[theora] Theora/Thusnelda features questions.

[Gregory Maxwell]

On Thu, Jul 23, 2009 at 9:03 PM, Ralph Giles<giles at xiph.org> wrote:
> On Thu, Jul 23, 2009 at 5:44 PM, Christopher Blizzard wrote:
>
>> To the untrained eye these seem like knobs that could be adjusted pretty
>> easily, but maybe with changes to the format.  Would they need changes to
>> the Theora format (i.e. a Theora 2) to bridge the gap to HD content?
>
> Yes, those could be fixed with a format revision. However, if you're
> going to go to the pain of upgrading everyone's decoder, why not use
> dirac? Or fix a whole bunch of other infelicities, less obvious to the
> untrained eye, and build a whole new codec? We've had this debate both
> ways several times. :)

It would be outright foolish to break compatibility and not fix more
than MVs. But where to draw the line is a hard question. I don't think
there is a natural breakpoint, and even some of the obvious fixes can
present gnarly IPR issues that need to be considered.  (I.e. Do we
really want long MVs in theora without differential encoding? ... if
not than the super-simple MV fix becomes legally and technically more
complicated)


> Anyway, it's not so much that theora can't do HD content, as that it's
> tuned for 'web video' bitrates. At high bitrates (higher than consumer
> bandwidth can stream) Dirac and h.264 will just do a better job
> encoding a 'visually lossless' version than theora will, and this is
> more obvious when you have more pixels.

For 30fps video the MV limit hits you when you pan faster than 465
pixels per-second.  I expect, but have not measured, at least for 720p
it's uncommon enough that it wouldn't be worth caring were it not for
the fact that the quality falls (or the bitrate rises) so enormously
when you cross that threshold.