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 3rd May 2016, 15:03   #41  |  Link
dapperdan
Registered User
 
Join Date: Aug 2009
Posts: 201
There was a video about VP10, I think it was linked from a thread here, which probably mostly applies.

https://www.youtube.com/watch?v=gkz1ZvejmEc

I've not watched it for a while, but I'm not sure there was anything particularly novel in the Daala sense of trying new things that won't have been patented, just the usual incremental improvements.

Last edited by dapperdan; 3rd May 2016 at 15:05.
dapperdan is offline   Reply With Quote
Old 5th July 2016, 20:36   #42  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
First tests codec, which will be released in March 2017.
http://www.cnx-software.com/2016/07/...te-with-h-265/
I am surprised the wording.
The command will encode all y4m files in the directory at 200 kbps up to 500 kbps at a 50 kbps increment. Encoding only uses one core, my machine is powered by AMD FX8350 processor, and you can see encoding is currently very slow well under 0.5 fps for a CIF video (352 x 288 resolution), but that should be expected because VP9 encoding is already slow (its successor is expected to require even more processing power), and first software implementations are usually not optimized for speed, they are just meant to show the encoding works.
Jamaika is offline   Reply With Quote
Old 6th July 2016, 00:44   #43  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by Jamaika View Post
First tests codec, which will be released in March 2017.
http://www.cnx-software.com/2016/07/...te-with-h-265/
I am surprised the wording.
The command will encode all y4m files in the directory at 200 kbps up to 500 kbps at a 50 kbps increment. Encoding only uses one core, my machine is powered by AMD FX8350 processor, and you can see encoding is currently very slow well under 0.5 fps for a CIF video (352 x 288 resolution), but that should be expected because VP9 encoding is already slow (its successor is expected to require even more processing power), and first software implementations are usually not optimized for speed, they are just meant to show the encoding works.
Reference encoders are always super-slow*. The official reference implementations for H.264 and HEVC are about 100th the speed of x264 and x265 and lack lots of critical features. All the SIMD and multithreading work goes into production implementations.

* The one exception being VC-1, since that was actually done from the WMV9 decoder porting kit, and thus accidentally was full of real-world features and optimizations .
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 6th July 2016, 03:15   #44  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
I didn't know that codec is reference. Supposedly it is based on the codec VP10, even Ligh once compiled, but may test off some function. In a word, there is nothing to build a codec AV1.
Jamaika is offline   Reply With Quote
Old 6th July 2016, 07:23   #45  |  Link
easyfab
Registered User
 
Join Date: Jan 2002
Posts: 332
It's very slow because he use run_tests that use best settings ( like placebo for x264/x265 ) and 1 thread.
He should test with more realistic setting --cpu-used=1 or 2 and add threads -t 4, with that it should be really faster.
easyfab is offline   Reply With Quote
Old 6th July 2016, 18:57   #46  |  Link
easyfab
Registered User
 
Join Date: Jan 2002
Posts: 332
I made a quick test :

source : Kimono1_1920x1080_24

x264 --preset veryslow --crf 23 : file output 6.38 Mo ( bitrate 5351 kbps ) ssim 0.955018 psnr 41.056804
x265 --preset slow 2 pass : 6.28 Mo ssim 0.960803 psnr 42.029059
x265 --preset slow 2 pass : 6.42 Mo ssim 0.961065 psnr 42.070003
vp9 -cpu-used 1 2 pass : 6.29 mo ssim 0.960090 psnr 41.922800

and for av1 -cpu-used 1 2 pass : 6.43 ssim 0.961722 psnr 42.229238

so only a little better than vp9 and x265 for the moment.
easyfab is offline   Reply With Quote
Old 6th July 2016, 21:20   #47  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by easyfab View Post
I made a quick test :

source : Kimono1_1920x1080_24

x264 --preset veryslow --crf 23 : file output 6.38 Mo ( bitrate 5351 kbps ) ssim 0.955018 psnr 41.056804
x265 --preset slow 2 pass : 6.28 Mo ssim 0.960803 psnr 42.029059
x265 --preset slow 2 pass : 6.42 Mo ssim 0.961065 psnr 42.070003
vp9 -cpu-used 1 2 pass : 6.29 mo ssim 0.960090 psnr 41.922800

and for av1 -cpu-used 1 2 pass : 6.43 ssim 0.961722 psnr 42.229238

so only a little better than vp9 and x265 for the moment.
And how did they look? With such close values, PSNR and SSIM aren't going to be that informative.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 7th July 2016, 04:51   #48  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
Quote:
Originally Posted by easyfab View Post
so only a little better than vp9 and x265 for the moment.
Maybe an interesting comparison, but ...
I don't know if there are any differences between the VP10 and AV1.
Nor do I know what the results seeks by the manufacturer. I don't know ranges of SSIM and PSNR. Could use a graph.
I would have also posting links encoders, because I understand that you tested the latest versions. (Next2)
https://aomedia-review.googlesource..../status:merged
https://chromium-review.googlesource.com/#/q/vp10

PS What puzzles me? New patches codec vp10 aren't included in the codec AV1.

