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 > High Efficiency Video Coding (HEVC)

Reply
 
Thread Tools Search this Thread Display Modes
Old 7th October 2020, 08:40   #21  |  Link
H2sixty
Not a Registered User
 
H2sixty's Avatar
 
Join Date: Aug 2020
Location: earth
Posts: 24
Quote:
Originally Posted by benwaggoner View Post
circa 2016
a link? i can not find what this is.
H2sixty is offline   Reply With Quote
Old 7th October 2020, 09:01   #22  |  Link
H2sixty
Not a Registered User
 
H2sixty's Avatar
 
Join Date: Aug 2020
Location: earth
Posts: 24
Quote:
Originally Posted by benwaggoner View Post
CRF really isn't meant to be used as a grain removal tool!
Quote:
Originally Posted by benwaggoner View Post
there will also be content where real detail gets lost, and artifacts will start to become visible before grain becomes invisible.
i had issues with artifacts & detail retention at first, luckly it didn't stay an issue, toughest issue i had though...

also, as i stated before---\/
Quote:
Originally Posted by H2sixty View Post
i stopped using filters such as nlmeans degrain and nr-inter/nr-intra=x since i finished putting together the custom settings i posted...
(the presets have heavy color banding on higher crf settings, there was no way to use crf to blend with presets)

degraining helped when i started this project, even helped retain detail, but, eventually, at some point, as my settings evolved, it just made loss of detail... i tried some more noise reduction tests and i see loss of detail and accuracy. maybe with the default presets it works better.

i'm all for improvements, however i already tested everything heavily except ip pb ratio settings, i didn't break it down to every possibility... any ip pb ratio ideas? or are the defaults 1.4 1.3 best? does anyone change them?

the animation setting adjusts them, but that adds blending without reducing file size, and in attempt to counter this by lowering crf allows more grain/noise, and well, its meant for cel animation...



i know how to properly use everything. i just found i didn't need to adjust so much for different content.
because without using zones, a single video can have fast motion, slow motion, blended skys or detailed trees or peoples faces close up, fine hairs, to an ocean of fluid water, to little tiny twinkling stars in the dark night sky, all in one video. so one setting for one video was the first goal, then it became a "1 for all" setting.

now with cel animation being an exception i can agree on.
cel animation mixed with detailed backgrounds would be the hardest, adding tskip should help the cel animation flat colors look as one color, etc. maybe deblock of 0,0 (witch i find to be next best to 0,-3...) -1,-1 messes with blending of some grainy content, while 0,0 could bring out better picture, even sometimes more detail. -1,-1 deblock is over rated, but my 3rd choice... (1,1 for cel animation as mentioned before)



[these settings are the result of thousands of encodes, and months of work]

even so, i still would love to improve the "1 setting for all" custom settings.

i just figured why ask questions when i can encode and see for myself, and did a lot of it. just small clips of a few seconds.

as an idea for anyone reading, you can take multiple test clips from same video and combine using MKVToolNix or some other program... good way to mix scenes up, like fast & slow, a scene were good blending is needed, detail, or that impossible dark scene. however choosing a crf value with a small encode doesn't seem to work well. maybe 1 or 2 minutes should be enough for long videos. find a crf good for all scenes, and encode. the way it should be, easy :)
H2sixty is offline   Reply With Quote
Old 7th October 2020, 10:31   #23  |  Link
Boulder
Pig on the wing
 
Boulder's Avatar
 
Join Date: Mar 2002
Location: Finland
Posts: 5,718
Do you do tests by frame-to-frame comparison or viewing from your normal viewing distance? What comes to hevc-aq, I only found it to blur the image a helluva lot. Aq-mode 1 has still been the best all-round choice compared to the automatic variance methods.

It's basically impossible to remove grain or noise without removing a lot of the details as well. Motion compensated methods do help there a bit but they are still not going to do wonders. That's why I myself only do slight motion compensated denoising.
__________________
And if the band you're in starts playing different tunes
I'll see you on the dark side of the Moon...
Boulder is offline   Reply With Quote
Old 11th October 2020, 10:42   #24  |  Link
H2sixty
Not a Registered User
 
H2sixty's Avatar
 
Join Date: Aug 2020
Location: earth
Posts: 24
Quote:
Originally Posted by Boulder View Post
Do you do tests by frame-to-frame comparison or viewing from your normal viewing distance

zoomed in, 1.5x, 2x, 2.5x + frame by frame and/or in motion, play back speed 30% 70% and normal speed. even blind testing as to avoid a bias decision i may have...
comparing multiple parts of a frame multiple times since some times its a close compare. if neither look better, the more accurate encode/setting is chosen... i started with the obvious choice, frame by frame, or still frames... but eventually realized still frames do not show quality in motion... also sometimes is a good idea to make multiple encodes with same settings, as they are most likely not identical visually.


