produce a sample with mencoder, but mencoder "fixes"
the timestamps so the whole problem cannot be seen. (And MP4Box doesn't
seem to accept the stream at all.)

Sorry again for the stupid report, Carl Eugen

[1]
rtsp://a2047.v1413b.c1413.g.vq.akamaistream.net/5/2047/1413/1_h264_110/1a1a1ae656c632970267e04ebd3196c428970e7ce857b81c4aab1677e445aedc3fae1b4a7bafe013/mw_2006_3_110.mov
If you want to test the stream, you would have to apply
http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/2007-March/050144.html