[Theora] Slowdown on lots of motion

[Stan Seibert]

<40D7EDB2.3010200 at forestfield.org>
Message-ID: <40D86756.7050300 at mailsnare.net>

J.B. Nicholson-Owens wrote:

>> (Disclosure: I produced the Honey theora files.)
>
>
> Due to quoting, it's not entirely clear to me who wrote this (Stan
> Seibert?) but I think I'm missing a part of the conversation.  I hope
> nothing pertinant to learning more about Theora or why "Honey" appears
> to stutter was discussed.  I'd like to learn why that happened to me,
> even when I play the small encoding of the movie.
>
> Thanks for providing us with a copy of "Honey" encoded in
> Vorbis+Theora.  This was most useful (and interesting to watch as
> well).  It was particularly generous to provide them under a CC
> license so we can share the movies.


Ah, maybe I should explain the full story:

At the South by Southwest (SXSW) music/film/interactive conference in
Austin this past March, I got a chance to see a panel hosted by Creative
Commons talking about CC and film.  One of the panelists was David Ball,
who was working with CC on releasing a film he wrote and directed under
a CC license.  Later, when we (some people at Xiph.org) were thinking
about Theora demo footage, I remembered David and suggested that Manuel
Lora (another Xiph.org guy) contacted him.

David was really excited by the idea.  He agreed to provide us with the
final DV cut of his movie (20GB!!) so we could try out Theora on a film
that had not been compressed to heavily already.  He uploaded the file
to my computer, and after some tinkering, I (Stan Seibert) made several
theora files with various frame sizes and quality levels.

So really, thanks to David Ball is in order.  Be sure to check out his
website at:

http://fourthwall.creativecommons.org

>> Trailers are a good idea.  We are actually working on getting
>> permission to distribute a trailer or two, though aren't going to be
>> blockbuster-grade movies like Harry Potter.
>
>
> That's fine with me -- I'm most interested in seeing what Theora does
> with video that hasn't been through any other codec (not transcoded)
> and I'm not sure that is possible with edited movie ads (which are
> delivered to the public in some lossy compressed encoding).  I'd like
> to see this because I'm interested in being able to learn (visually)
> what kinds of video Theora handles well and what it doesn't handle
> well.  I don't know that I could do this with transcoded video because
> it wouldn't be clear to me whether I was seeing the Theora codec not
> encode something well or if Theora had accurately encoded what it was
> handed -- and what it was handed was not encoded well.


Exactly.  We had some "source footage" available for people to test
video compression on http://media.xiph.org, but because of our temporary
hosting situation while our replacement server is fixed, that website is
not currently online.

Given the enormous size of video, most of the test clips are DV files,
which are compressed slightly.  (It seems to be about 1/4 the size of
raw YUV video.)  That's good enough to test compression for Internet
distribution, where you are looking at 20:1, 50:1 or higher compression
ratios.

> Is it possible to make this kind of video with consumer-grade equipment?

Sure!  DV cameras are pretty popular these days, and with a Firewire
cable and $15 Firewire PCI card, you can dump the digital video directly
to your PC.  Of course, camera quality varies depending upon your
budget, but the basics are fairly affordable.

> Would transcoding from a Flash animation help illustrate something
> along these lines?

Perhaps, though I would imagine the sharp color boundaries and solid
colors in most Flash animations would tend to be blurred.  It would be
sort of like testing JPEG compression on Illustrator documents rather
than actual photographs.  It would be useful for certain things, but
wouldn't be a normal test case.

> I'd also be interested to see portions of Honey encoded such that it
> traded away file size for accuracy.  A short scene with lots of motion
> which could be decoded without much computation would be interesting
> to me because then I could better discern if my computer just isn't
> fast enough to show Theora movies right now or if what I saw in Honey
> is the effect of encoder settings.


The theora encoder has a sliding quality scale just like Vorbis.  I
created several files using different quality settings, but we only
distributed one of them.  Honey is a bit long to use for quality
comparisons, but I'll see if we can dig up some nice video clips.  For
the next Theora release we'll try to have a single clip posted with both
the DV original, and several theora encodes at different quality
settings.  We'll also be sure to use a small enough frame size that
people's computers can keep up.

---
Stan Seibert