Adobe Rants Over Linux Video Acceleration APIs

[Hoodlum]

Can't find any references on that one.

Nope you're right, that's wrong. It seems that was something I assumed after they began digitising their library with it losslessly (around the time it became known). While reading I found that it was apparently used at the Beijing Olympics.

[bugmenot]

I dunno where I read it, but I also remember I read something like it's more for HD content and not for streaming. but Wikipedia says something different:

Dirac format aims to provide high-quality video compression from web video up to ultra HD[2] and beyond, and as such competes with existing formats such as H.264 and VC-1.

from http://en.wikipedia.org/wiki/Dirac_%28codec%29


Maybe you were searching for Dirac in a Matroska (MKV) container, and there you found that the .ogm container is better for streaming with low bit rates than .mkv? http://matroska.org/technical/guides/faq/index.html (last line).

Maybe apart from the fact that dirac is not widely supported now, it could be the alternative to theora? It has most probably no patent issues and the quality is very good. Actually perfect for a free and high quality web video standard.

[Zhick]

Maybe apart from the fact that dirac is not widely supported now, it could be the alternative to theora? It has most probably no patent issues and the quality is very good. Actually perfect for a free and high quality web video standard.

Is it? I've tried several times to encode video in Dirac/Schroedinger using ffmpeg, but the encoding was either unbearably slow (libdirac) or the quality was mediocre at best (libschroedinger). But maybe I've been doing something wrong.
Anyway I guess Apple and Nokia wont accept Dirac for the same reasons they don't accept Theora: Risk of submarine-patents (bullshit, especially in the case of theora) and the lack of dedicated hardware-decoders (if it was chosen as the definite standard that'd probably quickly change, but existing devices like the iPhone/N900 will have to do the decoding in software).

[bugmenot]

I converted a few videos with "OggConvert" and it's not that slow and the quality depends on the settings.

BBC uses it, so actually it should not be too bad. I will test a bit more

[Apopas]

There are two ways to encode Dirac video.
By using Schroedinger or by using dirac.
The first one is faster but has uglier results while the second one is much slower but has better results.

[yotambien]

I find the debate about video codecs very interesting, and did some reading because I was basically uninformed. This time, instead of closing the sites after reading them I kept them open so I can share the links. You probably have read most of this, but perhaps someone find it informative.

On the issue about Dirac, I think this summary from G. Maxwell, from xiph, put it pretty well. Basically Dirac was designed to achieve high quality (up to lossless), but will have problems at the low bitrates required for streaming applications.

It appears that many video codec comparisons that are still lingering out there and being cited used versions of Theora that don't quite reflect its current capabilities. This update, from May last year, is quite telling. The comments in that link about PSNR mearuements are very interesting. Perhaps you remember that Slashdot article reading something like "Theora surpasses H264 for the first time". That was the answer from Theora people to a flawed article (this one) whose comparison pictures I remember seeing somewhere. Too bad that the benchmarks the Slashdot article referred to was also flawed, but to a much lesser extent (from a 20dB difference between H264 and Theora to round about 4dB). All three versions in the link, plus the explanation as to why this happened. It is interesting to note too that due to some bugs in ffmpeg2theora, the differences would indicate a real bad performance of Theora. Following more links it would appear that other benchmarks were done using this tool. So, for instance, this can't be considered representative. Also, nobody is apparently saying that Theora can beat H264 for general usage, only that it could do its job good enough for the web.

Also in that site, we are pointed to an article were the validity of PSNR comparisons are valid as long as the material and codec type remain the same. From the abstract, though, it's not clear to me whether it is safe to use it to compare different codecs.

WHATWG mailing list. June last year. C. DiBona (Google), says this:

Comparing Daily Motion to Youtube is disingenuous. If yt were to switch to theora and maintain even a semblance of the current youtube quality it would take up most available bandwidth across the internet.

G. Maxwell's reply became what I guess is the most known 'test' in favour of Theora. Response and comparison.

To me, it really seems that at current Youtube quality standards, Theora is perfectly valid. A similar, one-off test followed in the mailing list.

From about the same time we have one more comparison. The author stripped the audio to have a more fair comparison, although I don't understand the logic behind this. If Vorbis performance is good and actually helps the encoding to have a lower size and thus leave some bits to Theora, I don't see why that shouldn't be taken into account.

From a x264 developer, we have this , on animated content. Theora is basically destroyed there. There are some notes at the end explaining that some bug is making it look worse than what it could be. Others argue that H264 deals particularly well with this kind of content. Problem is that without pictures it's difficult to have an idea of what those numbers really mean.

This long thread from LWN is very informative about all this. There are many bits here and there, but perhaps I'd highlight the answer somebody got from the MPEG LA about free and open source implementations of H264. On a similar vain, some funny license restrictions from Final Cut Pro and W7.

The ffmpeg developer in the LWN thread has a good point, though. Google and other players in this story are most likely thinking ahead, so it makes sense to bet for the best performant codec given that quality requirements (and bandwidth available) are bound to go up with time. I don't know what are the chances that Theora could stand up in that battle against a much more modern codec, let alone win of course. VP8 perhaps?

Oh, I almost left this one out, regarding the patent threat to Theora and Dirac.

[yotambien]

I converted a few videos with "OggConvert" and it's not that slow and the quality depends on the settings.

BBC uses it, so actually it should not be too bad. I will test a bit more

Note that the BBC used Dirac internally during the Beiging Olympics, this is, to compress HD video and send it to themselves. That says a lot about the capabilities of the codec. However, it also says a lot that BBC's iPlayer uses H264 to deliver content to the end user (before that it used VP7 apparently).