Last edited by Jamaika; 7th July 2016 at 05:09.
Jamaika is offline   Reply With Quote
Old 7th July 2016, 09:40   #49  |  Link
bstrobl
Registered User
 
Join Date: Jun 2016
Posts: 55
Quote:
Originally Posted by Jamaika View Post

PS What puzzles me? New patches codec vp10 aren't included in the codec AV1.
As far as I know, only the best patches that have been properly tested get merged into AV1.

AreWeCompressedYet shows minor improvements of AV1 when compared to HEVC/VP9 in areas of PSNR/PSNR-HVS/SSIM (FastSSIM seems to be more problematic). Not sure how they are going to achieve a doubling of efficiency when compared to HEVC by 2017 however.
bstrobl is offline   Reply With Quote
Old 7th July 2016, 13:42   #50  |  Link
mzso
Registered User
 
Join Date: Oct 2009
Posts: 930
Quote:
Originally Posted by Jamaika View Post
Maybe an interesting comparison, but ...
I don't know if there are any differences between the VP10 and AV1.
Nor do I know what the results seeks by the manufacturer. I don't know ranges of SSIM and PSNR. Could use a graph.
I would have also posting links encoders, because I understand that you tested the latest versions. (Next2)
https://aomedia-review.googlesource..../status:merged
https://chromium-review.googlesource.com/#/q/vp10

PS What puzzles me? New patches codec vp10 aren't included in the codec AV1.
VP10 is not AV1. AV1 is built only in part on VP10, at least it's supposed to be. How much of it is from Daala or Thor or whatever I can't say.
mzso is offline   Reply With Quote
Old 7th July 2016, 14:08   #51  |  Link
dapperdan
Registered User
 
Join Date: Aug 2009
Posts: 201
Quote:
Originally Posted by mzso View Post
VP10 is not AV1. AV1 is built only in part on VP10, at least it's supposed to be. How much of it is from Daala or Thor or whatever I can't say.
I think it's fair to say that AV1 is VP10 at the moment. As tech or patents or coding contributions come from other parties now then it may grow to be something different, but as a starting point it was and mostly still is the libvpx VP10 codebase. At what percentage of code or ideas it stops being VP10 is of course debatable, and they certainly want to be a collaborative project going forward, but yeah, I don't think it's a misleading shorthand to say AV1 is VP10.

Last edited by dapperdan; 7th July 2016 at 14:13.
dapperdan is offline   Reply With Quote
Old 7th July 2016, 14:59   #52  |  Link
Jamaika
Registered User
 
Join Date: Jul 2015
Posts: 697
Quote:
Originally Posted by dapperdan View Post
I think it's fair to say that AV1 is VP10 at the moment.
There was such a suggestion. Codec AOM replaces VP10, Daala, Thor, HEVC. Anyway what for they did to the alliance. VP10 disappears. I see that these are two paths of development, ie said Clare.
Quote:
Originally Posted by BadFrame View Post
The AOM (Alliance of Open Media) codec is based upon VP10 (which is still in heavy development as a separate branch), but they will use any useful technology which their members have access to, for example if you look at the AOM branch, there are Daala features being added:

https://chromium-review.googlesource...oject:webm/aom

I don't know if VP10 will actually materialise as a separate stable codec release or if it will be decprecated as it is the base of the AOM codec (depends on how long it will take to develop I guess).
It can and is based on the codec vp10, just don't know it or not version a few months ago.

Last edited by Jamaika; 7th July 2016 at 15:25.
Jamaika is offline   Reply With Quote
Old 8th July 2016, 08:53   #53  |  Link
Quikee
Registered User
 
Join Date: Jan 2006
Posts: 41
Quote:
Originally Posted by mzso View Post
VP10 is not AV1. AV1 is built only in part on VP10, at least it's supposed to be. How much of it is from Daala or Thor or whatever I can't say.
Some things from daala and thor is in AV1 AFAICS - mainly the deringing filter (daala) and CLP filters (thor) are and multi-symbol entropy coder from daala (which might be replaced by rANS as it is faster and has similar efficiency) PVQ is still a WIP to integrate into AV1 - which might be quite interesting..
Quikee is offline   Reply With Quote
Old 8th July 2016, 09:07   #54  |  Link
Quikee
Registered User
 
