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 9th December 2016, 18:55   #121  |  Link
dapperdan
Registered User
 
Join Date: Aug 2009
Posts: 201
Not read it yet, but that appears to be the same group that put out a very early analysis of VP9, they took a copy of the git master branch the day that the format was finalized and talked about it as if it was released code. I think it was over a year later before there was an official VP9 release.

edit: the previous work I refer to: http://iphome.hhi.de/marpe/download/...3_preprint.pdf

Last edited by dapperdan; 9th December 2016 at 19:08.
dapperdan is offline   Reply With Quote
Old 10th December 2016, 09:30   #122  |  Link
CruNcher
Registered User
 
CruNcher's Avatar
 
Join Date: Apr 2002
Location: Germany
Posts: 4,926
Not sure what's so surprising the coding tools are not that efficient yet (but in case of AV1 also not fully finalized and frozen yet) but therefore also the complexity is not as high and overall lower nothing unexpected to other studies also that the RC from the HEVC Reference is pretty damn weak compared to the reference VP9 encoder so their decission to look at the raw fixed Q performance is understandable especially seeing, that this is the area where they mostly contributed themselves to MPEG/AVC/HEVC

Though it's raw complexity which isn't the whole truth at all and H.266 will show that they know it very well themselves, we have arrived in diminishing returns space partly already no more energy to suck of the universe from
__________________
all my compares are riddles so please try to decipher them yourselves :)

It is about Time

Join the Revolution NOW before it is to Late !

http://forum.doom9.org/showthread.php?t=168004

Last edited by CruNcher; 10th December 2016 at 09:49.
CruNcher is offline   Reply With Quote
Old 10th December 2016, 15:41   #123  |  Link
easyfab
Registered User
 
Join Date: Jan 2002
Posts: 332
I just made a little test (latest git for all codec) with Kimono1_1920x1080_24.y4m @2500kbits :

x264 veryslow 3.00 Mo : ssim All:0.938488 (12.110405) psnr average:38.967018
x265 slower 2.87 Mo : ssim All:0.945886 (12.666905) psnr average:40.059902
vp9 cpu-used=2 2.82 Mo :ssim All:0.946726 (12.734856) psnr average:39.988449
and
av1 cpu-used=2 2.77 Mo : ssim All:0.947562 (12.803542) psnr average:40.220836

So AV1 is better than VP9 and x265 for this clip ( and slower ). And I don't build aomenc with experimental flag that should give even better ( and even slower) result .
easyfab is offline   Reply With Quote
Old 11th December 2016, 10:04   #124  |  Link
Clare
Registered User
 
Join Date: Apr 2016
Posts: 61
Quote:
Originally Posted by easyfab View Post
I just made a little test (latest git for all codec) with Kimono1_1920x1080_24.y4m @2500kbits :

x264 veryslow 3.00 Mo : ssim All:0.938488 (12.110405) psnr average:38.967018
x265 slower 2.87 Mo : ssim All:0.945886 (12.666905) psnr average:40.059902
vp9 cpu-used=2 2.82 Mo :ssim All:0.946726 (12.734856) psnr average:39.988449
and
av1 cpu-used=2 2.77 Mo : ssim All:0.947562 (12.803542) psnr average:40.220836

So AV1 is better than VP9 and x265 for this clip ( and slower ). And I don't build aomenc with experimental flag that should give even better ( and even slower) result .
What speed did you get for AV1? I'm trying to encode a small clip but I get less than 1 frame per minute! This is impossible to encode anything with such a slow speed.
Clare is offline   Reply With Quote
Old 11th December 2016, 10:51   #125  |  Link
easyfab
Registered User
 
Join Date: Jan 2002
Posts: 332
Quote:
Originally Posted by Clare View Post
What speed did you get for AV1? I'm trying to encode a small clip but I get less than 1 frame per minute! This is impossible to encode anything with such a slow speed.
I used -t 8 --good --cpu-used=2 --tile-columns=4 and get 2-3 fps, but under 30% cpu usage, I hope better multithreading in future.
I have not tested better settings cpu-used=1 or --best because it's too slow under 1fps
easyfab is offline   Reply With Quote
Old 15th December 2016, 19:49   #126  |  Link
Clare
Registered User
 
Join Date: Apr 2016
Posts: 61
At constant quality (measured by VMAF and PSNR-HVS on 30 short clips), I've got some impressive 33% reduction in size compared to x264, whereas with x265, I only achieved 5-10% at best. I'm gonna test it again in 10Bit and see if there is more room for improvement.

edit: I have --enable-dering experimental flag enabled to test. With PVQ enabled, computing was too slow.

