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 7th March 2019, 17:57   #1521  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by lvqcl View Post
Some time ago I was able to download "Childish Gambino - Feels Like Summer" (https://www.youtube.com/watch?v=F1B9Fk_SgI0) in 1080p AV1 format; now its max. available resolution for AV1 is 720p.

Maybe 1080p AV1 is just too heavy to decode.
It would have been in many cases. Perhaps it'll come back when Chrome has a highly optimized decoder.

Also, the CPU requirements for encoding 1080p are 2.5.x that for 720p. Perhaps quality compromises were made to get publishing time to be reasonable that reduced competitiveness, or there were limits on available encoder capacity?

A 1080p H.264 encode takes >>100x less compute than a 1080p AV1. Even Google only has so much free CPU capacity in a day.

Encoder performance improvements plus better quality/speed tradeoff modes will change things dramatically. With continued strong development at the current pace, we could potentially have encoders that can provide better quality than x264 at the same bitrate and encoding time by late 2020. That encoder wouldn't even try to do 99.9% of the stuff that libaom tries to do, but real encoders don't try exhaustive searches, but use lots of heuristics and early exits.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 8th March 2019, 19:56   #1522  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Rav1e versus x264 on cel/film anime source

I ran a new test of Rav1e on the same source as in this past post last year: https://forum.doom9.org/showpost.php...&postcount=882
(the motivation was a claim that Rav1e supposedly started to beat x264 "at any bitrate" so I thought I could as well repeat my highly specific test - note that this is by no means supposed to be authoritative. Unless you care about this type of content which I do.)

I used the last Rav1e release available at the time, at quality 20 and speed 0, tune psy. There are no other tuning options, so I could not do any tweaks - not my fault, that's the developers' policy. Resulting bitrate was 14526kbps. Encoding speed was 0.005 fps on Ryzen 3 2200G (~1 core used).

My x264 commandline is kind of tuned for this content - I copied what I last used on a similar bluray (but I dropped --qpmax which probably worsens efficiency). Note that three pass was used to get closer to Rav1e's bitrate but due to the short length probably, x264 still undershot (14337 kbps). Using three pass should not give quality boost.
The encoding speed was about 0.05 fps due to extremely placebo settings and 1 thread used (same as Rav1e). You could probably get 0.3 without much if any damage to quality.

Here are images from the encode: http://imgbox.com/g/yrCWrTYtF4
In the dust storm scene that uses higher bitrate than the rest and is at the start of the 910frame clip, Rav1e does well and it seems to be very close - I am not totally convinced it is as good as x264 as I think there are some signs of kinda low-passing the noisy blocks, but I am not completely confident it is worse either, although I am inclined to say x264 is in fact better here.

The rest of the clip has Rav1e clearly deficient despite the large bitrate though. It constantly smoothes the solid/flat areas and generally can't keep the milder grain and texture (which is a flaw). So its psychovisual decisions probably still aren't ready for high quality transparent encoding. This is a general problem with any non x264/x265 encoder probably, even x265 would drop texture detail everywhere before it got aq and psyrdo.

I'm trying to upload the source and streams (not fun to reproduce the Rav1e one hah), but uloz.to seems to fail on me. One thing that I should perhaps note about source - it has its chroma temporally denoised (not luma). In case that particularly matters for rav1e rate control... it only has constant quantizer mode though, afaik.


Commandlines:
rav1e.exe --speed 0 --quantizer 20 --tune Psychovisual n:\test.y4m -o psy-q20-speed0.ivf

x264-2935-64.exe n:\etr-testLL.mkv --qcomp 0.60 --aq-mode 1 --pass 3 --bitrate 14526 --no-mbtree --min-keyint 5 --keyint 240 --b-pyramid normal --deblock 0:0 --psy-rd 1.0:0.0 --aq-strength 0.8 --preset placebo --bframes 9 --direct auto --me tesa --merange 64 --subme 11 --threads 1 --colormatrix bt709 --sar 1/1 --chromaloc 0 --input-range tv --range tv --force-cfr --fps 24000/1001 -o n:\test-etr-x264-10bit-noqpmax.mkv --output-depth 10 --stats n:\noqpm.txt

Last edited by mandarinka; 8th March 2019 at 19:59.
mandarinka is offline   Reply With Quote
Old 9th March 2019, 13:44   #1523  |  Link
Atak_Snajpera
RipBot264 author
 
