[FFmpeg-devel] Regression in mxf decoding

[Matthieu Bouron]

On Mon, May 21, 2012 at 09:30:33AM +0100, Tim Nicholson wrote:
> On 18/05/12 18:08, Tim Nicholson wrote:
> > Just spotted that ffmpeg is failing to correctly determine the size of
> > V210 material in an mxf wrapper. Have been running a git bisect for a
> > while now and narrowed it down as follows, but have run out of time today.
> > 
> >[..]
> 
> OK its 70ca163bc558b2194fd9e73502bd13073c214ef5 ....

According to SMPTE 377M the values introduced by this patch are the
correct ones. I suspect the code which handle the frame layout values to be
wrong (maybe i am missing something).

In case of mixed fields layout the code should not double the fields heigh
to get the frame heigh since (quoting SMPTE):

MIXED_FIELDS (3): an interlaced lattice as for SEPARATE_FIELDS above, stored
as a *single* matrix of interleaved lines 1,2,3,4,5,6,...

So i beleive that doubling the field height to get the frame heigh is only
valid for separate fields layout.

Can anyone comment on this ?

> 
> ffprobe version N-39974-g70ca163-by_Tim Copyright (c) 2007-2012 the
> FFmpeg developers
> 
> [..]
> 
>     Stream #0:0: Video: v210, yuv422p10le, 720x1152, 25 fps, 25 tbr, 25
> tbn, 25 tbc
> 
> Whereas the correct answer is:-
> 
> 
> ffprobe version N-39973-gc90b8a7-by_Tim Copyright (c) 2007-2012 the
> FFmpeg developers
> 
> [...]
> 
>     Stream #0:0: Video: v210, yuv422p10le, 720x576, 25 fps, 25 tbr, 25
> tbn, 25 tbc
> 
> This is happening on all my samples of the BBC preservation format mxf's
> at SD.
> 
> The smallest sample I have is 5 seconds worth but 154M.

I can confirm the regression with the sample you gave me when i was working on
timecode extraction (BRD35721202.mxf).

Regards,
Matthieu