[MPlayer-dev-eng] [PATCH] unified timing patch for H264

[Alexander Strange]

On Aug 24, 2010, at 4:29 AM, Pásztor Szilárd wrote:

> Here is a modified version again, this time without touching libavcodec.
> Return value of the decoder was insufficiently handled in vd_ffmpeg.c anyway
> because NULL was returned in all cases of any type of "error", making it
> impossible to differentiate between different decoding "errors".
> 
> What works:
> * all samples in http://samples.mplayerhq.hu/V-codecs/h264/PAFF/ except the
> hdforum one (which is badly muxed and never worked without dumping video)
> * the sony-hdr-cx6-avchd-interlaced-decoding-problem samples too
> * my obscure m2hd.ts too
> * the danish_dvb.ts too, now finding correct A-V sync on its own.
> 
> I changed some of the PTS recalculation code so that it only kicks in after
> A-V sync has been settled. This, I believe, is working quite failsafe now.
> 
> I would require samples with 0-sized frames in AVIs or non-coded VOPs in mpeg4
> streams to check if they work correctly. Can anyone help me out?
> If they are broken, may these cases checked *before* passing data to the
> decoder? That could help out.

0-sized frames are removed by libavformat beforehand, so they just appear as VFR frames in the decoder.

There's two different kinds of non-coded VOP use in MPEG4 - one in packed b-frames and one used to drop frames directly.

I have the first, but not the second, and I'm not sure if you need that. In either case, you can generate them with mencoder+xvid by setting frame_drop_ratio very high.

Packed B-frames: http://astrange.ithinksw.net/ffmpeg/ef-op-bframes.mov
MPEG-2 field pictures: http://samples.mplayerhq.hu/MPEG2/mpeg2_field_encoding.ts