[FFmpeg-devel] full swscale relicensing to LGPL
Fri Mar 27 11:37:30 CET 2009
Hi Carl Eugen,
Carl Eugen Hoyos wrote:
> Michael Niedermayer <michaelni <at> gmx.at> writes:
>>>> -vsync 0' options still belong?
>>> I did not test yet (manually), because I wanted to ask first if you (or
>>> Michael) have an automated way doing it, but my next intention was to find
>> i dont, and we lack some timestamp handling code yet for it to work except
>> by luck and its not moving us forward if such luck is tested for by fate
> Could you explain the issue? Is this the reason current MPlayer fails for all
> PAFF samples, even the ones that (more or less) played fine before?
> Both ffmpeg and ffplay svn work fine with all my samples.
Actually, timestamp handling as such should be correct by now for H.264
(except some exotic things like discontinuities). But there are some
small deficiencies regarding frame rate detection in MPEG-TS/PS streams
(and possibly other containers). For instance, 25 fps *interlaced* H.264
coded as field pictures (50 fields per second) gets double frame rate
(50Hz instead of 25Hz) and produces a full frame in decode_picture only
for each second tick (i.e., actual full frame rate is 25Hz). This is
handled correctly by ffplay and ffmpeg, simply by copying the frame and
dropping it again when encoding back to 25 fps, but may confuse other
programs using lavf/lavc.
I provided a patch for this, which wasn't accepted by Michael.
Unfortunately, due to personal reasons (my daughter was born) I don't
have time to make it better now.
There are also some other deficiencies in H.264 handling regarding
detection of key frames when the media is lacking SEI recovery point.
Again, I didn't have time yet to address this issue. Files produced by
Canon AVCHD camcorders lack SEI recovery point.
Yet another problem is proper seeking support, but that's a general one.
BTW, can you give me a link to a sample which doesn't play properly, so
I can look at it (if/when I get some time)?
More information about the ffmpeg-devel