Re: [rtcweb] Congratuiations on the Cisco announcement

Agree.

Some previous tests between VP8 and H.264, like the one from the Atlanta presentation, have included a rate controller which does something similar to QP-toggling for VP8 but not for H.264. About two weeks ago we sent out a modified test where both codecs are allowed to QP-toggle, and that resulted in no difference between H.264 baseline and VP8 (http://www.ietf.org/mail-archive/web/rtcweb/current/msg09126.html). A more detailed analysis on the problems of previous tests can be found in http://www.ietf.org/mail-archive/web/rtcweb/current/msg09064.html. The result of no difference between H.264 baseline and VP8 is also consistent with the fixed QP test results we had published earlier (http://www.ietf.org/mail-archive/web/rtcweb/current/msg08052.html). We have long argued that compression efficiency tests are best made this way, i.e., without a rate controller, since small changes in the rate control parameters can have a huge impact on the end result, as the example of QP-toggling shows. 

But I fully agree to that quality isn't the thing to debate at the moment.

/Bo Burman

> -----Original Message-----
> From: rtcweb-bounces@ietf.org [mailto:rtcweb-bounces@ietf.org] On Behalf Of Mo Zanaty (mzanaty)
> Sent: den 7 november 2013 08:40
> To: Gregory Maxwell; David Singer
> Cc: Harald Alvestrand; rtcweb@ietf.org
> Subject: Re: [rtcweb] Congratuiations on the Cisco announcement - but we still prefer VP8
> 
> The previous comparisons used poor x264 settings and different rate controls.
> The latest comparisons used better x264 settings:
> http://www.ietf.org/mail-archive/web/rtcweb/current/msg09064.html
> ...and then more similar rate controls:
> http://www.ietf.org/mail-archive/web/rtcweb/current/msg09126.html
> 
> In the end, it is a draw, with or without rate control. Which was expected since the actual coding tools are virtually
> identical. We could have saved a lot of time and posturing if the codec experts on both sides would have just admitted
> this. Performance is a wash (when ignoring high profile).
> 
> Mo
> 
> 
> On 11/7/13, 4:47 AM, Gregory Maxwell <greg@xiph.org> wrote:
> 
> On Wed, Nov 6, 2013 at 10:35 PM, David Singer <singer@apple.com> wrote:
> > On Nov 7, 2013, at 15:19 , Harald Alvestrand <hta@google.com> wrote:
> >> Re performance, and just so that silence isn't taken as me agreeing
> >>with David Singer on this topic:
> >> I haven't wanted to distract from the debate at hand by tossing more
> >>numbers around, but we do think VP8 is significantly better than
> >>Baseline. The exchanges with Ericsson have shown that we need to be
> >>meticulously clear in defining what we mean by that and how we measure
> >>it, so I won't post more numbers until I feel that our description is
> >>precise enough.
> >
> > I also would like to see some tests which tell us where we really are.
> >A lot of us are engineers who like having facts, even if, as many of us
> >realize, the comparative performance is a minor part of the debate.
> >
> > I welcome Ericsson's efforts to get clarity in this area; they seem to
> >have been working very hard to get a level playing field measure, and
> >if you can help that effort, this would be all to the good.
> 
> Hm.  Maybe I'm confused: I thought slide 9 from the H.264 "joint presentation" in Atlanta (
> http://tools.ietf.org/agenda/85/slides/slides-85-rtcweb-10.pdf ) showed baseline needing 16% more bitrate than VP8 at
> the same quality, and the debate was more over high profile and the magnitude of VP8's lead wrt baseline.
> 
> (And, also in agreement: quality isn't the thing to debate in excruciating detail, but I did think at least this point wasn't
> controversial. Did I miss a post?) _______________________________________________
> rtcweb mailing list
> rtcweb@ietf.org
> https://www.ietf.org/mailman/listinfo/rtcweb
> 
> _______________________________________________
> rtcweb mailing list
> rtcweb@ietf.org
> https://www.ietf.org/mailman/listinfo/rtcweb