Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Video Encoding > New and alternative video codecs

Reply
 
Thread Tools Search this Thread Display Modes
Old 13th April 2020, 20:21   #201  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
Quote:
Originally Posted by hajj_3 View Post
https://bitmovin.com/compression-standards-vvc-2020/

A blog from 2 months ago that shows VVC is 35% more efficient in regards to PSNR than HEVC. Requires 1.7x more cpu than HEVC to decode and requires 10x more cpu than HEVC to encode.
For CTU128. This codec is only 32bit and has one thread core. Tested for 4K and 8K formats, but creators isn't currently developing speed cpu. Added my codecs don't have SIMD support.
Jamaika is offline   Reply With Quote
Old 13th April 2020, 23:44   #202  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,752
Quote:
Originally Posted by Jamaika View Post
For CTU128. This codec is only 32bit and has one thread core. Tested for 4K and 8K formats, but creators isn't currently developing speed cpu. Added my codecs don't have SIMD support.
MPEG doesn't even try to make performant reference encoders. There's much less of an emphasis on perf than in libvpx or libaom, for example.

To the extent they get SIMD and other speedups, it's really just to make testing iterations fast enough to be practical to do during development.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 14th April 2020, 06:22   #203  |  Link
ksec
Registered User
 
Join Date: Mar 2020
Posts: 117
Here is a recent paper [1] comparing VVC and HEVC along with AV1. Although it is using an ancient VTM 4.0. For the bitrate I care about ( which is sub 4Mbps ) it seems to be doing well at anywhere between -30 -40% in terms of PSNR and VMAF. These are reference encoder anyway so I think the stats are good enough for now.

They are using AOM AV1 0.1.0-9647-ga6fa0877f, I am not familiar with libaom build number? But 1.0 was released a long time ago, does anyone know what version does this represent?

Edit: The same presentation with updated results [2], the libaom used was basically initial AOM 1.0 release in 2018, the presentation included updated results with encoder from late 2019.


[1] https://arxiv.org/pdf/2003.10282.pdf

[2] https://aomedia.org/wp-content/uploa...niversity.pptx

Last edited by ksec; 14th April 2020 at 06:29.
ksec is offline   Reply With Quote
Old 14th April 2020, 20:55   #204  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,752
Quote:
Originally Posted by ksec View Post
Here is a recent paper [1] comparing VVC and HEVC along with AV1. Although it is using an ancient VTM 4.0. For the bitrate I care about ( which is sub 4Mbps ) it seems to be doing well at anywhere between -30 -40% in terms of PSNR and VMAF. These are reference encoder anyway so I think the stats are good enough for now.
Well, it's been well documented that each generation of MPEG codecs since at least H.264 shows bigger subjective quality improvements relative to PSNR and SSIM than the previous. And VMAF got baked into AV1 and libaom in some ways that appear to artifically inflate AV1 VMAF scores relative to actual subjective quality.

We've seen cases where even a 20% "advantage" in objective metric bd-curves can still wind up as a lower subjective ranking.

Bear in mind that VMAF is just machine learning trying to estimate subjective quality ratings based on some basic objective rankings per frame. But it was only trained on what it's been trained on. Since no one has ranked VVC versus AV1 artifacts in the training set, VMAF simply hasn't been exposed to those, and I am skeptical of its scores for novel codecs until there is an updated model trained on examples of the codecs in question.

We've seen VMAF provide lower scores when adaptive quantization is used in H.264 or HEVC even when subjective ratings improved.

VMAF is still the least-bad objective metric we've had, and much more useful than PSNR. And it's gotten successively better as new models have come out (which is why talking about "VMAF" without specifying a version number is dangerous, as the same video will have a different VMAF score with today's model than with older ones).

But it's far from reliable for novel codecs or psychovisual algorithms.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 14th April 2020, 21:16   #205  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,752
Quote:
Originally Posted by ksec View Post
They are using AOM AV1 0.1.0-9647-ga6fa0877f, I am not familiar with libaom build number? But 1.0 was released a long time ago, does anyone know what version does this represent?

Edit: The same presentation with updated results [2], the libaom used was basically initial AOM 1.0 release in 2018, the presentation included updated results with encoder from late 2019.


[1] https://arxiv.org/pdf/2003.10282.pdf

[2] https://aomedia.org/wp-content/uploa...niversity.pptx
This is really a quite well done academic paper, with some useful information. In particular, it compares objective and subjective ratings so we can see what the gap is like with modern implementations, in Table VII. We can see that VMAF is the best, but still short of the 0.9 correlation where things are generally thought to become more reliable. It's close enough that I can imagine a future VMAF could hit that threshold.

