Alliance for Open Media codecs

[birdie]

new macbook air has been released, it uses a 10th gen intel processor, no info about AV1 decoding or not: https://www.apple.com/uk/macbook-air/specs/

10th gen Intel CPUs don't support HW accelerated anything in regard to AV1.

[benwaggoner]

What is Qualcomms deal at this point?
Are they silently preparing another Patent Pool for AV1 like Sisvel, or are they just lazy and indecisive
AV1 decode takes a lot of transistors, which are expensive. Are consumers willing to pay an extra $5-10 for a phone with AV1?

AV1 really didn't get anywhere near the kind of tweaking and tuning needed to make for efficient HW and SoC decoders like the MPEG codecs have. I'd expect AV2 to be a lot better in this regard. But VVC will provide better compression efficiency than AV1, with less of an incremental cost increase than AV1. Even if VVC costs $1 a unit in codec licensing, it likely would still be cheaper to a SoC vendor than AV1.

Plus Qualcomm spent all those resources on VP8/9 to never see it being used meaningfully outside of YouTube, and not providing any customer benefit beyond what H.264 already delivered.

[soresu]

AV1 really didn't get anywhere near the kind of tweaking and tuning needed to make for efficient HW and SoC decoders like the MPEG codecs have. I'd expect AV2 to be a lot better in this regard. But VVC will provide better compression efficiency than AV1, with less of an incremental cost increase than AV1. Even if VVC costs $1 a unit in codec licensing, it likely would still be cheaper to a SoC vendor than AV1.

Eh? Wasn't AMD and nVidia involved in AV1 development + several mobile SoC developers?

From that alone it would seem AV1 had more input from actual hardware vendors vs a theoretical hardware/transistor cost in MPEG development schemes as I've never heard of hardware vendors being involved previously.

Of course time itself was a significant constraint on AV1 development.

[nevcairiel]

Honestly noone really expected wide-spread hardware deployment before middle/end of this year. That one SoC already has it and another doesn't seems perfectly normal to me. Adoption takes time, SoC development cycles take years. Something that is available today would've planned to include AV1 over a year ago.

[utack]

AV1 really didn't get anywhere near the kind of tweaking and tuning needed to make for efficient HW and SoC decoders like the MPEG codecs have.

People were also ranting that VP9 is even worse for hardware implementation
Lo and behold, Youtube started using it and every cheap chinese SOC managed to implement it

[Blue_MiSfit]

Still very much hoping for hardware decode.

This definitely can't be seen as anything other than a major blow towards AV1 adoption. If the latest Qualcomm SoC doesn't support it, why should anyone else care? :|

Maybe next gen? But I mean.. The bitstream spec has been out for years. I know there was an eratta that kind of reset the clock not that long ago but.... yikes.

If I was all-in on AV1 I'd be very nervous right about now. HEVC is delivering great results today, and LC-EVC and VVC both look quite promising in the near future, and hardware support is completely guaranteed for VVC.

[huhn]

just wait for nvidia this year.

and microsoft has to do there part too.

it took 19 month to get VP9 hardware decoding on maxwell the decoder was just not accessible just to show how this worked and there was no word about it too the card where just able to do that from one driver to another and the windows update 19 month after it release date...

BTW that 3 years after VP9 was released.

i don't see a reason to get nervous.
in the end it doesn't matter if youtube and/or what ever are forcing this type of content the hardware decoder will follow.

[Yups]

Intels upcoming Xe architecture presumably supports 12 Bit AV1 according to a new leak: https://videocardz.com/newz/exclusive-intel-rocket-lake-s-features-pci-express-4-0-xe-graphics

https://abload.de/img/intel-rocket-lake-s-vy5ktm.jpg


RKl-S will use the smaller GT1 from Intels Gen12LP.

[marcomsousa]

AV1 Encoding Now Available with AWS Elemental MediaConvert



https://aws.amazon.com/about-aws/whats-new/2020/03/av1-encoding-now-available-with-aws-elemental-mediaconvert/?nc1=h_ls

Pricing https://aws.amazon.com/mediaconvert/pricing/

Marco Sousa

[Blue_MiSfit]

Now THAT is some good news (Intel). Even if it's decode only.

