A Blog From Behind the Trenches

Originally posted by haavard:

And how does Google's acquisition of On2 fit into all of this?

This is really great news for all citizens of the web.

I really hope Youtube will start streaming video on Theora in some near future. I hate flash and want to get rid of it ASAP. But it's really hard when almost all major video sharing sites are using flash and I cant play h.264 in my beloved Opera.

http://files.myopera.com/Tamil/Smilies/Good.gif -

Originally posted by mgillespie:

Clearly the ARM port is because of Android handsets

Probable, but I like the implications for a certain other ARM-based handheld system I like

As I suggested in my blog, browser vendors should pump some money into Theora development. Theora is not quite good at the time. Looks like Google guys read my blog (: I think both Opera and Mozilla can donate some bucks too.

Another great thing to do is to create a special fund for support so all interested web devs could donate some money and/or code. I'm not good at video processing, never did anything in that field, but I would like to donate something to boost Theora improvement.

I just read this like five minutes back! .
.

And this is a shock....why? With newfound ownership of On2's existing project to make VP8 work with ARM, Google is the only body in a position to rapidly improve Theora.

Read this from On2's CTO: http://www.dspdesignline.com/showArticle.jhtml?articleID=214303691 It was published more than a year ago, while VP8 was announced even before that. Forget hardware acceleration for a moment; WHO is using VP8 on ANY platform in the real world?

Aside from a few gimmicky HTML5 sites and personal sites, WHO is using Theora right now? Of greater concern than any ARM software decoder, I would like to see Google working to improve Theora's container format. Or improvements to rate-based encoding (rather than quality-based encoding). On2's blurry, 1-1/2 year old comparison images make me sick. How did they make h.264 look so terrible at "2-pass HQ, 2mbps"?

Open video is great, free is great, platform support is great. Yet the Theora and VP8 comparisons only show small, blurry still images. When actual h.264 videos are heartbreakingly beautiful in real life, how do they mangle it so badly when they try to cast Theora as "better"?

EDIT: http://www.on2.com/index.php?342 This is just fake science, right? Perhaps like SpinRite's product claims? I think what they meant to say, was that unchanged blocks are "compiled to zero", similar to advances in browser javascript engines. If nothing has changed, no work needs to be done. However the way they present this technology, the "clocks" (wrong term) are "shut down", reducing total "power consumption" by 60% or more IN THEORY. Now if they were speaking of load-based allocation of POWER to necessary LOGIC UNITS, well even x86 CPUs have been doing that BY THEMSELVES IN HARDWARE for many years.

Originally posted by Aux:

but I would like to donate something to boost Theora improvement.

http://www.xiph.org/donate/ I gave $20 for Theora 1.1, and that turned out very nice

Originally posted by Aux:

I think both Opera and Mozilla can donate some bucks too.

+1 for Opera donating some, as stated by Haavard, Mozilla already has. Another great contributor is Red Hat, if I understand correctly, Red Hat pays Monty Montgomery to work on the codecs. He's the main commit contributor.

Originally posted by DanielHendrycks:

Mozilla already has. Another great contributor is Red Hat, if I understand correctly, Red Hat pays Monty Montgomery to work on the codecs. He's the main commit contributor.

Khm... I did not know about donate page. But now I have another quetsion - if Mozilla, RadHat and now Google donated money and code then why Theora still suck?

Originally posted by hellspork:

blah blah

Oh, quit trolling willya. Your constant walls of text whining about h264 are getting really boring. Stop trolling, and stop derailing all blog posts.

Originally posted by Aux:

if Mozilla, RadHat and now Google donated money and code then why Theora still suck?

Red Hat pays one guy, I think. Mozilla gave $100,000 which would cover some other people for some time, and Google is funding a fork, TheoraARM, not a Xiph project.

Originally posted by Aux:

why Theora still suck?

Not that many people is one thing, but Ptalarbvorm is looking wonderful, in some cases better than H.264. (They're at the pre-alpha 3 phase now) http://svn.xiph.org/experimental/derf/theora-ptalarbvorm/

Ptalarbvorm
H.264

(Look close)

Originally posted by hellspork:

Aside from a few gimmicky HTML5 sites and personal sites, WHO is using Theora right now?

Wikipedia immediately comes to mind as neither personal nor gimmicky.

Originally posted by DanielHendrycks:

Not that many people is one thing, but Ptalarbvorm is looking wonderful, in some cases better than H.264.

Well, I really hope Theora will be polished in a two-three years. But now it may look better sometimes, but what about resulting file size? According to many tests same quality as h.264 generates 2 times bigger files.

Originally posted by Aux:

But now it may look better sometimes, but what about resulting file size? According to many tests same quality as h.264 generates 2 times bigger files.

That's when you use bitrate to calculate it. That's the wrong way to do it.

Originally posted by Aux:

but what about resulting file size?

That we'll have to wait and see since I do not know how to encode something...

http://files.myopera.com/Tamil/Smilies/Good.gif -

Originally posted by prd3:

That's when you use bitrate to calculate it. That's the wrong way to do it.

Comparing quality with the same bitrate or comparing the bitrate at the same quality are the ONLY ways to two different videos. What other way would you advocate besides using bitrate? The size of the video is a function of the video bitrate + audio bitrate + (small) overhead from the container.

Originally posted by DanielHendrycks:

That we'll have to wait and see since I do not know how to encode something...

There have been a few comparisons already on Doom9 where h.264 still beats ptalarbvorm with the same quality at half the bitrate. Xiph's comparisons are by no means unbiased.
Here's one: http://img706.imageshack.us/i/lolfailt.png/
x264 beats Theora (Ptalarbvorm) at 1/2 bitrate. Theora is even slightly worse than Xvid at the same bitrate.

I'm hoping this is true:

http://newteevee.com/2010/04/12/google-to-open-source-vp8-for-html5-video/

Originally posted by ruario:

I'm hoping this is true

I also hope! And I hope codecs will be merged to some point. AFAIK they both have pros and cons. It would be nice if cons are merged and pros thrown away (:

Yeah that looks pretty cool. I know I come across as negative here but from appearances vp8 seems relatively competitive with h.264 (though if you read Dark Shikari's blog, it's likely a little bit worse). h.264 is already open source but having two high quality open source codecs will be great. Now hopefully Google can get rights to all of the patents on vp8 and make it patent-free!

Originally posted by MTKnight:

Wikipedia immediately comes to mind as neither personal nor gimmicky.

Thank you. I hadn't even realized that Wikipedia had begun hosting videos. Guess I should start digging around in there.

For the comparison images, I am most fascinated by the single four-way spread shot. Those two small and crappy stills from Greg are rather unprofessional. http://www.youtube.com/watch?v=MaCZN2N6Q_I Youtube looks pretty good in 480p these days.

I would be more interested to hear some better info about Greg's video experiments. I expect the (youtube) file was a video uploaded to, then downloaded from youtube. I would like to know more about the source file. Youtube certainly does compile for the bottom of the heap, explaining why it looks worse than anything an x264 user could make on their own. More information on ptalarbvorm's CPU usage and other matters would also be greatly appreciated.