Dirac 1.0.0 Released

dylan_- writes "According to their website, 'Dirac is an advanced royalty-free video compression format designed for a wide range of uses, from delivering low-resolution web content to broadcasting HD and beyond, to near-lossless studio editing.' Now a stable version of the dirac-research codebase, Dirac 1.0.0, has been released. The BBC have already successfully used the new codec during the Beijing Olympics and are looking to push it to more general use throughout the organisation. The latest version of VLC (the recently released 0.9.2) has support for Dirac using the Schroedinger library."

Great, but...

I tried using the Schrodinger library but I'm uncertain it works. Plus, I can't find my cat.

Re:Great, but...

Plus, I can't find my cat.

Be glad for that, as long as you dont you might still have one.

Re:Great, but...

I tried using the Schrodinger library but I'm uncertain it works. Plus, I can't find my cat.

I hate to tell you, but your cat is dead. And/or not.

0xBBCD

I see the first 4 bytes are 0xBBCD.
British Broadcasting Corporation Dirac.

Re:0xBBCD

Isn't that just 2 bytes? :)

*nibbles on parent's geek card*

Re:0xBBCD

the fourcc is 'drac'

Open source overkill

From the FAQ:

What are the license conditions?

The Schrodinger software is available under any of the GPLv2, MIT or MPL licences. Libraries may also be used under LGPL.

Sounds like someone wanted there to be no question about whether it was open source.

Re:Open source overkill

The Schrodinger software is available under any of the GPLv2, MIT or MPL licences. Libraries may also be used under LGPL.

Sounds like someone wanted there to be no question about whether it was open source.

Sounds to me like the license exists in multiple states at once, which may be exactly the way Schrodinger would have liked it.

Re:Open source overkill

The GPL makes assurances regarding patents that the MIT license doesn't:

"Each contributor grants you a non-exclusive, worldwide, royalty-free patent license under the contributor's essential patent claims, to make, use, sell, offer for sale, import and otherwise run, modify and propagate the contents of its contributor version."

So if you use it as a GPL licensed library you can't get sued by the BBC or other contributors to the code.

Content

I was wondering where I could find some vids to check out quality vs. file sizes and found this [bbc.co.uk] index of demo files. Looks great in VLC, quite impressive even at lower bitrates.

News from OGG Theora, too!

Dirac isn't the only royality-free, patent-unencumbered video codec there is - Xiph's OGG Theora has been around a while already, yet failed to impress quality-wise up until recently. There's some really cool development going on however, and you may see some of the results achieved over there: http://xiphmont.livejournal.com/35363.html [livejournal.com]

It's noteworthy that the changes made only affect the ENCODER, thus no changes to the DECODER (the part of a codec all applications used to play back files have included) are necessary. This bodes very well for HTML5, which will include some support for Theora on at least Mozilla (and iirc Opera) browsers.

Re:News from OGG Theora, too!

Theora gets a bad rap for being outdated technology, but it does have a few advantages over MPEG-4 ASP: the loop filter, adaptive block sizes, and multiple reference frames, putting it closer to H.264 than MPEG-4 ASP. With these features, it's really a pretty strong showing from Xiph, and things can only get better as the encoder nears 1.0.

Re:News from OGG Theora, too!

This is a big point. The Encoder is far more important that the rest of the codec. Folks talk about xvid and divx as if they are codecs when really they are different encoders for mpeg4.

Both Theora and Dirac have plenty of space to move with regard to encoders.

However there is no easy way to measure "distortion" of the encoded image that matches the human visual system all that well. (unlike audio). But I expect most codecs to get better in the next few years because of encoders. (including h264).

Ironically h264 does so well because of the availability of a free, fast and good quality encoder done my the community. Not the license owners.

Re:News from OGG Theora, too!

In sound the idea of masking works really well. That is if there is a loud sound at a particular frequency we tend to not be able to hear sounds that are a low in pitch and a bit quieter (IIRC). Its effectively masked. The other big advantage is also the linear nature of sound.

But the human visual system is a *lot* more complicated. IIRC about 1/3 or our brains are used for visual perception. Currently we use PSNR (Peek signal to noise ratio) as a measure. But this has been shown many times to be a very poor indication of what we perceive. One example is blocking. Blocking cause straight lines to form in the image and our brains lock on to them far more quickly that other artifacts.