I note some limitations that make extrapolation to real-world implementations quite limited.
  1. It uses standard test sequences. Many encoders, particularly reference ones, have been trained and tuned on those sequences endlessly. Comparing with novel source lets us see what an encoder can do with frames it hasn't been specifically developed to encode well.
  2. The comparison is done at fixed QP and fixed GOP, so no rate control, no VBV, no adaptive quantization, no adaptive IDR placement. So it's missing a lot of important facets of both encoders and codecs. For example, a key limitation of VC-1 was the relatively high overhead that adaptive QP signalling took relative to H.264. This kind of standard testing never would have discovered that.
  3. LibAOM has a lot more performance optimization than the reference HEVC and VVC encoders. So the perf disadvantage of AV1 would be a lot worse in the abstract. But these are also doing full searches pretty much, so it's quite challenging to extrapolate how fast practical production encoders would be in terms of quality @ perf @ bitrate.
These are very typical limitations for academic research in this area, and there really aren't any obvious better ways to do these comparisons at these early stages of codec development.

But in the real world, differences between encoders are bigger than between codecs. Recall x264 beat libaom for a lot of real-world scenarios early on, and x265 can beat libaom for a lot right now, especially when encoding time is constrained. This is at least as much about the maturity of the encoder implementation than fundamental advantages/disadvantages of the bitstream syntax and features.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 25th April 2020, 09:19   #206  |  Link
hajj_3
Registered User
 
Join Date: Mar 2004
Posts: 1,120
vvc vtm v8.1 out: https://vcgit.hhi.fraunhofer.de/jvet...leases/VTM-8.1
hajj_3 is offline   Reply With Quote
Old 25th April 2020, 11:22   #207  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697

https://nofilmschool.com/three-new-codecs-are-coming

I'm afraid this will be the last version this year. There is currently no interest in developing on the European market.
Jamaika is offline   Reply With Quote
Old 25th April 2020, 12:46   #208  |  Link
hajj_3
Registered User
 
Join Date: Mar 2004
Posts: 1,120
Quote:
Originally Posted by Jamaika View Post

https://nofilmschool.com/three-new-codecs-are-coming

I'm afraid this will be the last version this year. There is currently no interest in developing on the European market.
I'm not sure what your post means? There won't be a vvc vtm 8.2 or 9.0 this year? What information do you have about no interest in developing it in europe?
hajj_3 is offline   Reply With Quote
Old 25th April 2020, 13:12   #209  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
If you analyze the VVC library, you can see how much the addition of new functions or major corrections has decreased. Now these are minor corrections for a month. This has good sides, but the project has stopped. There are hardly any European creators although the project is German. It also amazes me. Disinterest? Why do I complain? This project isn't for hobbyist only for TV corporation and I have the impression after the advertisement that it is far from being implemented.
Jamaika is offline   Reply With Quote
Old 25th April 2020, 13:32   #210  |  Link
hajj_3
Registered User
 
Join Date: Mar 2004
Posts: 1,120
Quote:
Originally Posted by Jamaika View Post
If you analyze the VVC library, you can see how much the addition of new functions or major corrections has decreased. Now these are minor corrections for a month. This has good sides, but the project has stopped. There are hardly any European creators although the project is German. It also amazes me. Disinterest? Why do I complain? This project isn't for hobbyist only for TV corporation and I have the impression after the advertisement that it is far from being implemented.
It is supposed to be ratified this year so that would explain the lack of new features. Also there has been a covid19 shutdown over the past month. You are reading way too much into that. VVC delivers a huge bandwidth saving which is especially important for tv broadcasters using DVB-T2 and for streaming.
hajj_3 is offline   Reply With Quote
Old 25th April 2020, 14:05   #211  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
Quote:
Originally Posted by hajj_3 View Post
VVC delivers a huge bandwidth saving which is especially important for tv broadcasters using DVB-T2 and for streaming.
For me the codec will need graphics card with VVC 10bit support for streaming and we can work. It will take a while.
Jamaika is offline   Reply With Quote
Old 25th April 2020, 16:55   #212  |  Link
hajj_3
Registered User
 
