Slashdot Mirror
AV1 is Well On Its Way To Becoming a Viable Alternative To Patented Video Codecs, Mozilla Says (mozilla.org)
Here's a surprising fact: It costs money to watch video online, even on free sites like YouTube. That's because about 4 in 5 videos on the web today rely on a patented technology called the H.264 video codec. From a report: It took years for companies to put this complex, global set of legal and business agreements in place, so H.264 web video works everywhere. Now, as the industry shifts to using more efficient video codecs, those businesses are picking and choosing which next-generation technologies they will support. The fragmentation in the market is raising concerns about whether our favorite web past-time, watching videos, will continue to be accessible and affordable to all.
Over the last decade, several companies started building viable alternatives to patented video codecs. Mozilla worked on the Daala Project, Google released VP9, and Cisco created Thor for low-complexity videoconferencing. All these efforts had the same goal: to create a next-generation video compression technology that would make sharing high-quality video over the internet faster, more reliable, and less expensive. In 2015, Mozilla, Google, Cisco, and others joined with Amazon and Netflix and hardware vendors AMD, ARM, Intel, and NVIDIA to form AOMedia. As AOMedia grew, efforts to create an open video format coalesced around a new codec: AV1. AV1 is based largely on Google's VP9 code and incorporates tools and technologies from Daala, Thor, and VP10.
Mozilla loves AV1 for two reasons: AV1 is royalty-free, so anyone can use it free of charge. Software companies can use it to build video streaming into their applications. Web developers can build their own video players for their sites. The second reason we love AV1 is that it delivers better compression technology than even high-efficiency codecs -- about 30% better, according to a Moscow State University study.
Over the last decade, several companies started building viable alternatives to patented video codecs. Mozilla worked on the Daala Project, Google released VP9, and Cisco created Thor for low-complexity videoconferencing. All these efforts had the same goal: to create a next-generation video compression technology that would make sharing high-quality video over the internet faster, more reliable, and less expensive. In 2015, Mozilla, Google, Cisco, and others joined with Amazon and Netflix and hardware vendors AMD, ARM, Intel, and NVIDIA to form AOMedia. As AOMedia grew, efforts to create an open video format coalesced around a new codec: AV1. AV1 is based largely on Google's VP9 code and incorporates tools and technologies from Daala, Thor, and VP10.
Mozilla loves AV1 for two reasons: AV1 is royalty-free, so anyone can use it free of charge. Software companies can use it to build video streaming into their applications. Web developers can build their own video players for their sites. The second reason we love AV1 is that it delivers better compression technology than even high-efficiency codecs -- about 30% better, according to a Moscow State University study.
Forgotten? No. VP8 lead to VP9, which is used by YouTube and Netflix. The work on what was to be VP10 was merged into AV1, which also includes contributions from Cisco's Thor and Mozilla's Daala.
VP8 hasn't been forgotten so much as its development has been continued.
What is of more concern to me is how carefully AV1 has been constructed in terms of its coding tools to avoid patent trolling and patent submarining
I don't think you need to worry. When Google announced VP8, MPEG-LA publicly announced that they were setting up a patent pool for it; they encouraged all the patent holders who VP8 infringed to step forward and add their patents to the pool.
Nobody ever came up with anything, and after over a year, MPEG-LA accepted a small amount of money from Google in exchange for a promise to never sue over VP8. No patents, no royalties, just a one-time payment; that was pretty much unconditional victory for Google and VP8. The news coverage called this a "licensing agreement" but it was more like "here, take a small amount of money and go away forever."
https://techcrunch.com/2013/03/07/google-and-mpeg-la-sign-licensing-agreement-covering-googles-vp8-video-codec-clearing-the-way-for-wider-adoption/
When VP8 was first announced, many self-appointed experts here on Slashdot declared confidently that it just had to infringe on H.264 patents, as a reading of the standard revealed numerous similarities. I am not a patent expert but I was pretty sure they were mistaken about this... Google spent something like a year after they licensed the technology before they released the open-source VP8, and I assumed that they had paid patent lawyers to go over the standard and make sure it didn't infringe on anything. Also, it looked to me like the original developers of the code had deliberately studied the existing patents and implemented something just different enough not to infringe.
