Firefox 47 Beta Enables VP9, Embedded YouTube Videos Use HTML5

[hansg]

This is also the version of Firefox where mandatory extension signing finally gets turned on. Since I'm using at least one that hasn't been signed (nor updated since 2010), this is where we say goodbye... It's been many happy years, but it will be 46.something for me, and eventually another browser.

[Gusar]

the result has been slowly increasing quality, and has reached the point where, on some data set, the codec is slightly better than x265 on some metric.

The problem is that metrics, even PSNR-HVS and FastSSIM, *do not* correlate to visual quality at all. But my hope is that, with AV1 being actually open instead of a proprietary Google format, there will be encoder implementations that focus on visual quality rather than metrics. That's the biggest issue with VP9, libvpx is slow (no real multi-threading), focused solely on metrics, and has really crappy rate control.

This is also the version of Firefox where mandatory extension signing finally gets turned on. Since I'm using at least one that hasn't been signed (nor updated since 2010), this is where we say goodbye... It's been many happy years, but it will be 46.something for me, and eventually another browser.

You should stick to 45 ESR, it'll get security fixes for a lot longer than 46. Or you make use of the fact that Firefox is open source and use a build (either your own compile or someone else's) that has mandatory signing turned off.

[anarki2]

No, adding support for a codec that is a big giant mess licensing wise does not make sense. Wait for AV1, the codec from the Alliance for Open Media, that's where the stuff is. Until then, h264 will do.

Let me summarize that big giant mess: it's free to use. How messy indeed.

[plonoma]

Maybe this makes sense: https://github.com/mozilla/shumway

No and wasn't shumway discontinued?
http://www.i-programmer.info/news/86...placement.html

Your post does not answer any of my questions.
My questions are about if there is website specific code in Firefox.
My questions also are limited to video playing, nothing about other applications such as Flash/html5 games.
The questions are about video's using a combination of html5 and flash, either with flash as fallback or html5 as fallback.
Does it work with embed links from websites other than YouTube? e.g. embedding a Vimeo link in a website.
And if there is website specific code left currently in Firefox.

Which are completely separate questions from: is there a flash replacement I can use and automatic conversion of old flash video content:
http://www.ghacks.net/2016/01/16/fir...o-html5-video/
Is the code generic, thus working for any website (embed source) and not just YouTube?

[Gusar]

Let me summarize that big giant mess: it's free to use. How messy indeed.

H.265? Are you kidding? Let me give you a real summary...

H.264:
You go to MPEG LA, pay $0.20 for each unit above 100,000 (no fees for smaller number of units) with an annual cap of $6.5 million.

H.265:
You go to MPEG LA, pay $0.20 for each unit above 100,000, with an annual cap of $25 million. Then you go to HEVC Advance, pay $0.40 - $1.20 per unit depending on what exactly it is you're shipping, with an annual cap of up to $40 million and additional content fees up to $5 million per year. Then you go to Technicolor and negoitate licensing with them, as they used to be part of HEVC Advance but then decided to go their own way. After you've done all that, you should also make a large cash reserve, because those three cover about half of all h265 patents, and the reserve should be ready for when the other patent holders decide to start collecting, either individually (like Technicolor) or as part of a yet unformed third patent pool.

[tuuker]

This is also the version of Firefox where mandatory extension signing finally gets turned on. Since I'm using at least one that hasn't been signed (nor updated since 2010), this is where we say goodbye... It's been many happy years, but it will be 46.something for me, and eventually another browser.

For this there will be Cyberfox.

[marco-c]

You should stick to 45 ESR, it'll get security fixes for a lot longer than 46. Or you make use of the fact that Firefox is open source and use a build (either your own compile or someone else's) that has mandatory signing turned off.

Another simple solution is to fork the addon and sign it yourself by submitting it to AMO (addons.mozilla.org).

This is also the version of Firefox where mandatory extension signing finally gets turned on. Since I'm using at least one that hasn't been signed (nor updated since 2010), this is where we say goodbye... It's been many happy years, but it will be 46.something for me, and eventually another browser.

What is the addon in question?

[DrYak]

The problem is that metrics, even PSNR-HVS and FastSSIM, *do not* correlate to visual quality at all. But my hope is that, with AV1 being actually open instead of a proprietary Google format, there will be encoder implementations that focus on visual quality rather than metrics.

VP9 is not proprietary. Their libvpx *does* suck, but FFmpeg have their own implementation (and apparently it's better enough to warrant a switch for Firefox from the former to the later)

Speaking of perceptual quality, Xiph have shown some quite interesting results in Daala with their perceptual vector quantisation - PVQ, namely activity masking.
All this is going in AOMedia's futur codec (together with technology comming from Cisco's Thor and Google's VP9).
And that's very promising.

But indeed ABX and similar blind tests are best tools to measure the efficience of codec.
And given the results with Opus, we can set our hopes high.

(But in the meantime, PSNR-HVS and FastSSIM are a cheap way to track the evolution of code commit after commit).

[Gusar]

VP9 is not proprietary.

I've been through this before, search my previous posts if you're really interested, but VP9 is very much a proprietary format. The existence of an open source reference does not change that. Proprietary != closed source, the word has meaning beyond just openness of the source.

Their libvpx *does* suck, but FFmpeg have their own implementation

FFmpeg only has a decoder though.

(But in the meantime, PSNR-HVS and FastSSIM are a cheap way to track the evolution of code commit after commit).

But that's useless when an encoder optimizing for those will produce really bad visual results. It's why the VPx family of codecs was never competitive when they were still owned by On2 (before Google bought them).

x264 only got good when Dark Shikari started working on it, adding all sorts of psy enhancements, like variance based adaptive quantization and psychovisual rate-distortion decisions. When VAQ was being developed, there was a huge thread at the doom9.org forums where people tested out various experimental settings. Everyone used their eyes in those tests and posted example screenshots, no one posted metrics. *That* is how you make a good encoder, not by plotting PSNR and SSIM graphs.

[DrYak]

x264 only got good when Dark Shikari started working on it, adding all sorts of psy enhancements, like variance based adaptive quantization and psychovisual rate-distortion decisions. When VAQ was being developed, there was a huge thread at the doom9.org forums where people tested out various experimental settings. Everyone used their eyes in those tests and posted example screenshots, no one posted metrics. *That* is how you make a good encoder, not by plotting PSNR and SSIM graphs.

PSNR and SSIM graphs just give you a quick overview of how much distortion you have for a given compression factor (over-simplification: how many details you lose)
It's still useful to have a vague idea how things are heading (e.g.: new commit is now compressing more while losing same amount of details) which in the early phases of development *IS* a bit useful.
Of course what matter the most is *which* details are lost. And of course that is where human organs are needed for the analysis.
(It would be bad if the lost details are in the form of big freaking blinking rainbow colored dot in the middle of the screen. Yes the distorsion is mathematically small, but it's still annoying as hell)

Of course, I fear not for the overall end result quality-wise:
Given past experience with Xiph development and IETF approved codecs - OPUS codec got ABX-ed the shit out of it on hydrogenaudio forum - I'm pretty sure that, as Daala codec matures into AOMedia, it will get suitably analyzed (double blindly, big pool, etc.).

Xiph is already starting to consider some psychovisual properties (features of the PVQ like activity masking).