Next is the colour and the 2d nature of a image. Then add that the eyes do a bunch of preprocessing on motion perception and its getting quite difficult. Finally we have the method of comparison. Which often involves comparing still images from the video stream. Yet if thats a high motion scene the codec might be better off encoding these frames with low quality because we can't perceive the quality loss combined with fast motion.

Lets also not forget how many people think youtube is good quality or at worse, good enough!

Re:News from OGG Theora, too!

All the R&D papers I have read and from folk in the field working on this. Its well recognized that psychoacoustic models are far more developed than psycho visual models.

I don't doubt that some people can tell the difference between flac and mp3/ogg/aac. But the true number is far less than the claimed number (do a proper blind test to really find out). Also you don't design codecs for 0.5% of the population that can hear the difference, but for the 90% that can't and the other 9.5% that don't care.

Now its a fact that PSNR is used in most encoders. Its also widely recognized that it is not a good measure. I have done my own image compression and got better PSNR than jpeg per bit, and yet it looked far worse.

So I'm not really sure where you getting the idea that is even in the same category as audio.

Re:really?

Read their site. From the FAQ [diracvideo.org]:

Do you infringe any patents?

The short answer is that we don't know for certain, but we're pretty sure we don't.
We haven't employed armies of lawyers to trawl through the tens of thousands of video compression techniques. That's not the way to invent a successful algorithm. Instead we've tried to use techniques of long standing in novel ways.

What will you do if you infringe patents?

Code round them, first and foremost. There are many alternative techniques to each of the technologies used within Dirac.
Dirac is relatively modular (which is one reason why it's a conventional hybrid codec rather than, say, 3D wavelets) so removing or adding tools was relatively easy, even though this may mean issuing a new version of the specification.

Re:really?

Encoding and decoding is presently fat and slow. It's very much in development.

Re:really?

Since you claim this I assume that you tried the 1.0.0 already - I watched the promo vid, and it says the BBC is using the codec to handle HD content over their standard def infrastructure at very low latency (a few ms, if I remember correctly).

Nonetheless, this seems to be an interesting thing to keep an eye on, because the codec specs address good compression especially for very high bandwidths, which is going to be an important issue for movie post production/processing, HD content and the likes. The promo vid is well worth watching.

Re:Performance? Benefits?

We don't need another codec, per se, we need a royalty free codec, that can be legally implemented in FOSS situations, and others without a lot of legal overhead. Assuming it isn't markedly worse than others in performance terms, Dirac qualifies. If by some miracle(class II or greater) mpeg4 were available under such terms, there wouldn't be any point to Dirac; but that isn't exactly likely.

Re:Performance? Benefits?

How does it stack up to other codecs?

As I say below, unfortunately the quality is lacking compared to modern codecs like H.264 and even (dare I say) VC-1. Apparently that's just the nature of using wavelets. While they give a very natural style of compression on still images (JPEG-2000, etc), they do not translate well to moving sequences because, unlike all other current codecs, the image is not broken up into blocks that can then be tracked and diff'd in time. Still, it'll be interesting to follow Dirac, if only because they're taking a radical new approach with only Michael Niedermayer's Snow as a peer.

Re:Performance? Benefits?

As I state below. Most of codecs performance has to do with the encoder. At 1.0.0 its too early to tell if the format/codec design is limited.

However a great codec without a good encoder is no good at all. But its early days yet considering h.264 has been around for 5+ years.

Re:Performance? Benefits?

Dirac employs wavelet compression, instead of the discrete cosine transforms used in most older codecs (such as H.264/MPEG-4 AVC or SMPTE's VC-1). Dirac is one of several projects attempting to apply wavelets to video compression. Others include Rududu [2], Snow and Tarkin. Wavelet compression has already proven its viability in the JPEG 2000 compression standard for photographic images.

Yes it does :|

Re:For low values of success

Could it be that the BBC's slowness to offer HD is related to the fact that most license payers receive their broadcasts via analogue or "Freeview" digital, neither of which currently support it? I guess they have better things to spend their limited budget on.

Re:For low values of success

In a nutshell, yes. HD is also not much of a big deal to the vast majority of television viewers either. The only reason there is such a fuss over it in the United States is mostly because they are rolling out HD and digital at the same time: most of the improvement has come from the change to digital, not HD. In Europe it's not such a big deal because we've already switched to digital. HD is "nice" but it's not the huge leap in visual quality some people would like you to believe.

Re:For low values of success

Someone's evidentially not been watching Top Gear [topgear.com], which features some of the best camera work on TV and film.