Tuesday, 29 January 2008

BBC Festival of Technology

We had our Festival of Technology last week, where BBC Research and Innovation showed off some of the things that we've been doing over the last year. Dirac was there of course, showing some of our professional HW and 1920x1080 HDTV at 8Mb/s. But a big part of what we were showing came from the Virtual Production team: surround video, the iView system for virtual camera angles and a really neat image stabilisation tool for extracting a stable video image from tiny body-mounted cameras.

We've all seen these things on undercover reporter programs: cameras hidden in lapels or bags. Well, the quality is generally pretty awful as soon as the camera is moved. What the VP team have done is put a fish-eye lens on the camera and use real-time global motion compensation to stabilise it. Then you can pick a rectangular area to display and undo the geometric distortion. All this done in OpenGL on the graphics board in real time. Something that wouldn't have been possible a couple of years ago without some very powerful specialist hardware, done cheaply and easily to produce a great tool for TV.

Given sufficient resolution, maybe in a few years we won't even be using standard lenses in sports coverage : everything will be fish-eye lenses with virtual cameras cut out in the production gallery.

Monday, 21 January 2008

On the road to a Dirac standard ... at last

Ok, so I know that people think that Dirac disappeared into a black hole some while ago but we're still hanging in there and getting it done. We're just coming up to some really major milestones and things are looking really exciting.

First, Dirac (or part of it) is going to be an international standard. Yay! We made a cut-down version doing intra coding only and this has only just been submitted to the SMPTE. If it goes through it will become VC-2 (Windows Media 9 became VC-1 when they standardised it). After a lot of hard work fighting SMPTE's preferred Word format (yuk) it went in just before Christmas and is being voted on as a Committee Draft as I write this.

At the same time we've been updating the full spec and that's been published today. Version 1.0 covers the professional VC-2 stuff, whilst version 2.0 covers the whole system. If VC-2 is well-received we'll propose an extension so that it covers the whole of Dirac. Then at last there'll be a royalty-free video compression standard ...

There'll be a new release (0.9) of the Dirac reference software in the next couple of days to comply with the spec. The compression performance of the reference encoder keeps improving and we'll carry on adding tools in later spec versions (3.0 etc) so that it stays as competitive as possible.

The second really exciting thing is that Dave Schleef has been going great guns on the Schrodinger (NB the code is no longer in SVN - it's available from git clone git: //diracvideo. schleef. org/git/schroedinger.git) high-speed implementation of Dirac. This really is very fast and Wladimir van der Laan is doing great things to make it even faster with a CUDA implementation some of which Dave is merging in. Dave is getting really close to a 1.0 release that will be fast enough for realistic use for encoding and decoding.

The third thing is that our hardware partner Numedia have produced some Dirac hardware using the low-delay syntax. This isn't the stuff of streaming video but it allows the BBC and other broadcasters to re-use all our SD infrastructure to carry HDTV transparently, with almost no delay (so no audio sync cock-ups in production).

So it's been a long haul (the first software was released in 2004, although it feels like a decade ago) but we're getting there.