Quote:
Originally Posted by Boulder View Post
What comes to hevc-aq, I only found it to blur the image a helluva lot. aq-mode 1 has still been the best all-round choice compared to the automatic variance methods.
its a real synergy between most settings if not all. one affects the other, change one setting and all the previous tests are now irrelevant, because you now have a new "baseline"...

placebo with or without aq-mode=1 is far inferior to my custom settings. so i focused on my custom settings with & without aq-mode=1. aq-mode 1 is giving distortion on low quality and high quality sources. also where aq-mode 1 seems to have more detail, it was detail not in the source of sintel, hevc-aq was more accurate, had more detail & accuracy in numerous places. aq-mode=1 did sharpen the picture more than the source, however was blurry in areas that hevc-aq was not...


Quote:
Originally Posted by Boulder View Post
It's basically impossible to remove grain or noise without removing a lot of the details as well. Motion compensated methods do help there a bit but they are still not going to do wonders. That's why I myself only do slight motion compensated denoising.
i agree. thus i really like using one slider (crf) scale to go from blurry denoising to high detail. as incorrect as it sounds it works because the more blur, the smaller the file... and the more detail, the bigger the file... makes sense to save space if not saving detail. the custom settings acts as a motion compensated denoising.

i bet the blending could be better, but i didn't want to loose any quality and still be able to scale up to high bitrates for a "one for all" baseline setting.
-
i would like more complex encodes like tears of steel for comparison.
-
in the pix look under the persons left eye, aq-mode 1, has all this distortion, that's not detail.
or to the left of hand under face, looks like aqmode1 has a lot of detail, but its not in source. and their left arm, on the sleeve, less detail in aq-mode 1, more detail in custom settings...

pix--\/
cap_sintel_trailer_2k_1080p24____source____00.00.06_01.jpg
cap_sintel_trailer_2k_1080p24____crf=19.2____custom_____00.00.06_01.jpg
cap_sintel_trailer_2k_1080p24____crf=19.8____custom_+_aq-mode=1.aq-strength=1.00_____00.00.06_01.jpg

videos--\/
sintel_trailer_2k_1080p24____crf=18.4____placebo.mkv
sintel_trailer_2k_1080p24____crf=19.0____placebo_+_aq-mode=1.aq-strength=1.00.mkv

sintel_trailer_2k_1080p24____crf=19.2____custom.mkv
sintel_trailer_2k_1080p24____crf=19.8____custom_+_aq-mode=1.aq-strength=1.00.mkv

Sintel trailer (1253 frames 16:9 letterbox) y4m 3.8GB
https://media.xiph.org/
-
-
-
-
-
-
more pix--\/
(higher detail scene)

cap_sintel_trailer_2k_1080p24_source_00.00.19_04.jpg
cap_sintel_trailer_2k_1080p24____crf=19.2____custom_00.00.19_01.jpg
cap_sintel_trailer_2k_1080p24____crf=19.8____custom_+_aq-mode=1.aq-strength=1.00_00.00.19_01.jpg

aqmode1 in these pix just above, shows lack of detail & accuracy all over...
H2sixty is offline   Reply With Quote
Old 11th October 2020, 18:53   #25  |  Link
Boulder
Pig on the wing
 
Boulder's Avatar
 
Join Date: Mar 2002
Location: Finland
Posts: 5,718
Have you made the same tests with real life footage? CGI is a very different beast, and hevc-aq could well work better with that. A regular grainy/noisy movie is not as easy to deal with.. I did test your psychovisual settings + hevc-aq with a snip from Star Trek: Enterprise S02E01, and while they removed quite a lot of noise, they also caused the remaining noise "jump out" of the background every now and then. That was quite a lot more distracting than some floating noise.
Season 3 of that series is a total disaster quality-wise, it's like looking the video through a dirty screen because DNR has not been applied carefully. That is something I would like to be able to improve, but I don't think any encoder settings can do that.
__________________
And if the band you're in starts playing different tunes
I'll see you on the dark side of the Moon...

Last edited by Boulder; 12th October 2020 at 05:04.
Boulder is offline   Reply With Quote
Old 11th October 2020, 22:37   #26  |  Link
H2sixty
Not a Registered User
 
H2sixty's Avatar
 
Join Date: Aug 2020
Location: earth
Posts: 24
i can not find what you are talking about. what are the 2 command lines/settings you are comparing? can you share star trek? or i will have to get a snipit of live action from tears of steel, or is there something else i can encode with grain that is better, so i can share with the world a compairison... benwaggner mentioned something few posts back, but i can not find it...
Quote:
Originally Posted by benwaggoner View Post
With x265 circa 2016, I literally was unable to make those shots look anything short of unwatchable, even at 40 Mbps --preset-placebo, lots of --nr, preprocessing, anything.
is it free for share? i need live action with & without grain. ideas? links?
H2sixty is offline   Reply With Quote
Old 11th October 2020, 23:59   #27  |  Link
quietvoid
Registered User
 
Join Date: Jan 2019
Location: Canada
Posts: 570
Quote:
Originally Posted by H2sixty View Post
i can not find what you are talking about. what are the 2 command lines/settings you are comparing? can you share star trek? or i will have to get a snipit of live action from tears of steel, or is there something else i can encode with grain that is better, so i can share with the world a compairison... benwaggner mentioned something few posts back, but i can not find it...

