Alliance for Open Media codecs

[easyfab]

And after bitstream freeze I dream of others encoders like ffav1 ( ffmpeg team ) or xav1 ( like x264 ) with multi-threading and a lot of asm optimizations to enable better competition .

[bstrobl]

I am hoping that they will not rush out a bitstream with flaws in it. If it takes a bit longer than 2017 I guess that would be fine, since these codecs have much longer lifespans than companies like Google would like (remember the whole "every year a new codec" thing).

MP3 is still being used and progress in codecs is slowing gradually. Only so much Information can be crammed into a small file size. Getting this codec right the first time would be great.

[leandro]

I really hope they freeze bitstream this year too.

[nakTT]

Originally Posted by bstrobl

I am hoping that they will not rush out a bitstream with flaws in it. If it takes a bit longer than 2017 I guess that would be fine, since these codecs have much longer lifespans than companies like Google would like (remember the whole "every year a new codec" thing).

MP3 is still being used and progress in codecs is slowing gradually. Only so much Information can be crammed into a small file size. Getting this codec right the first time would be great.

Exactly. While i do hoping for it to be this year but in the interest of getting it right the first time, I would rather see them taking time to sort thing out for the better.

[mzso]

Originally Posted by bstrobl

If it takes a bit longer than 2017 I guess that would be fine, since these codecs have much longer lifespans than companies like Google would like (remember the whole "every year a new codec" thing).

Another reason I don't understand why they're still handicapping it with historic baggage like chroma subsampling and YUV. (At least interlacing seems to be gone for good.)
(Are they still doing the limited range nonsense too?)

[Quikee]

Originally Posted by mzso

Another reason I don't understand why they're still handicapping it with historic baggage like chroma subsampling and YUV. (At least interlacing seems to be gone for good.)
(Are they still doing the limited range nonsense too?)

What's the alternative? Chroma subsampling is a good way to shave of quite a bit of image data without most people noticing it or caring about it.

[mzso]

Originally Posted by Quikee

What's the alternative? Chroma subsampling is a good way to shave of quite a bit of image data without most people noticing it or caring about it.

It's not a good way, it's a very bad, primitive way.

The alternative: no subsampling. Let the encoder reduce color information in a more evolved way. An immediately obvious possibility is that you could change how color/luminosity is stored adaptively.

eg: when there's only the sky on the picture why dump half the color information for the blue range of the spectrum even before it reaches the encoder? Why "keep" anything for the red range? (The only color is blue, or white/grey aka no-color)

[nevcairiel]

Can we not repeat this discussion every couple months? You can't force content producers on a technical level, they'll just not use the codec then. Convince them that they want to use 4:4:4 or RGB, not that they have to.

Additionally, a lot of content already exists that is already subsampled. Being forced to upsample that before encoding isn't going to do anyone any good, and as such a codec that can't encode a vast amount of existing content is not going to see adoption. Its as simple as that.

[Clare]

Hey guys,

I've updated my image comparison website, now you can compare AV1 from oct 2016 to AV1 from now.

You can see the changes in metrics here: http://wyohknott.github.io/image-for...omparison.html

[wiak]

Originally Posted by Clare

Hey guys,

I've updated my image comparison website, now you can compare AV1 from oct 2016 to AV1 from now.

You can see the changes in metrics here: http://wyohknott.github.io/image-for...omparison.html

can you add VP9 and HEVC in that too?

[Quikee]

Originally Posted by wiak

can you add VP9 and HEVC in that too?

BPG uses x265 so there is already HEVC representation

[Clare]

Originally Posted by wiak

can you add VP9 and HEVC in that too?

I've updated all the codecs and added VP9. I also fixed some erroneous numbers. VP9 is doing pretty good, but I don't think we will see anyone making it an image codec. The guys doing WebP can't change codec because it would modify their bitstream and all efforts to be integrated in browsers would go to waste.

BPG is based on HEVC, although it's an older revision, I've tried a more recent git revision from x265 and there's no real gain.

Nice blog btw.

[IgorC]

Clare,

Interesting comparison. Thank You.

AV1 gets better. Ringing has disappeared. There are less artifacts and more details.

[wiak]

did anyone post these?



https://parisvideotech.com/wp-conten...ch-meet-up.pdf (English Slides)
https://www.youtube.com/watch?v=PVUIBbb-WcE (French
https://www.youtube.com/watch?v=t0oVlYcDAxM (French)

[Zebulon84]

In the videos, the AV1 developer says he estimated the availability of AV1 hardware chips around 18 month after codec freeze (so*mid 2019 if codec is realy frozen end of this year).

[easyfab]

@Zebulon84

18 months is for the very first products if I understand correctly.
And not all HW devices are planned to have it in the next generation, Apple now support HEVC, Qualcomm is not ( yet?) in the alliance...
if AV1 succeed against HEVC, It will take a long time for the adoption
Like Jean-Yves Avenard from Mozzila said : H264 will still be used for a very long time.

I think like for VP9 the first usage will be for youtube on desktop, perhaps Netflix.
but I hope the decoding will make progress, for now it's 3 times slower than VP9 and they target 2 times slower in the 6 months after bitstream freezing.
What CPU will be needed to decode 4K ?

[bstrobl]

IETF 99: https://www.youtube.com/watch?v=LJzvkOvPF30

[Clare]

The coding complexity is going through the roof. It would be nice they keep it reasonnable otherwise only big companies like Netflix or Google will be able to encode on their server. I'm running an encode and the eta is 15-20 days.

[wiak]

Originally Posted by Clare

The coding complexity is going through the roof. It would be nice they keep it reasonnable otherwise only big companies like Netflix or Google will be able to encode on their server. I'm running an encode and the eta is 15-20 days.

well how many cores? it slow because they
1. not started optimzing for SIMD yet
2. no multi-threading yet

[mzso]

Originally Posted by wiak

well how many cores? it slow because they
1. not started optimzing for SIMD yet
2. no multi-threading yet

Also there's a bunch of stuff they're experimenting with which they might take out or change.