So we have been working on the Schroedinger Dirac implementation for some time now and its starting to come together now.
The decoder is pretty fast and works well and the encoder is getting quite close too although its default settings needs to be moved away from developer settings. Not exactly sure where we stand in terms of being compliant with the latest Dirac specification, but we should be quite close as most of Dave’s commits recently have been about taking us the last few steps towards compliance. Anyway here is a screenshot I took today which is showing a video I created using this pipeline:
gst-launch-0.10 filesrc location=Dolphins_1080.wmv ! decodebin2 name="decode" decode. ! ffmpegcolorspace ! videoscale method=1 qos=false ! "video/x-raw-yuv, width=(int)640, height=(int)480" ! ffmpegcolorspace ! schroenc ! queue ! oggmux name=mux ! gnomevfssink location=file:///tmp/dolphins.ogg decode. ! audioconvert ! vorbisenc ! queue ! mux.
Two things stands out in this screenshot, one is that we need the Dirac plugins to report their codec and bitrate to the GUI :) The second is that the encoder needs tweaking so we don’t get the blur shadow at the ‘borders’ between different items in the image.
Dumb question I have never seen an answer to: why did BBC reinvent the theora wheel? Is theora just that bad? Or some other reason?
How do you see the two playing together in the future?
(Looking at the xiph website, I see that tarkin isn’t even mentioned anywhere anymore- I guess it is dead?)
BBC wanted something which could compete with H264 and WMV9 for high definition video, Theora can compete with MPEG2 on quality, but not with these newer codecs. I think in the future we will see these codecs offering both be used for a long while for different usecases.
Awesome news! Now all we have to do is convince the BBC to use their own codec for their content instead of Real :-)
Shameless self-promotion: the next version of OggConvert* will include support for encoding to Dirac (provided you’re running GStreamer 0.10.11 and have schroenc on your system). It’ll be out in the next couple of days, depending on whether I can fix the horrible pitfdll bug quickly.
* http://gnomefiles.org/app.php/OggConvert
How is the speed of the thing?
It is possible to play HD-sized stuff on a normal computer (say 3ghz P4)? I seriously doubt 1080p or whatever is the biggest, but maybe 1,280×720 at 30FPS?
Also when you guys have time a webpage with visual codec quality comparision vs bitrates would be entirely kick-ass.
If Dirac is actually fast enough to be usable on modern PCs and provides superior quality to Mpeg-4 ASP, and rivals Mpeg-4 AVC then your going to realy start to have something that people will be very interested in. Professionally interested in.
Luis: Theora is – at least in theory – roughly competitive with MPEG4-2 codecs (divx, xvid, etc. are implementations of that). Dirac is an attempt at jumping a generation-and-a-half beyond that (H.264 and WMV9 are generally considered the generation beyond MPEG4-2, Dirac’s being a little more ambitious than those).
The BBC people are also pushing for Dirac to be usable in the broadcast space – so they’re pushing it in directions quite different from what Theora is aimed at. We (the xiph people, that is) are interested in Dirac; we consider it complementary rather than a competitor.
And yes, Tarkin is very, very dead. It was an early-stage research prototype – and got more publicity than it should have. It was never planned as something that would be directly used.
That said, Xiph is very interested in designing and implementing a true next-gen video codec. We had a bit of discussion about this stuff at FOMS (before LCA). Unfortunately, that’s the sort of work that really requires someone working on it full-time for 6-12 months to get anything out of it at all – so we’d need funding from somewhere.
Of course, if you know anyone interested in funding R&D work on free codecs, please put us in touch! :-)