x266 VVC Encoder

[chenm001]

Originally Posted by tnti

Min Chen temporarily has no time to maintain this open source project.
I asked about this in July 2020.

I have enough time since September 2020, I focus on my x266 encoder and decoder after that.

[chenm001]

Originally Posted by quietvoid

Does anyone have more info about this repo here: https://github.com/chenm001/x266
Looks like just a research implementation but curious.

Thank you interesting of my projects.

That is my research tree, the goal is a hybrid video processing system since HEVC is a little complicated and WC almost can't be implemented as a real world product.
So I think software as a framework with hardware acceleration components is a good idea.

However, my previous employer is a China company, they almost prohibit any spare time work, either related to company products or unrelated.
As a result, my project has been frozen for almost 2 years, it is unfreezing now because I am unemployed since September 2020, I can continue to contribute to my projects.

[chenm001]

Originally Posted by FranceBB

The last release says "on Oct 21, 2017" when the standard wasn't even remotely a thing and was so far away from being finalized. I don't think it's something official nor from Multicoreware...

It is not related to Multicoreware, it is my research tree...

[FranceBB]

Originally Posted by chenm001

It is not related to Multicoreware, it is my research tree...

Oh, ok, got it. I was asking 'cause I know Multicoreware was working on x266 and I would have been surprised to find the repository on GitHub out of the blue, just like that.

Originally Posted by chenm001

However, my previous employer is a China company, they almost prohibit any spare time work, either related to company products or unrelated.
As a result, my project has been frozen for almost 2 years, it is unfreezing now because I am unemployed since September 2020, I can continue to contribute to my projects.

Oh... Sorry to hear that, man...
I hope you'll find something else.
By the way, I didn't know that Chinese companies had such strict regulations... but I guess I should have expected that, given that they banned pretty much everything on the internet...
Anyway, take care, I'm sure you'll find another job.

[chenm001]

Originally Posted by FranceBB

Oh, ok, got it. I was asking 'cause I know Multicoreware was working on x266 and I would have been surprised to find the repository on GitHub out of the blue, just like that.



Oh... Sorry to hear that, man...
I hope you'll find something else.
By the way, I didn't know that Chinese companies had such strict regulations... but I guess I should have expected that, given that they banned pretty much everything on the internet...
Anyway, take care, I'm sure you'll find another job.

Thank you.
I guess most China companies want t share nothing, no idea.

btw: Based on the Multicoreware scramble for x266 registering trademarks, I maybe rename my x266 to m266 or similar.
However, the project name is not important, I will still focus on make a really world high performance encoder and decoder products.

[tnti]

Originally Posted by chenm001

Thank you.
I guess most China companies want t share nothing, no idea.

btw: Based on the Multicoreware scramble for x266 registering trademarks, I maybe rename my x266 to m266 or similar.
However, the project name is not important, I will still focus on make a really world high performance encoder and decoder products.

According to past practice, it should be named T26X.：

[soresu]

Originally Posted by ksec

Hopefully x265 and x266 being developed by the same team would speed things up a little bit.

What that means is that we could see EVC, VVC and AV2 Encoder all appearing within a year.

AV2 is in deep dev mode at the moment but nowhere near finished at this point from what I gather from the commits on the AOM experimental branch.

I'd be surprised if we saw it even in 2022, more likely 2023 at the earliest?

[soresu]

Originally Posted by birdie

To be honest I'm a lot more excited by VVC than AV1. The former looks like a codec which will be embraced by the scene and which will be viable for end users. AV1 is so computationally expensive it's suitable only for Google with their insane compute resources. SVT-AV1 is fast and ... blurry at all presets. Of course, patents and everything, only it doesn't matter for your personal archival purposes.

Also, AV1/H.265/VP9 basically offer nothing/very little vs. H.264 in terms of transparent near-lossless compression where H.264 is still the king. I wonder if H.266 could change that.

From what I gather on the AV1 discord libaom has come on in leaps and bounds in both speed and memory consumption since SVT AV1 came out guns blazing on the speed front.

As for near lossless compression - I don't think that this is an aspect of the codec standard so much as the implementation.

x264 had an insane level of focus on replacing XviD at all levels, and it did eventually do that by about 2010/11 timeframe from what I remember - I think that this focus was the real reason that x264 is still the high bitrate king of the OSS encoders.

I've seen benchmarks that show x265 isn't even the best H265/HEVC encoder out there (HW265 was it?), so it seems there is definitely room for improvement.

II think that the rav1e people are focused on being a true open source x264 successor in terms of quality in their endgame, although it will probably take until well past the release of AV2 to reach that goal.

[benwaggoner]

"Best" is a very subjective word! For streaming, "transparent near-lossless compression" is a nice aspiration, but the focus is on "best quality within available bandwidth." Near lossless is more for content usable as source for additional reencoding, like ProRes, as a term of art.

The Beamr HEVC encoder is general thought of the highest quality option these days. Not sure what HW265 is.

Does rav1e have a team focused on psychovisual optimization algorithms? I wasn't sure what's happening with them after all the Mozilla layoffs.

It's amazing how influential x265's "Rate Factor" has been; we see --crf equivalents or straight --crf popping up all over the place these days. With the complexity of AV1 and VVC, I want to think that there are deeper ways to take advantage of the new codec features. Actual QP is less important any many ways given the novel ways that artifacts are suppressed. VVC's ability to encode high-QP motion without any block pattern becoming apparent is a game changer.

[Losko]

I stumbled upon this:
https://github.com/fraunhoferhhi/vvenc
Pure C++, open source (but I didn't read the license yet) and cross platform: looks promising!