Atak_Snajpera's Avatar
 
Join Date: May 2006
Location: Poland
Posts: 7,806
What is the point of using AV1 with such insanely (15Mbps) high bitrate for anime? Why not use XviD or even MPEG-2 if you have such high bitrate budget?
Atak_Snajpera is offline   Reply With Quote
Old 9th March 2019, 14:12   #1524  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Quote:
Originally Posted by Atak_Snajpera View Post
What is the point of using AV1 with such insanely (15Mbps) high bitrate for anime? Why not use XviD or even MPEG-2 if you have such high bitrate budget?
Transparent quality? I don't know what you expect from mpeg2 or xvid here but they would not give it to you. After all, the pictures clearly show that even 14 megabits isn't enough when encoder doesn't do the right decisions.
In any case, the point of it is stated in the post - it was claimed by certain people (and it was not just internet randoms) that Rav1e beats x264 at any bitrate, already. I wanted to test that claim.
(It's also a realistic case for me, but I would not use unfinished/untuned encoders normally for that, of course.)

Last edited by mandarinka; 9th March 2019 at 14:15. Reason: Sorry for the typos and edit
mandarinka is offline   Reply With Quote
Old 9th March 2019, 14:14   #1525  |  Link
Atak_Snajpera
RipBot264 author
 
Atak_Snajpera's Avatar
 
Join Date: May 2006
Location: Poland
Posts: 7,806
Quote:
Originally Posted by mandarinka View Post
Transparent quality? I don't think what you expect frommpeg2 or xvid here but they would not give it to you.
In any case, the point of it is stated in the post - it was claimed by certain people (not just internet randoms) that Rav1e beats x264 at any bitrate already. I wanted to test that claim.
(It's also a realistic case for me, but I would not use new/untuned encoders normally for that, of course.)
OK. I will have to check how bad is this codec in parkjoy at 5Mbps. I'm expecting similar disaster to that crappy SVT-HEVC encoder.
Atak_Snajpera is offline   Reply With Quote
Old 9th March 2019, 14:18   #1526  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Parkjoy might benefit from the compression strength advantage of the format, same as it usually shows in low bitrate tests. Rav1e has no AQ yet though so that will probably be a big disadvantage because parkjoy iirc benefited a lot from it?
But you would not run into this "can't have transparent dirt on a flat color area in cel anime" issue, that's for sure.
mandarinka is offline   Reply With Quote
Old 9th March 2019, 14:20   #1527  |  Link
Wolfberry
Helenium(Easter)
 
Wolfberry's Avatar
 
Join Date: Aug 2017
Location: Hsinchu, Taiwan
Posts: 99
I recommend to use the official AppVeyor builds here: https://ci.appveyor.com/project/tdaede/rav1e/history
__________________
Monochrome Anomaly
Wolfberry is offline   Reply With Quote
Old 9th March 2019, 14:41   #1528  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Quote:
Originally Posted by Wolfberry View Post
I recommend to use the official AppVeyor builds here: https://ci.appveyor.com/project/tdaede/rav1e/history
At the time I went to #aomedia to ask for these binaries (I recall I could not find the artifact button that leads to the binary there/it was not working atm... not sure now) and somebody there recommended to get the release one. So I did that.
mandarinka is offline   Reply With Quote
Old 9th March 2019, 14:50   #1529  |  Link
Wolfberry
Helenium(Easter)
 
Wolfberry's Avatar
 
Join Date: Aug 2017
Location: Hsinchu, Taiwan
Posts: 99
The pre-release is just a weekly snapshot and the latest one is already kinda old, so...
__________________
Monochrome Anomaly
Wolfberry is offline   Reply With Quote
Old 9th March 2019, 14:52   #1530  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Yeah, obviously. At the time I ran this test though, it was just 9 or 10 days old. You have to remember I used speed 0, so just getting 910 frames encoded took three or four days (PC hibernated overnight). The following week I was kind of busy which added more delay to this post. So it wouldn't really be a big difference if I used appveyor then.
mandarinka is offline   Reply With Quote
Old 9th March 2019, 15:16   #1531  |  Link
Wolfberry
Helenium(Easter)
 
Wolfberry's Avatar
 
Join Date: Aug 2017
Location: Hsinchu, Taiwan
Posts: 99
AFAIK, assembly is disabled on windows at the moment, so you are basically running on rust code.
I am recommending AppVeyor builds in case someone want to try it out now.
__________________
Monochrome Anomaly
Wolfberry is offline   Reply With Quote
Old 9th March 2019, 16:03   #1532  |  Link
Atak_Snajpera
RipBot264 author
 
Atak_Snajpera's Avatar
 
Join Date: May 2006
Location: Poland
Posts: 7,806
OH MY GOD! The encoding speed is just INSANELY SLOOOOOOOOOOOOOOOOOOOOOOOW. 90 minutes have passed and my 8C/16T managed to encode only 350 frames from just first pass.
How can you even test any codec if encoding takes so much time?

