7 ms·
Mozilla and Xiph's new Daala video codec to compete with h.265
- y0ghur7_xxx 14y agoI'm a bit out of the loop, but what was wrong with Theora? Daalas goal "is to provide a free to implement, use and distribute digital media format and reference implementation with technical performance superior to h.265." Wasn't that exactly what Theora was as well?
- jfb 14y agoWell, Theora wasn't, was the problem.
- Avshalom 14y agomostly theora lost it's momentum because during the web video "debate" there were many articles shitting on theora for not having hardware support and being slightly to fairly worse in various cases. and now they need a new attempt to get people back on the open codec wagon.
- cpeterso 14y agoSince H.265 doesn't (?) have hardware decoders yet, maybe Daala will have a fair chance competing.
- nullc 14y agoThe folks working on H.265 have already been producing material (http://phenix.int-evry.fr/jct/doc_end_user/documents/10_Stockholm/wg11/JCTVC-J0128-v1.zip http://phenix.int-evry.fr/jct/doc_end_user/documents/10_Stoc...) to counter the belief that you must have hardware accelerators for video decoding. And they have a point: the _screen_, radio, etc. (http://www.ietf.org/mail-archive/web/rtcweb/current/msg03785.html http://www.ietf.org/mail-archive/web/rtcweb/current/msg03785...) of most mobile devices is already using something like 90+% of the power, so even in the impossible case that hardware accelerators make decode itself use no power they can't make that much of a battery life difference, so long as the device was fast enough to decode in the first place. And this trend will only continue as devices becomes faster and process shrinks lower the computing per joule. But it's amusing to see some of the same people arguing that hardware accelerators were _must have_ are now arguing the opposite now that its their latest format which is suffering for the lack of them.
- ubernostrum 14y agoRemember that it wasn't just hardware decoders in end-user devices. H.264 already had a lot of buy-in from video producers. With support at both ends of the chain, trying to replace it just for the sake of patent issues (which most people didn't care about) was a fool's errand.
- Xcelerate 14y agoTheora was never very good. Not really the fault of its creators -- rather, all the good methods of compression are being hoarded by patent trolls (MPEG LA and friends), even though most of the methods are obvious. I believe this new attempt at a codec will also face the same issues.
- nullc 14y ago>rather, all the good methods of compression are being hoarded by patent trolls Nah, that is simply untrue. Theora was designed in the late _90s_ and targets a different computational envelope than AVC— it's pretty darn good compared to MPEG-2 (or even MPEG4 part-2/DivX) which are more contemporary. Today there is a much large computational budget available, plus new experience and understanding.
- Xcelerate 14y ago>> rather, all the good methods of compression are being hoarded by patent troll > Nah, that is simply untrue. Could you provide something to back this up? I cannot find it now, but I seem to recall an x264 developer discussing the difficulties in implementing an open source codec equivalent to or better than H.264 and the problem was mainly due to the fact that all the best algorithms had been patented. Perhaps if you are a Theora developer (or know one), you can clarify.
- hermanhermitage 14y agohttp://x264dev.multimedia.cx/archives/377 http://x264dev.multimedia.cx/archives/377 (might be the link?)
