Slashdot Mirror
Cisco Developing Royalty Free Video Codec: Thor
An anonymous reader writes: Video codec licensing has never been great, and it's gotten even more complicated and expensive in recent years. While H.264 had a single license pool and an upper bound on yearly licensing costs, successor H.265 has two pools (so far) and no limit. Cisco has decided that this precludes the use of H.265 in open source or other free-as-in-beer software, so they've struck out on their own to create a new, royalty-free codec called Thor. They've already open-sourced the code and invited contributions.
Cisco says, "The effort is being staffed by some of the world's most foremost codec experts, including the legendary Gisle Bjøntegaard and Arild Fuldseth, both of whom have been heavy contributors to prior video codecs. We also hired patent lawyers and consultants familiar with this technology area. We created a new codec development process which would allow us to work through the long list of patents in this space, and continually evolve our codec to work around or avoid those patents."
Cisco says, "The effort is being staffed by some of the world's most foremost codec experts, including the legendary Gisle Bjøntegaard and Arild Fuldseth, both of whom have been heavy contributors to prior video codecs. We also hired patent lawyers and consultants familiar with this technology area. We created a new codec development process which would allow us to work through the long list of patents in this space, and continually evolve our codec to work around or avoid those patents."
Why couldn't they contribute to Theora since that was the entire point of it was to be a royalty free video codec.
Your hair look like poop, Bob! - Wanker.
The Daala team has also experimented with integrating some Thor's features into Daala. It's likely that the codec developed by the IETF Internet Video Codec working group will be built from the best features of Daala, Thor and any additional contributions.
I checked with a couple of compression and video processing patent pools and the computer says no.
There is no way not to infringe on pretty much any kind of video compression tech by now
http://validator.w3.org/check?uri=http%3A%2F%2Fwww.slashdot.org Errors found while checking this document as HTML5!
Doesn't it all come down to math? That is, the algorithms being used for compression is just math. How can these patents remain valid?
They shouldn't, but companies have discovered that if they state a patent as a machine that implements said algorithm, then any other machine that implements it, even if done in a completely different way, is considered to be infringing. An computers that run a program are considered machines that implement an algorithm.
Theora, based on VP3, is roughly H.263-class technology comparable to Sorenson Spark (FLV) and MPEG-4 ASP (DivX and Xvid). H.264 and VP8 are a generation ahead of it in rate/distortion performance at Internet bitrates, and Thor is intended to be a generation ahead of H.264.
Marvel shares a parent company with Pixar, which has released OpenSubdiv as free software and just announced plans to do the same with Universal Scene Description. On that basis, I can think of companies more evil than Disney, such as The Tetris Company co-founded by Alexey "free software should never have existed" Pajitnov.
You could argue everything in the world boils down to math and physics, and that ALL patents should therefore be invalid.
But honestly - if someone IS going to patent software at all, a novel *algorithm" is one of the first things that should be allowed since it's actually a specific IMPLEMENTATION of an idea. The really shitty patents are those that just patent semi-abstract ideas (which shouldn't be allowed by patent rules anyway) or obvious user interface elements (electronic TV program guides, rounded corners...)
They're working on Thor through the IETF Internet Video Codec working group and committing to royalty-free licensing for those patents. It will be difficult for Cisco to walk back from that. Many codecs make use of patents which are licensed under royalty-free terms. Baseline JPEG does, Opus does, VP8 and VP9 do.
The propriety codecs like H.264 and HEVC are developed by the same companies which manufacture almost all the world's video hardware ---
Studio production. Theatrical distribution. Home video. Industrial applications and so on. In this pond, even Cisco is a very small fish,
If Mitsubishi is not on board, if Philips, Samsung, and a half dozen or so other global giants in manufacturing are not on board, your brightly polished license-free codec is going nowhere.
Really sad that one has to work *around* a patent - so much for patents encouraging innovation, at least not the 'get-around-it' kind.
AC comments get piped to
If that works well, it is an extremely good advertisement for Cisco.
Cisco is a company that is run by MBAs and powered by crappy offshored Indian labor. These clowns are not interested in anything except their next quarterly earnings report.
So... why are they creating an open source codec? Lets see...
Maybe they want everyone everywhere to be watching streaming video all the time, and so all of the service providers will want bigger and faster cisco routers for all of the extra traffic.
...and Cisco will just be Thor losers.
...precisely until the moment that Cisco decides it isn't free any more.
Dare I say "until they drop the hammer"? /ducks
John
If they want the extra traffic, they would have staid with MPEG-2
From the summary:
We also hired patent lawyers and consultants familiar with this technology area. We created a new codec development process which would allow us to work through the long list of patents in this space, and continually evolve our codec to work around or avoid those patents.