[benwaggoner]

Originally Posted by Losko

I stumbled upon this:
https://github.com/fraunhoferhhi/vvenc
Pure C++, open source (but I didn't read the license yet) and cross platform: looks promising!

Yeah, definitely more practical performance than the reference encoder. Although the efficiency hit from multithreading is quite high. It's probably better for making test streams than in demonstrating potential bitrate savings versus other codecs, at least so far. Fraunhofer has a history of making good quality encoders.

[PCU]

Friends, what format do you think Google will use in the future: H.266 or AV1?
Please compare: Elecard/MainConcept H.266 with x266.

[nevcairiel]

With Google, you mean YouTube? I don't think they are likely to adopt a brand new MPEG codec, they have avoided HEVC entirely and only begrudgingly offer H.264 as a fallback when VP9 isn't available.

[PCU]

Originally Posted by nevcairiel

With Google, you mean YouTube? I don't think they are likely to adopt a brand new MPEG codec, they have avoided HEVC entirely and only begrudgingly offer H.264 as a fallback when VP9 isn't available.

What about: Facebook, Twitter, Instagram?

[birdie]

Originally Posted by PCU

What about: Facebook, Twitter, Instagram?

As far as I know there are no representatives of these companies here on these forums, so unlikely anyone could give you a definitive answer.

[FranceBB]

Originally Posted by birdie

As far as I know there are no representatives of these companies here on these forums, so unlikely anyone could give you a definitive answer.

Which is a shame.
On the other hand, we know that Facebook is still stuck on H.264 and it won't allow anything over HD 30p.
So for instance I generally record simple 3840x2160 60p videos in H.265 100 Mbits with my phone. If I upload them to Facebook they will be downscaled to 1280x720 using a simple Bicubic as resizing kernel, frames will be dropped to 30p and it will be re-encoded to low bitrate H.264.
This is nowhere near what a user expects in 2021.
We've been complaining a lot about YouTube and the fact that videos are low bitrate, 8bit only and full of banding, but at least you can upload up to 8K 60p there and if you upload HLG or PQ it will be preserved and passed through to supported devices (while showing a puny LUT conversion on SDR ones). I mean, this is at least something. Facebook on the other hand is simply ridiculous, but I think they don't care as 95% of their users is using Mobile Phones on relatively poor networks and tiny displays so they don't care...

[ksec]

Facebook is pretty much Anti-MPEG, or at least certain SVP seems to be this way. Twitter doesn't really care about these sort of thing. They have never even participate in the input of any codec discussion, on or offline AFAIK. Instagram tends to follow Facebook, although both Facebook and Instagram is supporting JPEG XL rather than AVIF so at least they seems to be pragmatic about it rather than ideological.

At the end of the day I think ( or I would like to think ) it really is some market force and business decisions. With Apple now supporting VP9 via partial hardware acceleration, and I would imagine better hardware acceleration in the future ( No one has tested the new Hardware Decoder Block on A15 yet ), may be VP9 HDR will be fully hardware accelerated.

Note: Some "site" decide to label Apple as supporting AV1 hardware acceleration and gave zero fact or evidence that is the case, and somehow someone continuously edit this information to be included as fact on wiki.

Modern Internet Video Giant like Facebook and Youtube now have different set of priorities most may not realise. Bandwidth Cost is falling, and it is falling faster than we expected. With enough peering, bandwidth is quite literally free. ( Cloudflare's CEO made this point very very clear. ) It is storage and compute that are now the cost centre. Storage prices hasn't moved at all. And Compute hasn't gotten faster anywhere near the complexity of a new codec. Introducing a new codec means adding another storage cost for every video they have. For Netflix, they can afford to re-encode their whole catalog every three months just to gain on additional image quality via encoder improvement. Their priorities are very different to Facebook or Youtube. But Netflix doesn't like MPEG. Or at least their engineers doesn't like it due to patents and royalty.

Basically we might very well end up with something similar to what we currently have. H.264 and HEVC / VP9.

I would love to be wrong though since I quite like what VVC is showing so far. But the patent pool situation from VVC, MPEG-LA has yet to announce their pricing, 10 months after their initial announcement.

That is from Internet Video Companies perspective. From broadcasting, I gather the industry is quite excited especially those from Asia.

[benwaggoner]

Originally Posted by PCU

Friends, what format do you think Google will use in the future: H.266 or AV1?
Please compare: Elecard/MainConcept H.266 with x266.

Google was a founder of AOM and primary contributor of AV1, which started with the libvpx codebase and VP9 design. They've been vocally opposed to patent-bearing codecs for years now, and specifically coded out Chrome's support for passing HEVC decode to a system decoder.

Google's YouTube has always been the primary user of VP 7, 8, and 9, and now AV1, which they've used to drag devices into supporting their players to get access to new YouTube quality tiers.

I think it is very safe to assume that Google will continue to support AV1, and will work hard to make AV2 the next big codec. It would be a profound about-face for them to even allow VVC to be decodable in Chrome, let alone actually publish content using it.

[PCU]

Originally Posted by benwaggoner

Google was a founder of AOM and primary contributor of AV1, which started with the libvpx codebase and VP9 design. They've been vocally opposed to patent-bearing codecs for years now, and specifically coded out Chrome's support for passing HEVC decode to a system decoder.

Google's YouTube has always been the primary user of VP 7, 8, and 9, and now AV1, which they've used to drag devices into supporting their players to get access to new YouTube quality tiers.

I think it is very safe to assume that Google will continue to support AV1, and will work hard to make AV2 the next big codec. It would be a profound about-face for them to even allow VVC to be decodable in Chrome, let alone actually publish content using it.

Thanks.
What about Netflix & others? discs are dead these days.

[PCU]

Who do you think abandons a royalty-free codec and pays for a codec like H.266?