- nullc 14y ago> Perhaps if you are a Theora developer (or know one), you can clarify. I am a Theora and Opus developer, although I'm not exactly sure what clarification you'd like. I can tell you that in my codec experience the patents have seldom (never?) been a major direct barrier to progress by precluding an essential technique... By their nature they tend to not absolutely foreclose anything except outright copying a technique— and even then only in a specific context, and video coding and signal processing are old enough and mathematical enough that the basics are unpatentable. (Keep in mind— the patent office does not believe it allows patenting pure mathematics, their definition of "does not" and "pure mathematics" and mine may not agree, but their evil is finite and thus surmountable) The impediment from patents seems to most often take the form having to spend time and effort in patent research and negotiations, not being sure that you can just implement some randomly discovered research, erroneously discarding some useful techniques which could be used but isn't worth the effort to clear, and the cost of spending time with attorneys teaching them enough codec engineering— or, frankly, spending time correcting misconceptions on the Internet— rather than coding. Or in short, _A patent_ is almost never a big problem for a designer, but _the patent system_ wastes a lot of effort by creating big largely non-engineering overheads that sap engineers time and energy. The system also complicates or discourages cooperation by creating odd business motivations and incentives to be secretive (especially about defense strategies). But the non-free codecs suffer from some of these costs and pressures too— plus additional ones like arguing over which winners and losers get their techniques in the format and thus a share of the fees (and access to cross licensing). This mess also exists just as much outside of media codecs. But the enforcement is less active— I suspect partially because codecs are unusually attractive to monetize due to network effects (switching is MUCH more expensive) and because they make nice attorney-understandable atomic units of infringement that map to visible features. "I own h264, you have h264. Pay up!" works better than "I own computing the absolute value by this series of bit operations. You may or may not do this, I can't tell because of your obfuscated binary. Pay up! Maybe?". The network effect also makes very narrow patents more useful— it's much easier to write a patent that reads on implementations of H.264 (a single format with a fairly exact specification) than one that reads on any format similar to H.264. Very narrow patents are less costly to obtain and enforce (Less risk of invalidation), and the network effect says that you must implement H.264 not almost-H.264 so they are no less good at extracting royalties. But no one really cares if their kernel uses xor-linked-lists or not, and it's usually no compatibility problem to switch if someone starts making threatening noises. If you were thinking of Jason Garrett-Glaser's early technical analysis of VP8, I don't think he was saying quite what you walked away with... but it's also important to note that Jason didn't (at least at the time) have substantial video coding experience outside of his work on x264 (and some related things in ffmpeg), and didn't have substantial experience working with patents, and had never been involved in a RF codec effort of any kind (including the ones attempted in MPEG). He's "just" a particularly brilliant guy that came in writing assembly code like a force of nature and made x264 much better. To the extent that he could have been emitting the impression that everything in video is patented he would have been just repeating the not-very-well-informed conventional wisdom. Patent infringement is all about the fine details— so even a patent expert's off-the-cuff comments are going to be somewhat useless. I'd take Jason's thoughts on the latency of PAVGW as the word of God, codec patents, when he'd not even looked at the involved patents? meh. Later revisions of his analysis substantially softened some the remarks, but few people went back to read them. (I recall that I was especially amused by some of the things he derided as being 'unnecessary', as I thought I had a reasonable guess which patents Google/On2 had been specifically dodging).
- butterflyhug 14y agoH.264 != H.265. Theora tried to compete with the former, Daala will try to compete with the latter.
- shmerl 14y agoTheora can't compete with H.264. If something, VP8 could.
- jfb 14y agoI'd be interested in some documentation. I wish them luck.
- netvarun 14y agoObligatory XKCD link: http://xkcd.com/927/ http://xkcd.com/927/
- tete 14y agoA codec isn't exactly(!) a standard.
- Klinky 14y agoI am going to have to say that MPEG-4, and specifically MPEG-4 Part 2 and MPEG-4 Part 10 are indeed video encoding standards. The different individual parts even have their own ISO numbers. Thus I do feel the "relevant XKCD comic" is relevant, even if some find the posting of such comics annoying.
- Dylan16807 14y agoThe thing is, the reason for new codecs isn't better covering existing use cases, it's using better knowledge and techniques and wildly changing space/computation/power requirements to fit a new profile. There's no need to replace telnet. There is a need to replace gif wrt animated images.
- Klinky 14y agoA) H.264 replaced MPEG2 largely to make video on disc look better, i.e. an existing use case. B) I am not sure what use cases have to do with the point the XKCD comic was making. The open web people wanted Theora, Google offered WebM and most everyone else settled on continuing to use H.264. Unless the Daala people can get some good lobbying, or can make their codec in such a way that minimal effort is required by hardware vendors to support both H.265 and Daala, then I think it's likely to end the same way as previous open video codec attempts.