I'm so glad that patents are doing their intended purpose of encouraging progress. Nothing fosters progress like taking a long, circuitous route instead of the straight and patently obvious one.
It all comes down to levers. Any machine made of metal or wood is a bunch of levers, arranged in some way to be useful. (Realizing that cranks and gears are simply levers that go all the way around).
If someone invents a NEW way of arranging levers to make something that does something USEFUL, they can apply for a limited-time"monopoly on that invention. If someone invents a NEW way of arranging of arranging goto statements to create a new thing that something something useful, they can similarly also apply for a patent.
Such is necessary because software is a large series of logic 'gates'. That exact same arrangement of gates can be built in C, silicon, PHP, or copper coils. A scissor is a scissor whether it's made in steel or in brass. An mpeg decoder is an mpeg decoder whether it's made in silicon or made in C - it's the same machine.
Because patents are issued by government bureaucrats, there is little incentive for the patent workers to do a good job, so we end up with bad patents, patents on "inventions" which are not new, and are not useful. That happens with "inventions" made with wood and"inventions made with C#.
The anti-patent activists have spread a misconception that "math can't be patented ", and therefore any use of math can't be patented. That's plain false. The actual text is "the laws of nature, including the laws of science and mathematics" are not patentable. In other words, you can't patent gravity, but you CAN patent an invention based on gravity, such as a new type of elevator. You can't patent magnetism, you can patent a new type of motor which uses magnetism. You can't patent multiplication, you can patent a new invention which uses multiplication to create some useful new thing (such as a better code ).
Cisco might not need to. They are in the unique position of benefiting from more video streaming.
Slashdot social media options: AIM, ICQ, Yahoo, Jabber and Mobile Text. Why no MySpace?
If it becomes popular, they'll also be in a unique position of holding whatever "defensive" patents that may come about from this effort.
If they run into a quarter that isn't performing as well, and they're sitting on this super popular technology that can't be (easily) converted to alternatives.. it might start smelling like a gold mine.
But whatever.. making an effort is better than not making an effort, and as long as the license (or patents) isn't too restrictive there's nothing stopping someone from forking the project should they decide to close it in future.
Because the mathematics is not what's being patented. The implementation of a process to do a specific task is patented.
If you find a use for these techniques outside of video and audio compression then you can do so without infringing.
You could argue everything in the world boils down to math and physics, and that ALL patents should therefore be invalid.
That's a pretty obvious red herring. Of course everything is physics (math is an abstraction we use to describe physics.. nothing in the real world is "made of" math.) Not that physics isn't patentable. You can't patent a physical law, but you can certainly patent a device you invent that makes use of said law.
As for algorithms.. yes algorithms probably should be patentable. The concept really isn't all that bad. Where things get ugly in the real world is that currently:
a) They're allowed to be extremely vague and thus often cover many many similar algorithms that aren't actually the same.
b) They have an extremely narrow interpretation of "obvious."
c) They have absurdly long lifespans. 20 years may be reasonable for real world engineering advances but software changes so fast that 20 months might be closer to a useful exclusion period. 20 years ago, "the internet" was barely a known word outside academia and nerds. We're still working around patents from an era where you could choose to solder in your own math co-processor. 20 years for a software patent is just insane, regardless of how well the system is implemented.
Of these, I would say (c) is by far the worst. (a) and (b) are definitely issues as well, but both of them would be significantly mitigated if (c) was dealt with in a reasonable manner.
I like it! *smash* another!
Slashdot: Where a thread about Thor goes from Marvel to Stargate SG-1, without ever going near norse mythology.
I am TheRaven on Soylent News
I agree, of course we don't know how open google would be to collaboration. Also, I wonder what ever happened to dirac? Is that now a lame duck project?
I agree, of course we don't know how open google would be to collaboration.
In some ways we do. Participation in the Internet Video Codec working group is open to everyone. The final codec that comes out of NetVC will be built from the best features of all contributions. Cisco has contributed Thor and Mozilla\Xiph has contributed Daala. Google is supposed to be working on VP10 at the moment so it will be good if they contribute it to NetVC. As far as I know, they haven't yet.
The difference between physics and something that makes use of the physical law is that the law doesn't have to deal with anything other than said law, whereas a REAL DEVICE has to deal with ALL the laws, including any we haven't found yet.
This means that a REAL set of gears will have to deal with friction, torque, gear locking, grit and grind, lubgrication,vibration, carriage, rust, etc, etc, etc. Whereas a PROGRAM of a set of gears could have a billion teeth and never fail.
Likewise an algorithm is not having to deal with getting good production yields, interference, heat dissipation, feedback from nearby circuits, variations in capacitance on different paths, and so on, whereas an ASIC implementing said algorithm has to deal with all that and more.