[marcomsousa]

AOMedia Member Spotlight with Twitch’s(Amazon) Principal Research Engineer, Video, Yueshi Shen


http://aomedia.org/aomedia-member-spotlight-with-twitchs-principal-research-engineer-video-yueshi-shen/

Marco Sousa

[benwaggoner]

AOMedia Member Spotlight with Twitch’s(Amazon) Principal Research Engineer, Video, Yueshi Shen


http://aomedia.org/aomedia-member-spotlight-with-twitchs-principal-research-engineer-video-yueshi-shen/

Yueshi is an awesome, brilliant guy.

[Tadanobu]

@benwaggoner I can't help you with compiling but in case it may be useful here are the Windows builds most of us are using over at AV1 Discord : https://jeremylee.sh/bin.html

[benwaggoner]

Awesome, thank you both!

[hajj_3]

Is this an extra download or does Win10 comes with this AV1 extension decoder ?

If it is an extra download, any url ?

https://www.microsoft.com/en-us/p/av1-video-extension-beta/9mvzqvxjbq9v?activetab=pivot:overviewtab

[littleD]

i have actualized recently (not remember a day) av1 extension from store and decoding is stilll slow.

[soresu]

i have actualized recently (not remember a day) av1 extension from store and decoding is stilll slow.

SIMD optimisations for 10+ bpc content on x86 are still completely missing, you might get better performance on a recent W10 on ARM laptop due to the NEON optimisations for ARM64 added more recently.

With 8 bpc content though both the ARM64 and x86 platforms are mostly done now for SIMD optimisation (excepting AVX512), I would not expect any dramatic speedups for that now without adding more cores.

[benwaggoner]

With 8 bpc content though both the ARM64 and x86 platforms are mostly done now for SIMD optimisation (excepting AVX512), I would not expect any dramatic speedups for that now without adding more cores.
I'd be surprised to see material perf improvements by adding AVX512, at least on current gen CPUs, due to aggressive thermal throttling.

This will hopefully change over time, just like we saw AVX2 become much more useful in the second major generation of implementations.

[VincAlastor]

Unfortunately you still can't search/skip the svt-av1 stream quickly if you use the default value --irefresh-type 1 (CRA; Open GOP).

Since there is no hint if it is a svt-av1 encoder or dav1d decoder problem, I tested the latest version.

Does anyone have a hint?

[Beelzebubu]

Unfortunately you still can't search/skip the svt-av1 stream quickly if you use the default value --irefresh-type 1 (CRA; Open GOP).

Since there is no hint if it is a svt-av1 encoder or dav1d decoder problem, I tested the latest version.

Does anyone have a hint?

How did you mux the file? The problem suggests that the muxer did not mark the keyframes in the container index.

[VincAlastor]

How did you mux the file? The problem suggests that the muxer did not mark the keyframes in the container index.

i muxed with mp4box Feb 2020 and mkvmerge 45.

mkvmerge i tested --cues iframes and --cues all

nothing was working.

[Beelzebubu]

i muxed with mp4box Feb 2020 and mkvmerge 45.

mkvmerge i tested --cues iframes and --cues all

nothing was working.

You need to file a bug report (or feature request) with mp4box to add support for invisible keyfames, which is how AV1 calls non-IDR / open-GOP keyframes.

[benwaggoner]

i have actualized recently (not remember a day) av1 extension from store and decoding is stilll slow.
They have updated the extension pretty regularly, alas without release notes.

[Blue_MiSfit]

https://www.blog.google/products/duo/4-new-google-duo-features-help-you-stay-connected/

Pretty cool, Google has announced using AV1 in their Duo voice chat app, targeting 30 Kbps!

[soresu]

https://www.blog.google/products/duo/4-new-google-duo-features-help-you-stay-connected/

Pretty cool, Google has announced using AV1 in their Duo voice chat app, targeting 30 Kbps!

Wow 30 kbps, that is pretty incredible, even old 2-2.5g mobile networks and ancient DSL could field that, which is probably exactly the point.

Sounds like they are also trying to compete with that Zoom app and Skype for group video calls with the current demand for it skyrocketing.

Even with them looking to expand to 12 participant group calls it would only be 330 kbps down and 30 kbps up.