Last edited by Atak_Snajpera; 9th March 2019 at 16:15.
Atak_Snajpera is offline   Reply With Quote
Old 9th March 2019, 16:17   #1533  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Quote:
Originally Posted by Wolfberry View Post
AFAIK, assembly is disabled on windows at the moment, so you are basically running on rust code.
Are you sure? I was asking about the speed in the channel too, exactly for this reason, but didn't get this information there. Maybe it would be a good idea if the encoder displayed ASM used like x264 does.
Quote:
Originally Posted by Atak_Snajpera View Post
OH MY GOD! The encoding speed is just INSANELY SLOOOOOOOOOOOOOOOOOOOOOOOW. 90 minutes have passed and my 8C/16T managed to encode only 350 frames from just first pass.
How can you even test any codec if encoding takes so much time?
I don't think Rav1e even threads much. That's why I ran x264 at 1 thread too, to get even ground in the multithreading quality trade-offs (with the idea that Rav1e doesn't make them yet but will in the future when it threads).

Edit: Or not - looks like threading as well as some other options were added after my testing.

Last edited by mandarinka; 9th March 2019 at 17:12.
mandarinka is offline   Reply With Quote
Old 9th March 2019, 16:22   #1534  |  Link
Atak_Snajpera
RipBot264 author
 
Atak_Snajpera's Avatar
 
Join Date: May 2006
Location: Poland
Posts: 7,806
Atak_Snajpera is offline   Reply With Quote
Old 9th March 2019, 16:23   #1535  |  Link
Wolfberry
Helenium(Easter)
 
Wolfberry's Avatar
 
Join Date: Aug 2017
Location: Hsinchu, Taiwan
Posts: 99
Take a look at the source code: https://github.com/xiph/rav1e/blob/master/build.rs#L25
So yes, nasm is explicitly disabled on windows.
__________________
Monochrome Anomaly
Wolfberry is offline   Reply With Quote
Old 9th March 2019, 17:16   #1536  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Seems threads and other new options got introduced in after the build I used. Though that should not matter for quality, I used the slowest possible one before the speed levels rebalancing.

Does anybody have any explanation about what the --train-rdo flag does?
And also, what does bottom-up encoding mean?
mandarinka is offline   Reply With Quote
Old 9th March 2019, 19:52   #1537  |  Link
Atak_Snajpera
RipBot264 author
 
Atak_Snajpera's Avatar
 
Join Date: May 2006
Location: Poland
Posts: 7,806
Why rav1e ignored my 5Mbps bitrate and gave me 10Mbps ?!?


When I was watching encoded file I was quite impressed by quality but then I realized that I was comparing 10Mbps AV1 with 5Mbps x264

comparision
AV1 10Mbps vs x264 10Mbps
https://www.mediafire.com/file/t3hws...0Mbps.mkv/file
https://www.mediafire.com/file/bhyul...0Mbps.mkv/file

quick screenshots
AV1 -> http://atak-snajpera.5v.pl/images/AV..._00.03.000.png
x264 -> http://atak-snajpera.5v.pl/images/x2..._00.03.000.png

spoiler AV1 is still in (blurry) woods...

Last edited by Atak_Snajpera; 9th March 2019 at 20:26.
Atak_Snajpera is offline   Reply With Quote
Old 9th March 2019, 20:29   #1538  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Rav1e rate control is rather basic I think, they only just landed a first attempt on it. You generally want to encode AV1 (even with 1pass cqp) and then run 2-pass x264 to match, like I did

Edit: looks like it doesn't do too well on PJ either, the grassy bank is obvious but also the people are still artifacty. x264 has sharp artifacty edges on the figures while Rav1e has them blurrer which might be more pleasant, but it still artifacts a lot there. (and I wonder what's with the superstrong artifacting in the first few frames).

Last edited by mandarinka; 9th March 2019 at 20:47.
mandarinka is offline   Reply With Quote
Old 10th March 2019, 03:21   #1539  |  Link
Wolfberry
Helenium(Easter)
 
Wolfberry's Avatar
 
Join Date: Aug 2017
Location: Hsinchu, Taiwan
Posts: 99
Information about train-rdo: https://github.com/xiph/rav1e/pull/948
It is currently disabled at all speed levels since it is not usable yet.
__________________
Monochrome Anomaly
Wolfberry is offline   Reply With Quote
Old 10th March 2019, 21:20   #1540  |  Link
mandarinka
Registered User
 
mandarinka's Avatar
 
Join Date: Jan 2007
Posts: 729
Or so it is approximate (worse) RDO that uses cost estimates based on a trained model instead of actual values.
(So a speed option, not something improving quality).

I guess I could have figured that out from the name, should have put more thinking into it.
mandarinka 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 15:18.


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