It may be possible that a patent could pop up from seemingly nowhere, some weird patent nobody was paying attention to, and AV1 would be found to infringe upon it. If this scenario is possible for AV1, what makes it impossible for H.265? In fact, I'd argue it might be more likely for H.265, which is a complicated thing to which many companies tried to contribute (so they could get a share of royalties). I would be interested to hear an expert's opinion on whether AV1 is less complex than H.265... I bet that it is. And more complexity would suggest greater danger from overlooked patents.
As for submarine patents, again I am not very worried. The USA changed its patent laws between 1995 through 2000 to prevent abuses like submarine patents. Patents are 20 years from the date of filing, so playing games with paperwork extensions can't keep a patent alive forever anymore; and since 2000 patent filings are public, so the secrecy needed for submarine patents is gone.
So unless someone has a suitable patent application, filed before the year 1995, that they have kept alive with paperwork wizardry in the patent office, and nobody knows about it, and they get it granted... unless all of that is true, it shouldn't be possible for a submarine patent to torpedo AV1.
lf(1): it's like ls(1) but sorts filenames by extension, tersely
and since 2000 patent filings are public, so the secrecy needed for submarine patents is gone.
Since 2000 the default has been to publish a patent application around the 18-month mark, but under 35 U.S.C. Section 122(b)(2)(B)(i) the inventor can override that by filing a request for nonpublication, certifying that they're not seeking foreign patents.on that same invention. So inventors can still submarine if they're willing to limit themselves to U.S. patents.
Mozilla employs people from Xiph such as Chris Montgomery, Timothy Terriberry, Jean-Marc Valin, and Thomas Daede. I don't think paying the bills is laughable. Mozilla has funded development of Opus, Daala, and AV1.
If it helps, here's a recent blog post from Chris Montgomery on AV1's contstrainted directional enhancement filter.
So, I point out that ideas were thrown around and nothing become a standard, and it is not generally accepted and implemented throughout the VP codec family, and you repeat your link with the one method of encoding via the one codec that implements the hack and you think *I* have my head up my arse?
Perhaps you should actually read the link your own link points to, from 2012, then you may realize that this hack involved using a supplemental frame of content for Alpha, by substituting it for Y and zeroing U,V, and encoding in the same lossy way as the primary frame (and only in yuva420). I really hope you dont want any well defined alpha edges in your lower thirds..
Here is a good reference for how 'clean' this solution is:
https://lists.ffmpeg.org/pipermail/ffmpeg-user/2014-January/019283.html
I see in 2017 this VP8 hack was pushed in to an adobe export path, the discussion is enlightening as to how nicely official this has always been:
https://github.com/fnordware/AdobeWebM/issues/17
ffmpeg forced you to sample in yuva420p, son you are in subsampled hell, perhaps ok for final pass for the web, not useful for production editing.
and again, it uses the double-frame hack, without specific rate control for alpha - which is just broken. it took them years to even bother to support
the same hack in vp9..
AV1 is seems is playing a similar and foolish/broken game, trying to treat alpha channel as a separate image.
This inevitably leads to artifacts, and incompatible codec implementations.
To be clear, what I was is thought out, first class alpha support, in the official spec, that doesnt just exist as a hack though up in 2012 on a whim and half
implemented in some of the codecs.
Does that not make sense?
AVI supports alpha to about the same level of quality - actually slightly better, as RGBA is supported, it is just not standardised as to how the specify the A
doing in and out, so different applications handle it or not, randomly.
That is why people also dont use AVI for alpha content - quicktime is usually used, unless something else is forced, because it just works (tm) for codecs that claim they support it.
So, perhaps instead of being an insulting douche, you could perhaps consider that *proper* alpha support would be a good thing.