[Blue_MiSfit]

Yeah I wonder what encoder they're using (presumably libaom), and how it's configured! I imagine they're doing like 15 fps and some very small frame size like 240x430 or 160x284

[soresu]

Yeah I wonder what encoder they're using (presumably libaom), and how it's configured! I imagine they're doing like 15 fps and some very small frame size like 240x430 or 160x284

Even with a small frame size they may have a dynamic res DNN scaler to sweeten the pot given their AI/ML focus in the last decade.

[utack]

https://www.blog.google/products/duo/4-new-google-duo-features-help-you-stay-connected/

Pretty cool, Google has announced using AV1 in their Duo voice chat app, targeting 30 Kbps!

And yet their blog is using.....gif to showcase the moving components :rolleyes:

[Blue_MiSfit]

Even with a small frame size they may have a dynamic res DNN scaler to sweeten the pot given their AI/ML focus in the last decade.

That very much makes senes, though I wonder how good these can really be.... LC-EVC seems to be a much better solution than smart upscaling.

[soresu]

That very much makes senes, though I wonder how good these can really be.... LC-EVC seems to be a much better solution than smart upscaling.

If it only needs to work for a phone size screen, or at worst a 10 inch tablet then it won't be much of an issue anyways.

[benwaggoner]

Yeah I wonder what encoder they're using (presumably libaom), and how it's configured! I imagine they're doing like 15 fps and some very small frame size like 240x430 or 160x284
Video chat content is generally quite simple to encode. No camera motion, faces don't move that fast. So pixels/bit can typically be a lot higher than for TV/film content.

384x288 24 fps at 30 Kbps of video chat is feasible with a good HEVC and persumably AV1 encoder. Not pretty, but with lip sync and some sense of non-verbal communication. Lower frame rate and fps mean a lot of MIPS/pixel so AV1's broader encoding speed challenges shouldn't be a substantial issue.

[marcomsousa]

YouTube for Android TV enable AV1 video codec support on capable plaforms
https://www.xda-developers.com/youtube-for-android-tv-adopts-av1-video-codec-in-certain-devices/

[hbbs]

YouTube for Android TV enable AV1 video codec support on capable plaforms
https://www.xda-developers.com/youtube-for-android-tv-adopts-av1-video-codec-in-certain-devices/Just tested on my Nvidia Shield TV.

Only getting VP9. Also, apparently, there is no way to enable AV1 on the app settings.

Sent from my Moto Z3 Play using Tapatalk

[hajj_3]

Just tested on my Nvidia Shield TV.

Only getting VP9. Also, apparently, there is no way to enable AV1 on the app settings.

Sent from my Moto Z3 Play using Tapatalk

I assume it is just for android tv devices that have an av1 hardware decoder or maybe it is only used for low resolutions if your internet connection is slow.

[foxyshadis]

Just tested on my Nvidia Shield TV.

Only getting VP9. Also, apparently, there is no way to enable AV1 on the app settings.

Sent from my Moto Z3 Play using Tapatalk

Based on a look through the decompiled source, it does seem to require hardware support, and also requires Android 10. I don't think the Shield has either.

[Blue_MiSfit]

Shield definitely does not have a hardware AV1 decoder. It's an nVidia Tegra X1 / X1+ SOC.

[soresu]

YouTube for Android TV enable AV1 video codec support on capable plaforms
https://www.xda-developers.com/youtube-for-android-tv-adopts-av1-video-codec-in-certain-devices/

I believe this is pre emptive support for the rumoured Android TV powered Chromecast Ultra supposedly coming out later this year - if it uses Amlogic S905x4 as I believe it would, it should do 4K AV1 in hardware.

If that happens I would expect a ramp up in AV1 content on Youtube, Google Play Movies/TV, and Vimeo.

[foxyshadis]

Shield definitely does not have a hardware AV1 decoder. It's an nVidia Tegra X1 / X1+ SOC.

I would've loved to have been a fly on the wall on the meetings held over whether the AV1 block would be ready in time to meet the already delayed X1+ tape-out or not. "Just two more weeks, it's almost done!"

[Selur]

Is there an alternative to https://aomedia.googlesource.com/aom/+/master/examples/noise_model.c to build a film grain table for aomenc?
Preferably something that does only support humongous raw yuv files as input.
Having to:
1. create a raw yuv file from the source and from a denoised version of the source
2. create the film grain table with noise_model
(3. encode using the film grain table)
just seems like too much pain for this to be usable for anything but small samples.