Virtual objects therefore need no inventive step to make it work, whereas a real physical object DOES. If you don't, then your product may be more failure prone, more expensive, less productive or just plain not work because you never bothered with the other things, and someone else who implements an inventive step to deal with some or all of these problems has produced A BETTER "MOUSETRAP".
Algorithm patents are patenting the IDEA of catching mice. When the patent says "Catch mice", how the hell are you supposed to "build a better mousetrap"? If it ever is meant to catch mice, you'd be infringing on the patent and the invention is not yours.
So, yes, algorithms SHOULD NEVER be patented.
ASIC and circuit designs, or even mechanical/analogue products that make USE of those algorithms SHOULD be, since their production requires invention around the limited fidelity of the virtual world the algorithm is produced in compared to the reality the ASIC/whatever is reified on.
The animated gif is a testament to the importance of compatibility and adoption for a file format. It sucks at compression and quality, and doesn’t support any sound whatsoever.
How many expert committees and standards organizations and patent wars have revolved around implementing and promoting dozens of “superior” video formats (including codecs and containers and server/client software)? Despite all that effort and conflict, the animated gif reigned supreme as THE most widely used video format of the internet until the rise of Youtube (and it was arguably still competitive for awhile afterwards, and still hasn't gone away). Because it works absolutely everywhere, since the 90s.
The pro industry can get on whatever they like, it doesn't matter as to what people will use. That'll be whatever the big streaming services go for. Youtube, Netflix, Twitch, Amazon, these are the places that matter.
Hardware makers will get on board too. If Youtube needs a certain kind of acceleration to work well, smartphone processors will get it. Doesn't matter if the media industry says it should be something different, they want to sell new phones and something that'll do that is it working well with the streaming services people want.
Also there's the simple issue that they are fighting against existing codecs that already work well. H.264 gives a "good enough" result at speeds that modern Internet connections can cope with. So if new shit costs a lot more and/or has unacceptable restrictions, people just won't use it. Connections aren't getting slower. So you'll have a hard time convincing people to switch to save bandwidth if their bandwidth keeps going up and what they have now is "good enough".
Really, Google is the only company that matters. And maybe Netflix.
H.264 blew past Google and YouTube to become the dominant codec for HD distribution.
The geek is still fixed on web distribution through the browser and viewing video on his monitor.
But the action in HD is shifting to the app and the HDTV set.
That is where cable-cutting comes in, the smart tv, the set top box and dongles like the Amazon Fire Stick and Chromecast which bypass the browser and increasingly the computer itself.
Why can't the GPU do the work? Or do video codecs require special kinds of processing that GPUs are bad at?
It's not enough to make a good CoDec, a plan to propagate its adoption should be in the plans.
https://www.youtube.com/c/BrendaEM
There is a lot of stuff going on with HEVC:
1) Ultra HD Blu-ray is about to roll-out based on HEVC
2) ATSC 3.0 new digital broadcast standard with HEVC is being finalized
3) DVB and others are considering HEVC for digital broadcast
4) UHD/4K with HEVC is being deployed by OTT like Netflix as well as direct broadcast satellite like DirecTV and wireline like BT.
The HEVC Advance patent pool unlimited content royalties that was recently announced are giving content distributors a lot of concern. In the professional content world, it is understood that enabling technology intellectual property needs to be paid, but when you are talking about unlimited percentages of "all direct & indirect revenue" from content, not only is the cost too high, but the accounting is impossible.
Meanwhile from a bit rate versus quality level, VP9 is clearly not performing as well as HEVC.
If Cisco can show that Thor can perform nearly was well as HEVC, there are a lot of content distribution companies that will take it more seriously than they would have just a few months ago because of the HEVC Advance content royalty.
However the enabling factor would be if Cisco (and other Thor implementers) will indemnify users (i.e. content distributors) from any infringement by the use of their encoders/decoders.
Are they better than the merely foremost codec experts?
That was the turning point of my life--I went from negative zero to positive zero.
Dirac low-latency ("Dirac Pro", aka SMPTE VC-2) may come back as a mezzanine compression for production video over IP (Snell showed this at NAB 2014).
But "long GOP" Dirac never provided enough quality per bit per second compared with H.264, and certainly not with HEVC. I don't think anyone at BBC R&D is actively working on it now.
A competing codec to VP8? Isn't that reinventing the wheel some?
... they hammer this out eventually.
They shouldn't, but companies have discovered that if they state a patent as a machine that implements said algorithm, then any other machine that implements it, even if done in a completely different way, is considered to be infringing. An computers that run a program are considered machines that implement an algorithm.
