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 > VP9 and AV1

Reply
 
Thread Tools Search this Thread Display Modes
Old 12th April 2018, 03:45   #621  |  Link
foxyshadis
ангел смерти
 
foxyshadis's Avatar
 
Join Date: Nov 2004
Location: Lost
Posts: 9,558
Quote:
Originally Posted by Shevach View Post
AV1 supports 1/8-pel motion precision (optional, specified by frame header parameter - allow_high_precision_mv). i wonder what's a gain in coding efficiency to exploit high MV precision? In HEVC and AVC the precision is 1/4-pel (for luma). In case of high-frequency video (HFR) or even for 60 fps 1/8-pel probably is redundant. Perhaps, 1/8-pel MV precision is beneficial for 4K resolution with 30 fps rate?
Unlike to HEVC development (where all was public and all discussions/contributions were located at jct-vc repository), some AV1's decisions are non-graspable to me. Who knows - under what circumstances 1/8-pel motion vector precision is beneficial vs. 1/4 pel one?
Doesn't seem to have ever been discussed in public. I bet it was a test an engineer inserted and it was just never removed; I wonder if it's ever been used in real life.
foxyshadis is offline   Reply With Quote
Old 12th April 2018, 06:25   #622  |  Link
nevcairiel
Registered Developer
 
Join Date: Mar 2010
Location: Hamburg/Germany
Posts: 10,344
VP9 already had 1/8 pel MV precision, it was probably just inherited from there.
__________________
LAV Filters - open source ffmpeg based media splitter and decoders
nevcairiel is offline   Reply With Quote
Old 12th April 2018, 21:23   #623  |  Link
colinhunt
Registered User
 
Join Date: Dec 2002
Posts: 1,022
Harmonic showed a comparison demo at NAB, pitting AV1, AVC, HEVC and JVET against each other. At the equivalent bitrate of 1.9 Mbps they got the following results:

Codec -- PSNR -- VMAF

AVC -- 32.7 -- 58
HEVC -- 36.7 -- 80
AV1 -- 37.2 -- 83
JVET -- 38.5 -- 88

That's all the info I got.
colinhunt is offline   Reply With Quote
Old 12th April 2018, 22:52   #624  |  Link
LigH
German doom9/Gleitz SuMo
 
LigH's Avatar
 
Join Date: Oct 2001
Location: Germany, rural Altmark
Posts: 6,753
They still use PSNR? Don't we prefer SSIM/dB today? Well, VMAF is more like that.

What does it prove? ... More computational complexity is the solution? We need more supercomputers.

This statement might contain sarcasm.
__________________

New German Gleitz board
MediaFire: x264 | x265 | VPx | AOM | Xvid

Last edited by LigH; 12th April 2018 at 22:55.
LigH is offline   Reply With Quote
Old 14th April 2018, 12:39   #625  |  Link
iwod
Registered User
 
Join Date: Apr 2002
Posts: 756
Quote:
Originally Posted by colinhunt View Post
Harmonic showed a comparison demo at NAB, pitting AV1, AVC, HEVC and JVET against each other. At the equivalent bitrate of 1.9 Mbps they got the following results:

Codec -- PSNR -- VMAF

AVC -- 32.7 -- 58
HEVC -- 36.7 -- 80
AV1 -- 37.2 -- 83
JVET -- 38.5 -- 88

That's all the info I got.
VMAF 80 vs 83......

