[Ffmpeg-devel] [RFC] rtjpeg and NuppelVideo decoder/demuxer

Reimar at mplayerhq.hu Reimar
Wed Feb 8 10:29:04 CET 2006

On Wed, Feb 08, 2006 at 02:07:48AM +0100, Michael Niedermayer wrote:
> yes, if the tables may change they _must_not_ be in extradata! you should
> somehow figure out if they are truly constant or not (RTFS/RTFM/ask author)

I have not seen any sample where they change, and if they are created by
e.g. mencoder they cannot change. But on the other hand there is not
really anything that would prevent them from changing. That's basically
why I did that two-way approach, in the hope of getting the benefits of
Also, decoding with the wrong tables is still better than decoding with
none at all, IMHO (can get an issue when I implement seeking) :-)

> > And one last thing concerning the demuxer: since nuv files contain
> > timestamps given in milliseconds for each packet, I set r_frame_rate to
> > the fps and used av_set_pts_info(vst, 32, 1, 1000);
> > Is this the correct way to do it? It is a bit annoying, since this way
> hmm you shouldnt need to set r_frame_rate, lavf should figure that out
> for you ...

Hmm okay, though in theory the fps given in the nuv header and the real
framerate do not really _have_ to be related...

> > MPlayer will not display the true framerate but 1000 fps instead...
> -> bug, someone needs 10l, probably myself

I don't think so. Since nuv uses msec timestamps it is essentially
variable framerate (or could be misused in that way). I think MPlayer
can only handle that by using 1/timebase as fps (same issue with wmv).
It's just a pain when remuxing/transcoding/whatever.

Reimar Doeffinger

More information about the ffmpeg-devel mailing list