Join Date: Mar 2004
Posts: 1,120
Quote:
Originally Posted by Jamaika View Post
For me the codec will need graphics card with VVC 10bit support for streaming and we can work. It will take a while.
Most users of VVC will indeed be using hardware decoders, which is the same as AV1. Youtube doesn't stream in AV1 by default, you need to enable it in your settings. Phone apps don't use it by default either except for netflix at low resolutions.
hajj_3 is offline   Reply With Quote
Old 25th April 2020, 19:20   #213  |  Link
ksec
Registered User
 
Join Date: Mar 2020
Posts: 117
There were lots of draft update in between Jan and March. So it is in the very final stage of the standard. VVC was suppose to be ratified by July, with first official release of the standard in Oct 2020. Judging from the way things are handled it doesn't seems COVID will cause any delay to the standard.

MC-IF on the other hand looks a little uncertain.
ksec is offline   Reply With Quote
Old 26th April 2020, 23:25   #214  |  Link
IgorC
Registered User
 
Join Date: Apr 2004
Posts: 1,315
Quote:
Originally Posted by hajj_3 View Post
Youtube doesn't stream in AV1 by default, you need to enable it in your settings.
Youtube have started to stream AV1 by default already some time ago.
I have default settings and Youtube frequently streams in AV1 https://www.youtube.com/testtube
IgorC is offline   Reply With Quote
Old 27th April 2020, 23:38   #215  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,752
Quote:
Originally Posted by Jamaika View Post
If you analyze the VVC library, you can see how much the addition of new functions or major corrections has decreased. Now these are minor corrections for a month. This has good sides, but the project has stopped. There are hardly any European creators although the project is German. It also amazes me. Disinterest? Why do I complain? This project isn't for hobbyist only for TV corporation and I have the impression after the advertisement that it is far from being implemented.
It is absolutely true that VVC is being designed for professional use, like in TV, streaming distribution of premium content, and all the industries and scenarios that went MPEG-2 -> H.264 -> HEVC. And it's getting a lot of support from the usual MPEG stakeholders (although some of those are contributing even more heavily to EVC).

I don't know why VVC wouldn't be any less hobbyist-friendly than other MPEG codecs, although, since the patent licensing issues don't really apply to home use. There are certainly a huge number of tools that allow for deep experimentation and tweaking like we saw in x264's development for H.264.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 27th April 2020, 23:58   #216  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,752
Quote:
Originally Posted by hajj_3 View Post
Most users of VVC will indeed be using hardware decoders, which is the same as AV1.
I'd say AV1 is somewhat unique as it has a much stronger emphasis in use on web browers given Google and Mozilla are big supporters. We still don't have announced GPUs with AV1 for Win/Mac or high volume SoCs for mobile. The only HW decoders I know of are in a few high end TVs, as Google requires AV1 for 8K streaming.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 28th April 2020, 06:04   #217  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
For maniacs: jvetvvc v8.2 + little fixes
https://www.sendspace.com/file/g3uurw

https://www.multichannel.com/pr-feed...d-supports-vvc
https://www.allegrodvt.com/video-ip-...-vvc-standard/
https://arxiv.org/abs/2003.10282
https://www.ibc.org/manage/2020-crun...s/5569.article

Since VVC will require hardware implementations it will likely take at least a couple of years for initial adoption. It is also hard at this stage to predict the cost of the licence for VVC but everyone is keen to avoid a repeat of what happened with HEVC.

Last edited by Jamaika; 28th April 2020 at 06:26.
Jamaika is offline   Reply With Quote
Old 28th April 2020, 19:01   #218  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,752
Quote:
Originally Posted by Jamaika View Post
Since VVC will require hardware implementations it will likely take at least a couple of years for initial adoption. It is also hard at this stage to predict the cost of the licence for VVC but everyone is keen to avoid a repeat of what happened with HEVC.
I don't know that it'll still be 2+ years. VVC is getting lots of input from SoC and other HW companies to make decoder implementation more performant and parallelizable than AV1 was. We had HEVC HW encoders in the wild in 2014, within a year of HEVC's final standard being approved.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 29th April 2020, 13:15   #219  |  Link
utack
Registered User
 
Join Date: Apr 2018
Posts: 63
Quote:
Originally Posted by benwaggoner View Post
I don't know that it'll still be 2+ years. VVC is getting lots of input from SoC and other HW companies to make decoder implementation more performant and parallelizable than AV1 was. We had HEVC HW encoders in the wild in 2014, within a year of HEVC's final standard being approved.
But that was a time when we had actual progress in phones that cost $400, not when the $900 phones big feature was removing the headphone jack
I don't think either codec will spread as quickly as the previous ones
utack is offline   Reply With Quote
Old 30th April 2020, 04:14   #220  |  Link
FranceBB
Broadcast Encoder
 
