[FFmpeg-user] ffmpeg-55ed91c and x264-d9d2288. Profile setting differs from output.

[Joseph Rosensweig]

On Sat, Jul 21, 2012 at 7:44 AM, James Darnley <james.darnley at gmail.com>wrote:

> On 2012-07-20 21:32, Joseph Rosensweig wrote:
> > Hey all,
> >
> >    I assume I'm missing something really simple here.  But I'm not quite
> > sure what I'm doing wrong.  I checked out the ffmpeg and x264 git
> versions
> > indicated in my subject header.  I wanted to test out encoding H264 with
> > high, high10, high422, and high444 using libx264.  However when I try to
> > encode with anything above high, the output always just shows high rather
> > than say high 4:2:2.  Here are a few examples:
> >
> > Input #0, rawvideo, from 'pipe:0':
> >   Duration: N/A, start: 0.000000, bitrate: N/A
> >     Stream #0:0: Video: rawvideo (I420 / 0x30323449), yuv420p, 1280x720,
> 30 tbr, 30 tbn, 30 tbc
> >
> > Any help here :-)?
>
> Why would anything higher than High Profile be used for 8-bit 4:2:0
> video?  If you want to use High422 Profile then give x264 4:2:2 video
> (-pix_fmt yuv422p).  If you want to use High10 Profile then compile a
> 10-bit build of x264 and link ffmpeg to that.
>
> Setting the profile option for x264 does not force it to turn on all
> particular encoding features available in a profile.  In fact it does
> just the opposite: it disables all features that are not allowed or
> returns an error in cases where it cannot.
>

   Thanks for the replies.  I'm pretty new to FFMPEG so I'm picking these
things up as I go.  So for the first issue with the high444 and high422
coming out as high.  In an earlier version of FFMPEG, when I took raw 4:2:0
YUV data and encoded it using the ultrafast lossless preset.  FFMPEG would
show "High 4:4:4 Predictive", despite the fact that I was feeding it 4:2:0
YUV.  That is where this confusion came from for me.

   Basically what I'm gathering from this is that High10P, High422, and
High444 don't add anything that would be useful to an 8-bit 4:2:0 YUV
frame.  Using a high profile is sufficient.  I think I was under the
impression that in addition to adding support for more Sample Depths and
Chroma Formats, there would be more algorithmic features as well.