Um.....
iwod is offline   Reply With Quote
Old 14th April 2018, 13:14   #626  |  Link
LigH
German doom9/Gleitz SuMo
 
LigH's Avatar
 
Join Date: Oct 2001
Location: Germany, rural Altmark
Posts: 6,753
For this one selected sample, maybe? Many show demos pick the one comparison with the results making the own preferred product look best. I do not even see any link to this show here (maybe colinhunt was there in person?).
__________________

New German Gleitz board
MediaFire: x264 | x265 | VPx | AOM | Xvid
LigH is offline   Reply With Quote
Old 15th April 2018, 19:35   #627  |  Link
colinhunt
Registered User
 
Join Date: Dec 2002
Posts: 1,022
^ No, I wasn't, just happened to bump into a short video of the demo (and the scores) on Twitter.
colinhunt is offline   Reply With Quote
Old 16th April 2018, 08:48   #628  |  Link
enctac
Registered User
 
Join Date: May 2017
Posts: 8
AV1 vs x265 / libvpx-vp9 / x264 / QSV(H.264,LA-ICQ)

Clip: Animation, 1920x1080, 335frames(13.972sec)
Metric: VMAF, SSIM


Last edited by enctac; 16th April 2018 at 12:05. Reason: upload image to imgur
enctac is offline   Reply With Quote
Old 16th April 2018, 10:46   #629  |  Link
mzso
Registered User
 
Join Date: Oct 2009
Posts: 930
Quote:
Originally Posted by enctac View Post
AV1 vs x265 / libvpx-vp9 / x264 / QSV(H.264,LA-ICQ)

Clip: Animation, 1920x1080, 335frames(13.972sec)
Metric: VMAF, SSIM

Attachment 16311
The attachment will never get approved, you need to upload it somewhere else.
mzso is offline   Reply With Quote
Old 16th April 2018, 12:13   #630  |  Link
iwod
Registered User
 
Join Date: Apr 2002
Posts: 756
That is x265 doing surprisingly well.
iwod is offline   Reply With Quote
Old 16th April 2018, 17:14   #631  |  Link
Phanton_13
Registered User
 
Join Date: May 2002
Posts: 95
enctac for Vp9 in comparisons I recomend to disable frame-parallel as is at outdated feature and use row-mt because in my tests it don't hurt quality and inprove speed (in reality row-mt has highther metrics but the variation is so low that statistically is insignicant, it has a variance of 0.0002 for ssim scores and 0.04 for vmaf scores). Plus apart of CPU 0 is also interesting to use CPU 1 an least one time in the comparison.
Phanton_13 is offline   Reply With Quote
Old 16th April 2018, 23:29   #632  |  Link
dissory
Registered User
 
Join Date: Mar 2017
Posts: 51
Can anyone please explain this for dummies?

How good is AV1 in comparison to HEVC in terms of quality and size right now (and will these get better over time or are they pretty much done with most major improvements)?
dissory is offline   Reply With Quote
Old 16th April 2018, 23:35   #633  |  Link
LigH
German doom9/Gleitz SuMo
 
LigH's Avatar
 
Join Date: Oct 2001
Location: Germany, rural Altmark
Posts: 6,753
As usual, it is impossible to give a guaranteed ratio of efficiency between two codecs, it always depends on the material processed with it, as well as how annoying a specific person rates the loss.

The development of both HEVC (x265) and AV1 is not yet done, and AV1 is in a much earlier stage (it has not even many speed-up techniques implemented yet). But it is already considerably more efficient than x265. Unfortunately, it also requires a lot more time to encode.
__________________

New German Gleitz board
MediaFire: x264 | x265 | VPx | AOM | Xvid
LigH is offline   Reply With Quote
Old 16th April 2018, 23:39   #634  |  Link
dissory
Registered User
 
Join Date: Mar 2017
Posts: 51
Quote:
Originally Posted by LigH View Post
But it is already considerably more efficient than x265.
Any rough estimate % on how much more efficient on average?
dissory is offline   Reply With Quote
Old 17th April 2018, 09:05   #635  |  Link
LigH
German doom9/Gleitz SuMo
 
LigH's Avatar
 
Join Date: Oct 2001
Location: Germany, rural Altmark
Posts: 6,753
At its current speed, I don't have the patience to run elaborate test sequences, they would probably take months without using a "render park".