Last edited by Clare; 15th December 2016 at 19:51.
Clare is offline   Reply With Quote
Old 20th December 2016, 21:28   #127  |  Link
Clare
Registered User
 
Join Date: Apr 2016
Posts: 61
Kinda disappointed by experimental settings on pictures (dering + clpf):

Clare is offline   Reply With Quote
Old 20th December 2016, 23:13   #128  |  Link
CruNcher
Registered User
 
CruNcher's Avatar
 
Join Date: Apr 2002
Location: Germany
Posts: 4,926
Are you kinda disappointed about the Graph result or does it mirror also your HVS MOS ?
Are you gonna to include the preview results in the main image compare database at the different bpp/size targets ?
__________________
all my compares are riddles so please try to decipher them yourselves :)

It is about Time

Join the Revolution NOW before it is to Late !

http://forum.doom9.org/showthread.php?t=168004

Last edited by CruNcher; 20th December 2016 at 23:18.
CruNcher is offline   Reply With Quote
Old 21st December 2016, 09:09   #129  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
@Clare Can you present a graph of results VMAF for Daala?
Jamaika is offline   Reply With Quote
Old 21st December 2016, 09:44   #130  |  Link
Clare
Registered User
 
Join Date: Apr 2016
Posts: 61
Quote:
Originally Posted by CruNcher View Post
Are you kinda disappointed about the Graph result or does it mirror also your HVS MOS ?
Are you gonna to include the preview results in the main image compare database at the different bpp/size targets ?
Visually, there is indeed less ringing around the edges, but some details are phased out.

I'm not gonna touch the comparison tools.

@Jamaika: Graphs with Daala are included in the website in my signature/
Clare is offline   Reply With Quote
Old 21st December 2016, 22:56   #131  |  Link
MoSal
Registered User
 
Join Date: Jun 2013
Posts: 95
@Clare

Don't dering and clpf have similar functionalities?

You should try dering alone if you have the time. It's supposed to be better, especially at low bitrates.
__________________
https://github.com/MoSal
MoSal is offline   Reply With Quote
Old 22nd December 2016, 09:28   #132  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
@Clare @Mosal Give some commands {apps}. What are the most beneficial for vmaf?
usage: vmaf app fmt ref dis w h
apps:
adm, ansnr, motion, vif, all

How do the best on ffmpeg?
Code:
ffmpeg.exe -loglevel verbose -i "input_yuv.mp4" -an -f yuv4mpegpipe -vf scale=1920:1080,format=yuv420p - | 
vmaf_main.exe ??? yuv420p - output.yuv 1920 1080 --output json
aomenc.exe --codec=av1 --good --threads=4 --cpu-used=4 --target-bitrate=6000 --kf-max-dist=60 --auto-alt-ref=1 --frame-boost=1 --aq-mode=0 --color-space=bt709 --verbose --pass=1 --passes=1 --output=output.webm output.yuv
Jamaika is offline   Reply With Quote
Old 22nd December 2016, 10:37   #133  |  Link
Clare
Registered User
 
Join Date: Apr 2016
Posts: 61
Quote:
Originally Posted by MoSal View Post
@Clare

Don't dering and clpf have similar functionalities?

You should try dering alone if you have the time. It's supposed to be better, especially at low bitrates.
They re supposed to work together in the final product, I've even seen a patch a while back where they reversed the order of their execution because it was giving better metrics.
Clare is offline   Reply With Quote
Old 22nd December 2016, 10:45   #134  |  Link
Clare
Registered User
 
Join Date: Apr 2016
Posts: 61
Quote:
Originally Posted by Jamaika View Post
@Clare @Mosal Give some commands {apps}. What are the most beneficial for vmaf?
usage: vmaf app fmt ref dis w h
apps:
adm, ansnr, motion, vif, all

How do the best on ffmpeg?
Code:
ffmpeg.exe -loglevel verbose -i "input_yuv.mp4" -an -f yuv4mpegpipe -vf scale=1920:1080,format=yuv420p - | 
vmaf_main.exe ??? yuv420p - output.yuv 1920 1080 --output json
aomenc.exe --codec=av1 --good --threads=4 --cpu-used=4 --target-bitrate=6000 --kf-max-dist=60 --auto-alt-ref=1 --frame-boost=1 --aq-mode=0 --color-space=bt709 --verbose --pass=1 --passes=1 --output=output.webm output.yuv
I don't know what you're doing. I don't know what is your "vmaf_main.exe". The executable to use is vmafossexec.