- Dylan16807 14y agoA) Video on disc, with hardware a decade more advanced, is not the same use case any more. B) I agree that WebM fits with the XKCD comic because it's doing the same thing as an existing widespread standard but 'ours is more awesome trust us'. But changing hardware and math make a legitimate need for the creation of 265/daala. If 265 was already finished and implemented you could make the XKCD argument, but it's not.
- tommyka 14y agoI don't see how they should have a better chance of adoption than Web.M which I havnt seen much support and usage since Google announced it.
- pnathan 14y agoIMO, Mozilla has a better trust with people than Google.
- mtgx 14y agoI think it can beat h.264 in the same way h.264 beat DivX and others. We will soon have to switch to a new, more efficient codec anyway, and why not switch to a free and open source one like Daala or VP9, especially if they end up better than h.265? h.264 won over VP8 because VP8 came too late, and it was unfinished and not even that good when Google bought it. By the time Google made it on par with h.264, h.264 already won the format war. But with a new format war coming up soon, and with other codecs like Daala and VP9 being ready to compete (hopefully) from day one, h.265 might not be the default option as h.264 was. Also I think all new chips that came out this year had hardware acceleration support for VP8, and most media players support it now, too, as well as WebRTC. So even if h.265 wins again, hardware manufacturers may also support at least one of these open source codecs in the future, alongside h.265.
- shardling 14y agoThe problem with WebM seems to be that even Google didn't embrace it. I've seen a lot of speculation that they invested in it mainly to have more leverage when negotiating with the H.264 folk.
- kevingadd 14y agoMore details: http://wiki.xiph.org/Daala http://wiki.xiph.org/Daala
- ksec 14y agoWell it is not very new as it has been discussed for a while. But at least now more people knows about it. But as the recent Reference Implementation has shown H.265 is moving ahead faster then the previous H.264 has happened. Its quality is already quite good, and many manufacturers are ready to show their Encoder and Decoder in CES. So for both Software and Hardware Encoder, they are waiting for the final draft and final validation before making them for sale. VP9 is finally shaping up to be good. VP8 was never up to x.264 encoder standard. At least VP9 is making some quality improvement. But most of the current VP9 material just reads like marketing BS. While i would love to see Daala, or a codec from Mozilla rather then Google to succeed. It is just not going to happen. Mozilla has never been known for fast moving, and the amount of engineering required for a Video Codec is just huge. Even Google couldn't pull it off. And i doubt Mozilla can. Since Daala is still in research stage and it would take at least 2 year before anything done. By the time H.265 is already well ahead.
- mtgx 14y agoI wonder why they feel the need to create a completely new open source video codec, though. Can't they just collaborate with Google on VP9? Or is their technology fundamentally different? Or maybe they think Google is controlling VP9 development more than they should. Was it a technical or political decision?
- dietrichepp 14y agoThis is basically the same way that standards like H.264 get developed. A bunch of people have good ideas, make reference implementations, and get together and choose one (or mix and match). If you skip the last step, you end up with multiple standards instead of one. Which is fine. It's less work. See: http://x264dev.multimedia.cx/archives/360 http://x264dev.multimedia.cx/archives/360 I don't think there's a problem with creating new codecs. It's not that rare to get new codecs, and there are a lot of good reasons to create new codecs. Besides codecs like H.264, Theora, and VP8, there are also ones like FFV1, Dirac, Snow, Apple ProRes, and Apple Intermediate Codec. These can fill specific needs not met by H.264 or they can be used for research. However, that's not all. VP8 was a bit of a mess when it was released. I'm sure VP9 will be better, but the VP8 spec was little more than a reference implementation. Bugs in the reference implementation therefore get baked into the spec. Xiph.org has more experience producing viable compression specs than Google, so if I were them, I wouldn't camp out under the Google brand if I were making a new video codec. If anything, we should be asking why Google didn't just collaborate with Xiph for VP8. That's not to say Xiph hasn't screwed some things up too, like the Ogg container format, but at least they make it easier on people writing competing implementations.
- shmerl 14y agoHow does it compare to VP8?