And how much difference does it mean to you whether it is 10% or 15% or 20%? Where is your personal threshold for ... which action?
__________________

New German Gleitz board
MediaFire: x264 | x265 | VPx | AOM | Xvid
LigH is offline   Reply With Quote
Old 17th April 2018, 14:23   #636  |  Link
mzso
Registered User
 
Join Date: Oct 2009
Posts: 930
What are the odds that youtube won't use AV1's increased efficiency to decrease bandwidth further instead of increasing the crummy quality?
I expect they're not good.

Last edited by mzso; 17th April 2018 at 16:21.
mzso is offline   Reply With Quote
Old 17th April 2018, 14:38   #637  |  Link
LigH
German doom9/Gleitz SuMo
 
LigH's Avatar
 
Join Date: Oct 2001
Location: Germany, rural Altmark
Posts: 6,753
Because AV1 software is not yet mature enough, neither as encoder nor as decoder. If you have a device which has a hard time playing FullHD HEVC and can't handle UHD HEVC in real time, then do not even consider trying to play AV1.

The AOM is just even in the process to release specifications. There was no time yet to speed-optimize any software implementation. And without specifications being officially finalized, no hardware producer would be a fool to already release accelerating decoder chips.
__________________

New German Gleitz board
MediaFire: x264 | x265 | VPx | AOM | Xvid
LigH is offline   Reply With Quote
Old 17th April 2018, 17:58   #638  |  Link
Blue_MiSfit
Derek Prestegard IRL
 
Blue_MiSfit's Avatar
 
Join Date: Nov 2003
Location: Los Angeles
Posts: 5,988
Clearly a very long way to go.

Development of psy is of course critical in getting good subjective quality. In the tests I've seen the encoder still favors blurring heavily which smells a lot like simple PSNR tuning

Also, hopefully AV1 avoids the pitfall of VP9 - terrible rate control.

Provided these issues are handled, the encoding speed comes up and fast software decoders plus hardware decoders come out, I'd see AV1 being quite useful on the web - specifically in Chrome and Firefox, since these will likely never get HEVC support.

As a premium OTT service operator, I struggle to see the value in it, however, since DRM standards only allow for delivery of 480p (and in some cases 720p) on Chrome and Firefox due to the insecurity of Widevine on these platforms. One of the biggest value propositions for using a new codec is being able to deliver high resolution video. If my DRM requirements prevent this, there's less incentive to spend the cost to make yet another video format.

The way I see it, we'll have to keep making H.264 basically forever, we have to make HEVC today to deliver UHD + HDR video to all the 10 foot endpoints plus some mobile devices, and that is an increasingly mature ecosystem with high quality today. VP9 is an option, but it's moot for us since everything that supports VP9 also supports HEVC, and practical HEVC encoders are better than libvpx.

What does adding AV1 get me?
Blue_MiSfit is offline   Reply With Quote
Old 17th April 2018, 18:32   #639  |  Link
iwod
Registered User
 
Join Date: Apr 2002
Posts: 756
Quote:
Originally Posted by dissory View Post
Can anyone please explain this for dummies?

How good is AV1 in comparison to HEVC in terms of quality and size right now (and will these get better over time or are they pretty much done with most major improvements)?
If you look at the graph, it is not too hard to tell, at 5Mbps or below, x265 is about the same ( or better ) as AV1 in terms of quality for this clip. ( Anime )

My Personal Opinion, judging from small amount of testing and limited information, AV1 is the same or ( 10 - 20% ) better. ( Bitrate Reduction ). That is at the expense of 500x to 1000x+ encoding time. And as said, AV1 encoder is not optimised for speed yet, expect the final to be around 6 - 10x encoding time.
iwod is offline   Reply With Quote
Old 17th April 2018, 22:19   #640  |  Link
MoSal
Registered User
 
Join Date: Jun 2013
Posts: 95
Quote:
Originally Posted by iwod View Post
If you look at the graph, it is not too hard to tell, at 5Mbps or below, x265 is about the same ( or better ) as AV1 in terms of quality for this clip. ( Anime )
Look closer (--cpu-used=2 and --cpu-used=1).
__________________
https://github.com/MoSal
MoSal 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 21:34.


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