Join Date: Jan 2006
Posts: 41
Quote:
Originally Posted by bstrobl View Post
Not sure how they are going to achieve a doubling of efficiency when compared to HEVC by 2017 however.
It may possibly be that they won't double the efficiency but just be marginally worse or better. It is much more important that they release a codec "soon" so they establish a base and offer a viable alternative to HEVC.
Quikee is offline   Reply With Quote
Old 8th July 2016, 12:16   #55  |  Link
mzso
Registered User
 
Join Date: Oct 2009
Posts: 930
Quote:
Originally Posted by Quikee View Post
Some things from daala and thor is in AV1 AFAICS - mainly the deringing filter (daala) and CLP filters (thor) are and multi-symbol entropy coder from daala (which might be replaced by rANS as it is faster and has similar efficiency) PVQ is still a WIP to integrate into AV1 - which might be quite interesting..
So, no chance for lapped transformation from daala?
mzso is offline   Reply With Quote
Old 8th July 2016, 12:33   #56  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Quote:
Originally Posted by Quikee View Post
Some things from daala and thor is in AV1 AFAICS - mainly the deringing filter (daala) and CLP filters (thor) are and multi-symbol entropy coder from daala (which might be replaced by rANS as it is faster and has similar efficiency) PVQ is still a WIP to integrate into AV1 - which might be quite interesting..
I'm not sure Daala stuff is integrated in the sense that it is definitely going to land in the final format. They adapted it or are adapting it into the same codebase so that it can be evaluated at all. The tools will have to prove their worth and/or fight their way into it in the committees/politics/visions clashes Lots of stuff currently in the codebase will be dropped from the final format, likely.

If On2/Google has main say in development, they might hamper adoption of stuff from Mozilla/Xiph because NIH mentality. (Well, let's hope such pety things will be kept in check there, but people often tend to behave like that sadly.)
mandarinka is offline   Reply With Quote
Old 8th July 2016, 14:10   #57  |  Link
mzso
Registered User
 
Join Date: Oct 2009
Posts: 930
Quote:
Originally Posted by mandarinka View Post
I'm not sure Daala stuff is integrated in the sense that it is definitely going to land in the final format. They adapted it or are adapting it into the same codebase so that it can be evaluated at all. The tools will have to prove their worth and/or fight their way into it in the committees/politics/visions clashes Lots of stuff currently in the codebase will be dropped from the final format, likely.
But, how will they finalize the format by 2017 like this?
mzso is offline   Reply With Quote
Old 10th July 2016, 03:48   #58  |  Link
Quikee
Registered User
 
Join Date: Jan 2006
Posts: 41
Quote:
Originally Posted by mandarinka View Post
I'm not sure Daala stuff is integrated in the sense that it is definitely going to land in the final format. They adapted it or are adapting it into the same codebase so that it can be evaluated at all. The tools will have to prove their worth and/or fight their way into it in the committees/politics/visions clashes Lots of stuff currently in the codebase will be dropped from the final format, likely.
I never said any of it will be final in AV1 format - just that it currently is or isn't part of the AV1 codebase yet. Of course things that won't show gains (or show gains but increase complex too much) will probably be dropped.

Quote:
Originally Posted by mandarinka View Post
If On2/Google has main say in development, they might hamper adoption of stuff from Mozilla/Xiph because NIH mentality. (Well, let's hope such pety things will be kept in check there, but people often tend to behave like that sadly.)
If they want inclusive development this won't happen. If there are two tools that have similar gain they will together decide which one to keep and which to drop (that's why they have regular meetings). IPR and what hardware vendors say would be a big factor in such a decision.
Quikee is offline   Reply With Quote
Old 10th July 2016, 03:56   #59  |  Link
Quikee
Registered User
 
Join Date: Jan 2006
Posts: 41
Quote:
Originally Posted by mzso View Post
So, no chance for lapped transformation from daala?
No, lapped transform is too invasive - they would need to change almost everything, which would not make sense to select VP10 as the base.
Quikee is offline   Reply With Quote
Old 12th July 2016, 16:17   #60  |  Link
wiak
Registered User
 
Join Date: Jul 2003
Location: somewhere north
Posts: 260
if anyone is interested i have a compiled win64 version here
https://awesome.nwgat.ninja/aomedia/...f841b-win64.7z
__________________
Woah! Ninja?! http://nwgat.ninja/ (AV1 Overview)
"Not available in your region" has now been redefined as "Go Pirate, you filthy scum" Nwgat
wiak 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 21:55.


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