Re: [rtcweb] H.261 encoding samples at typical bitrates - sign language example

[Maik Merten <[e-mail]>]

Am 05.12.2013 22:05, schrieb Justin Uberti:
> Ow, my eyes...

Indeed, but old codecs such as H.261, MPEG-1 Part2, H.263 etc. somewhat 
rely on postprocessing at the receiving end for acceptable perceived 
quality in low-bitrate scenarios.

I attached a decoded frame of the ~256 kbps clip: One without 
postprocessing, one with mplayer's "pp" video postprocessing filter, and 
one with mplayer's "pp7" postprocessing filter. Clearly the filters 
cannot recover information lost in the low-bitrate encoding, but make 
things much less hurtful to look at, especially in motion.

Real-world performance thus also depends on the receiving end. This 
unfortunately is hard to assess and completely depends on implementation 
details of the receiver.


Maik



> The 256 kbps and lower clips are unusable. The 512 kbps clip is
> borderline, but might be usable if the framerate was cut in half.
>
> Remember also that these test clips are far better than what would be
> obtained from consumer webcams (i.e. good lighting, no shake, no
> temporal noise), so real-world performance is likely to be worse than
> what you see here.
>
>
> On Mon, Dec 2, 2013 at 8:23 PM, Basil Mohamed Gohar
> <[e-mail] <mailto:[e-mail]>> wrote:
>
>     On 12/02/2013 10:03 PM, Basil Mohamed Gohar wrote:
>      > Let's let any further discussions about the usability of H.261,
>     or any
>      > other codec for that matter, use actual examples going forward.
>      >
>      > The following is a VERY quick test of ffmpeg's h261 encoder in the
>      > context of the IETF's rtcweb working group's discussion of an MTI
>      > (mandatory-to-implement) video codec.
>      >
>      > sine_irene_cif.y4m taken from derf's collection:
>      >
>      > http://media.xiph.org/video/derf/y4m/sign_irene_cif.y4m
>      >
>      > ffmpeg version N-58565-gc122e69
>      >
>      > bitrate=64k,128k,256k,512k
>      >
>      > ffmpeg -i sign_irene_cif.y4m -codec:v h261 -b:v $bitrate -g 30
>      > sign_irene_cif.y4m-$bitrate.h261
>      >
>      > http://media.basilgohar.com/rtcweb/sign_irene_cif.y4m-64k.h261
>      > (real rate: 157.8kbits/s)
>      >
>      > http://media.basilgohar.com/rtcweb/sign_irene_cif.y4m-128k.h261
>      > (real rate: 165.6kbits/s)
>      >
>      > http://media.basilgohar.com/rtcweb/sign_irene_cif.y4m-256k.h261
>      > (real rate: 289.5kbits/s)
>      >
>      > http://media.basilgohar.com/rtcweb/sign_irene_cif.y4m-512k.h261
>      > (real rate: 541.8kbits/s)
>      >
>      > I apologize for the bitrate inflation, but if I had more time I can
>      > tweak the settings for a more accurate number.  These are simply the
>      > rates that ffmpeg produced with such a short clip at the given
>     requested
>      > rates.
>      >
>
>     I've updated the encoding settings as follows to get more accurate
>     resulting bitrates, but ffmpeg's h261 encoder seems to bottom-out at
>     around ~140kbps, so the only examples from above that are close (after
>     using the new settings) are 256k and 512k.
>
>     for bitrate in {1..512}k; do ffmpeg -i ../sign_irene_cif.y4m -codec:v
>     h261 -b:v $bitrate -minrate $bitrate -maxrate $bitrate -bufsize $bitrate
>     -qmax 1024 -g 30 -y sign_irene_cif.y4m-$bitrate.h261; done;
>
>     All the above posted examples can be viewed with mplayer and a bash
>     command line using the following, if you're interested:
>
>     mplayer
>     http://media.basilgohar.com/rtcweb/sign_irene_cif.y4m-{64,128,256,512}k.h261
>
>     The full integer range of bitrates from 1 to 512 can be found here:
>
>     http://media.basilgohar.com/rtcweb/h261/
>
>     Target bitrate and actual bitrate start to match around 150kbps with
>     these new settings.
>
>     I am currently exploring other codecs with the same methodology and will
>     share the results accordingly.
>
>     --
>     Libre Video
>     http://librevideo.org
>     _______________________________________________
>     rtcweb mailing list
>     [e-mail] <mailto:[e-mail]>
>     https://www.ietf.org/mailman/listinfo/rtcweb
>
>
>
>
> _______________________________________________
> rtcweb mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/rtcweb
>