So does anyone know a better tool or a better workflow to create the filmgrain tables? (may be a modified ffmpeg version?)

Cu Selur

[marcomsousa]

Libaom have a new branch applejack some people are saying that this is for Libaom 2.0? Anyone know anything about this?

Marco Sousa

[marcomsousa]

Libaom have a new branch applejack some people are saying that this is for Libaom 2.0? Anyone know anything about this?

Marco Sousa

OK, so this branch is new branch from master some time ago, to make it stable and release libaom 2.0.0

https://bugs.chromium.org/p/aomedia/issues/detail?id=2545

[Beelzebubu]

Is there an alternative to https://aomedia.googlesource.com/aom/+/master/examples/noise_model.c to build a film grain table for aomenc?
Preferably something that does only support humongous raw yuv files as input.
Having to:
1. create a raw yuv file from the source and from a denoised version of the source
2. create the film grain table with noise_model
(3. encode using the film grain table)
just seems like too much pain for this to be usable for anything but small samples.

So does anyone know a better tool or a better workflow to create the filmgrain tables? (may be a modified ffmpeg version?)

Cu Selur

aomenc --denoise-noise-level= does the same as a FFT denoiser + examples/noise_model + aomenc --film-grain-table=.

[Scope]

AVIF (AV1 Image File Format): experimental support
https://bugzilla.mozilla.org/show_bug.cgi?id=1625363

Firefox 77.0a1 (2020-05-03)
https://i.imgur.com/uonsh58.png
https://i.imgur.com/5awSd0t.png

[Tadanobu]

aomenc --denoise-noise-level= does the same as a FFT denoiser + examples/noise_model + aomenc --film-grain-table=.

Are you sure ? I thought --denoise-noise-level adds the same amount everywhere while the film grain table would calculate how much it adds for each frame/scene.

[Beelzebubu]

Are you sure ?

Yes.

I thought --denoise-noise-level adds the same amount everywhere while the film grain table would calculate how much it adds for each frame/scene.

That depends on the film grain table in the input argument. But generally speaking, if the film grain table was generated using examples/noise_model, then it invokes the exact same code as --denoise-noise-level= for film grain table generation.

[Selur]

Thanks for the info. :)

[mzso]

Can I get FFmpeg with all AV1 encoders?
It seems like media autobuilds is something like that, but it also looks like a worthless POS. Spams me with a hundred questions (with stuff I don't care for maybe don't even know what they are) only to fail at the first step of trying to do anything.

What I mainly want is ffmpeg with all AV1, HEVC encoders. (and maybe other non-free good codecs)

[benwaggoner]

Can I get FFmpeg with all AV1 encoders?
It seems like media autobuilds is something like that, but it also looks like a worthless POS. Spams me with a hundred questions (with stuff I don't care for maybe don't even know what they are) only to fail at the first step of trying to do anything.

What I mainly want is ffmpeg with all AV1, HEVC encoders. (and maybe other non-free good codecs)
media-autobuild-suite is the only way to get it I know of other than finding someone who's already made a binary. That said, media-autobuild-suite has something wrong that keeps it from fully building about 85% of the time....

[LigH]

MABS allows you to configure the set of linked codec libraries with a file build/ffmpeg_options.txt where all known supported options are listed, grouped and partially disabled due to known issues or your choice to disable codecs during your configuration of MABS as a whole (stored in build/media-autobuild_suite.ini).

[mzso]

media-autobuild-suite is the only way to get it I know of other than finding someone who's already made a binary. That said, media-autobuild-suite has something wrong that keeps it from fully building about 85% of the time....

I only got to:
- Download and install msys2 basic system

-------------------------------------------------------------------------------
-------------------------------------------------------------------------------

- Downloading and unpacking msys2 basic system

-------------------------------------------------------------------------------
The Win32 internal error "Egy rendszerhez csatlakoztatott eszköz nem működik" 0x1F occurred while writing to the console output buffer at
the current cursor position. Contact Microsoft Customer Support Services.

'7za' is not recognized as an internal or external command,
operable program or batch file.

Even though nothing is specified as requirement besides OS, RAM/diskspace and Powershell.

At the very fist step it actually tries to do something, after wasting me a quarter hour asking questions for nothing.