is it free for share? i need live action with & without grain. ideas? links?
https://forum.doom9.org/showthread.php?t=175776
quietvoid is offline   Reply With Quote
Old 12th October 2020, 05:29   #28  |  Link
Boulder
Pig on the wing
 
Boulder's Avatar
 
Join Date: Mar 2002
Location: Finland
Posts: 5,718
For example these clips from Hot Fuzz are good for testing a rather good quality source.
https://drive.google.com/file/d/19cI...ew?usp=sharing
https://drive.google.com/file/d/1_we...ew?usp=sharing

For a more noisy one, this is a bit nasty to get to look good without a floating noise effect or banding.
https://drive.google.com/file/d/1N53...ew?usp=sharing
__________________
And if the band you're in starts playing different tunes
I'll see you on the dark side of the Moon...

Last edited by Boulder; 12th October 2020 at 05:33.
Boulder is offline   Reply With Quote
Old 12th October 2020, 06:55   #29  |  Link
H2sixty
Not a Registered User
 
H2sixty's Avatar
 
Join Date: Aug 2020
Location: earth
Posts: 24
Quote:
Originally Posted by Boulder View Post
For example these clips from Hot Fuzz are good for testing a rather good quality source.
https://drive.google.com/file/d/19cI...ew?usp=sharing
https://drive.google.com/file/d/1_we...ew?usp=sharing

For a more noisy one, this is a bit nasty to get to look good without a floating noise effect or banding.
https://drive.google.com/file/d/1N53...ew?usp=sharing
something more free to distribute please.

creative commons, etc...
H2sixty is offline   Reply With Quote
Old 12th October 2020, 12:37   #30  |  Link
Boulder
Pig on the wing
 
Boulder's Avatar
 
Join Date: Mar 2002
Location: Finland
Posts: 5,718
Quote:
Originally Posted by H2sixty View Post
something more free to distribute please.

creative commons, etc...
Unfortunately I have none of those available. Perhaps there are some found over the internet, but they tend to be mostly CGI. I've always considered short samples as being under the fair use policy so never bothered worrying about copyright issues.
__________________
And if the band you're in starts playing different tunes
I'll see you on the dark side of the Moon...
Boulder is offline   Reply With Quote
Old 12th October 2020, 17:05   #31  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by H2sixty View Post
a link? i can not find what this is.
I doubt there is any way to get access to those old encodes anymore. But it would be pretty easy to synthesize by just putting a really high strength small grain noise filter on top of some low contrast 4K video.

I think using --hme might be particularly helpful with that kind of content, as the noise will get largely averaged out by the downscaling so the 960x540 layer will be able to pick up vectors for the actual motion.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 12th October 2020, 17:14   #32  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by quietvoid View Post
Thanks! Although there is some grain in Tears of Steel.

Grain is a great way to mask less-than-perfect compositing, and the combination of 2K VFX with 4K source. Grain adds a psychovisual sense of sharpness that can be well beyond the actual real detail of the underlying content.

Netflix's Sol Levante is a short CGI HDR 4K anime available under the Creative Commons "Attribution-NonCommercial-NoDerivatives 4.0 International" license. It doesn't have any plot or dialog or anything, but is a fun clip to test with.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 13th October 2020, 04:42   #33  |  Link
H2sixty
Not a Registered User
 
H2sixty's Avatar
 
Join Date: Aug 2020
Location: earth
Posts: 24
Quote:
Originally Posted by Boulder View Post
I did test your psychovisual settings + hevc-aq with a snip from Star Trek: Enterprise S02E01, and while they removed quite a lot of noise, they also caused the remaining noise "jump out" of the background every now and then. That was quite a lot more distracting than some floating noise.
i think i see what you are talking about, i'll throw up some encodes in a few days and spew out the details



Quote:
Originally Posted by Boulder View Post
Season 3 of that series is a total disaster quality-wise, it's like looking the video through a dirty screen because DNR has not been applied carefully. That is something I would like to be able to improve, but I don't think any encoder settings can do that.
maybe you could share a small 30-60 second clip of this
H2sixty is offline   Reply With Quote
Old 13th October 2020, 05:36   #34  |  Link
Boulder
Pig on the wing
 
Boulder's Avatar
 
Join Date: Mar 2002
Location: Finland
Posts: 5,718
Quote:
Originally Posted by H2sixty View Post
maybe you could share a small 30-60 second clip of this
You can use these to test:
https://drive.google.com/file/d/1f_-...ew?usp=sharing
https://drive.google.com/file/d/1G8w...ew?usp=sharing

The problem is less obvious in the original, but gets enhanced in the encode. It's just too strong denoising which makes the grain so much less random that it starts to resemble a static dirty screen in some parts of the frame. The fact that it gets worse after even very slight denoising makes it even harder to handle well. One way to battle it is to add noise before the encode to hide the artifacts, but it will start requiring quite a lot more bitrate.
__________________
And if the band you're in starts playing different tunes
I'll see you on the dark side of the Moon...
Boulder is offline   Reply With Quote
Reply

Tags
audio, handbrake, hevc, opus, x265

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 13:44.


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