Just build it like that:
Code:
git clone https://github.com/Netflix/vmaf
cd vmaf
make
The executable "vmafossexec" is then available in the wrapper directory.

Usage: ./wrapper/vmafossexec <pix_fmt> <width> <height> <ref_path> <dis_path> <model_path>

The model path to use is "resource/model/nflxall_vmafv4.pkl"
Clare is offline   Reply With Quote
Old 22nd December 2016, 11:50   #135  |  Link
MoSal
Registered User
 
Join Date: Jun 2013
Posts: 95
Quote:
Originally Posted by Clare View Post
They re supposed to work together in the final product,
I don't know about that.

I would still use one of the two if the results i'm getting are too smooth.
__________________
https://github.com/MoSal
MoSal is offline   Reply With Quote
Old 9th January 2017, 13:17   #136  |  Link
leonccyiu
Registered User
 
Join Date: Feb 2015
Posts: 45
Does anyone have any idea as to how much longer it'll take for the av1 format to be frozen and how much work still needs to be done?
leonccyiu is offline   Reply With Quote
Old 9th January 2017, 14:18   #137  |  Link
Clare
Registered User
 
Join Date: Apr 2016
Posts: 61
Quote:
Originally Posted by leonccyiu View Post
Does anyone have any idea as to how much longer it'll take for the av1 format to be frozen and how much work still needs to be done?
Quote:
When will the codec ship? The group is aiming to freeze the bitstream sometime between the end of the 2016 and March 2017. Expect to see browser-based support soon thereafter, with the first hardware support within 12 months after that.
http://www.streamingmedia.com/Articl...ec-110383.aspx

Although I'm personally skeptical about these delays.
Clare is offline   Reply With Quote
Old 9th January 2017, 14:40   #138  |  Link
mzso
Registered User
 
Join Date: Oct 2009
Posts: 930
Quote:
Originally Posted by leonccyiu View Post
Does anyone have any idea as to how much longer it'll take for the av1 format to be frozen and how much work still needs to be done?
Doubt that anyone will give anything other then guesses. They don't provide any information whatsoever on their progress or state of the codec.

So only someone within could give any information. I don't think there's anyone on this forum besides jmvalin who has any sort of attachment to AV1.
mzso is offline   Reply With Quote
Old 10th January 2017, 21:34   #139  |  Link
Mr.Radar
Registered User
 
Join Date: Nov 2002
Location: Minnesota, USA
Posts: 34
Quote:
Originally Posted by leonccyiu View Post
Does anyone have any idea as to how much longer it'll take for the av1 format to be frozen and how much work still needs to be done?
The last public update on AV1 I'm aware of is from Netflix's forum on royalty-free video codecs back in October which mentioned a target of "H1 2017" for freezing the bitstream. A talk giving a new status update on AV1 is scheduled for FOSDEM in early February which should hopefully provide a bit more detail.

Last edited by Mr.Radar; 10th January 2017 at 21:37.
Mr.Radar is offline   Reply With Quote
Old 15th January 2017, 11:06   #140  |  Link
VincAlastor
Registered User
 
Join Date: Sep 2009
Location: Berlin
Posts: 173
Quote:
Originally Posted by 06_taro View Post
Vanilla nightly builds for Win64, with high bit depth enabled. Should be able to output 8 / 10 / 12 bit av1. Encoder consumes too much RAM, so no 32bit build.
http://tmod.nmm-hd.org/aom/

Built daily if there are new commits. Old builds will be removed occasionally.

Also added av1 decoding (8/10/12 bit) for LAVFilters in http://tmod.nmm-hd.org/LAVFilters/ with latest libaom, thanks to Libav patches #61173 and #61174 and VFR-maniac's patch for LAVFilters. Both were somewhat modified and placed in patches folder.

BTW, the encoder seems not stable enough that 4K encoding usually causes crashes on windows, and even 720p 10-bit encoding results in abort trap 6 on OS X. Decoding with LAVFilters seems to be fine now, but further test is still needed.


for easier testing with logs and statistics i want to use a GUI. Is there already a GUI for aomenc.exe?
VincAlastor 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 18:21.


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