FranceBB's Avatar
 
Join Date: Nov 2013
Location: Royal Borough of Kensington & Chelsea, UK
Posts: 2,883
Quote:
Originally Posted by hajj_3 View Post
VVC delivers a huge bandwidth saving which is especially important for tv broadcasters using DVB-T2 and for streaming.
It does, however realistically speaking we're far from seeing an implementation in the broadcast world. If we go back way before this coronavirus madness happened, there were experiments about 8K workflows. After all, other companies were doing it as well and Japan even began to air in 8K recording things like documentaries or theater acts like they did in Milan, Italy at "la scala" (which is a very important theater). Anyway, long story short, there was a lot of confusion about this and about how we were going to do it ourselves as some people were saying that the new standard was supposed to be 8K 100fps progressive 12bit HDR in PAL and 120fps progressive in NTSC but we had a lot of doubts. Besides, we didn't even know how we could deliver that and with which color curve... This all reminded me of the very early tests with 4K HLG when we didn't have it implemented in the cameras yet (heck, there weren't even cameras with it) so we had to shoot live events in Log and then apply a matrix of linear transformation in real time in the studio before Encoding and airing which was madness.
Anyway, back to 8K.
There has always been a lot of confusion about how things were going to be but we couldn't rest cause Tokyo Olympics were getting closer and Japanese were 100% sure that they were going to shoot and air it in 8K thus providing an 8K feed to broadcasters who bought the rights. So we tried to do what we could with what we had available.

The plan:

Sony UHC-8300 cameras, different video servers and two teams.


The reason for these cameras is that they provide simultaneous 8K/4K/FULL HD streams in HDR
and SDR and you can choose between S-Log3, HLG and the plain old linear BT709. There's been quite a debate about doing simultaneous streams as I didn't quite agree. My suggestion was to get a single 8K HDR stream out of octal 12G-SDI in HLG and then downscale and frame convert it live to 4K and also downscale, frame convert and divide in fields (interlace) it live to FULL HD and apply a linear transformation with a LUT made ad hoc. They didn't like it, they said that I shouldn't bring back old arguments and that we've been over this again, so the plan they wanted to go for was to have different outputs from the camera and then two different teams working on the HDR HLG 4K/8K stream and the other working on the SDR FULL HD one like we do now for 4K/FHD events 'cause the idea is to get as much as possible from both technologies and NOT to penalize SDR viewers only because of HDR workflows since (believe it or not) they still make a big fat chunk of our viewers. Now comes the funny part: how are we going to encode it? H.266 is still at a very early stage and it's really not practical to use it (heck, the standard isn't defined yet and there are no hardware encoders for it), so what do we do? Go with H.265? Well that's... Impossible...
Although H.265 is the best thing we have, you gotta keep in mind that a master 8K stream is over 48 Gbit/s and if you wanna encode it live you gotta do it fast so you don't have much complexity anyway unless you wanna add a huge delay to your customers (and many live encoders are between 1 and 4 seconds but even if you set it to the maximum, they don't let you choose anything higher than 20 seconds which is a huge amount of time). So, even with H.265 set for live encoding at its best but with a reasonable delay we couldn't really get anything good out of it lower than 70 Mbit/s and if we push the delay really up we can squeeze it to around 50 Mbit/s but that's it, really, so the question is: how the hell are we going to air it?!
Sadly everything stopped because of COVID-19 now but this is to give you an overview on how things were going before this disease... Honestly, I have no idea about what Japanese wanted to do with the Olympics, but still...
I also heard that BTSport guys were ready to offer the Champion's League final in 8K HDR10+ PQ H.265 50fps but via internet only and not on their linear channels, which is reasonable considering that with internet you can overcome pretty much everything as long as you're willing to pay for the CDN bandwidth.
Now everything is frozen and we're in this limbo waiting for things to go back to normal: Olympics have been postponed for a year and we don't even know if there will be a Champion's League final this year as it becomes less and less likely as day passes...
This is all to say: we're not ready, but I'd very much like to see H.266 begin to be implemented in real hardware encoders and TV decoders...

Last edited by FranceBB; 30th April 2020 at 04:27.
FranceBB is offline   Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 07:37.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.