My suggestion is something I would call "compulsory counter licensing". Let's say we have two video codecs A and B, each with some patents, and each using some feature that is patented by the other. Then the owner of A should be able to offer B the use of A's patents in exchange of A's use of B's patent, and that offer is compulsory - B has to accept. And likewise, B can use A's patents for free by allowing A the use of B's patents in exchange. Again, compulsory.
See how that would work out in the video codec market: Suddenly, everyone who invested some serious work and got patents is free to use the results of their work.
That was supposed to say "codec", not "code".
If you invent a _useful_ _new_ way of compressing video, you can in fact patent it. It doesn't matter if you implement your method in pure silicon (an asic), vacuum tubes, C, or copper relays. If it's new and and useful, it's patentable.
You can't patent the laws of physics, such as Newton's first law, and you can't patent multiplication. You can patent things that use gravity, and you can patent things that use multiplication. BTW, gears are for doing multiplication. If you could NOT patent anything that uses multiplication, you couldn't patent anything that uses gears.
Certain parties regularly point to patent applications that purport to have invented something that isn't new, and claim "therefore software patents should be outlawed". Well, patents on things that aren't new are ALREADY not legal. They exist - the patent office isn't perfect, but the law is that such are not valid. Similarly, non-useful "inventions" which are no more than wishes or concepts such as "time travel" or "compress video" are already not allowed. You have to invent a useful new way to doing time travel, or a useful new way of doing video compression, in order to have a legally valid patent.
> no invention is necessary
An invention is a novel (new) and useful thing. Is it your position that nobody has ever done anything new and useful with computers?
No, numbers aren't new. Prime numbers aren't new. A big prime number isn't new, it's existed since before humans.
Taxis aren't new. Uber is a taxi service, run much like many taxi services before it. When I was a kid, my uncle owned a cab, and contracted with a cab company to send him customers. It was pretty much exactly Uber, 35 years ago. Marketing the same old thing by applying a hippy image isn't new. See Ben and Jerry's ice cream as an example of prior art. All Uber did was apply an old style of marketing to an old product, exactly like Birkenstock and many others. Heck, even ILLEGAL taxis aren't at all new.
> NOT the job being done or the idea of doing the job.
The idea of doing the job is generally not useful, without a good way of getting it done.
> no maths can be patentented
Sorry, they lied to you. That's not what the law says. The anti-patent activists have spread a misconception that "math can't be patented ", and therefore any use of math can't be patented. That's simply false. The actual text is "the LAWS of nature, including the LAWS of science and mathematics" are not patentable. In other words, you can't patent gravity, but you CAN patent an invention based on gravity, such as a new type of elevator. You can't patent magnetism, you can patent a new type of motor which uses magnetism. You can't patent multiplication, you can patent a new invention which uses multiplication to create some useful new thing (such as a better codec.). "Addition is commutative" is a law of mathematics. "Compress video this way" isn't a law of anything, it's a suggestion which may a good suggestion may not.
You could argue everything in the world boils down to math and physics, and that ALL patents should therefore be invalid.
That's a pretty obvious red herring.
Exactly, that was my point...
c) They have absurdly long lifespans. 20 years may be reasonable for real world engineering advances but software changes so fast that 20 months might be closer to a useful exclusion period.
Totally agree - my general argument on software patents is that they should be valid for enough time to give a company a clear advantage in the marketplace, but not enough to completely stifle competition for decades. Even 5 years is probably ok, if not ideal.
One thing you didn't mention about shorter SW patents that's also key: it most invalidates patent trolling as a business. The vast majority of trolls acquired (what most people thought were) nearly useless patents and used their teams of lawyers to analyze them and find applications that were completely unintended by the inventor. If you limit the patent to 3-5 years vs 20, they won't be able to submarine them and sue companies a a decade later.
but patents AREN'T just one specific algorithm
Arguing that algorithms should or should not be patentable vs. whether a patent contains an algorithm are two completely different things.
The problem is not with the concept of patenting specific, workable algorithms, it's that non-algorithm patents - or those that are overly vague - are being issued.
It's a fault with the poor implementation, not the theory. Much like our entire governmental process in general these days...
I am in the professional content industry, and the last thing we want to do is to use a new codec. But for feature-quality 2160p24, HEVC is a must to keep bit rates from peaking above 100 Mbps (VBR). Also there are HEVC software decoders available, or else no one would be able to watch the Netflix "4K" content in HEVC. I've even seen HEVC software decoders running on an iPhone.
Here is one analysis of HEVC HM versus H.264 x264 quality, and finds: For video compression, the performance of VP8 were competitive with x264, while, interestingly, the new HEVC technology under definition usually showed the best performance.