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 > MPEG-4 AVC / H.264

Reply
 
Thread Tools Search this Thread Display Modes
Old 8th September 2022, 20:24   #1  |  Link
andiandi
Registered User
 
Join Date: Aug 2017
Posts: 58
Some questions about psy-rd

Hi,

I often tend to prefer lower psy-rd values, but first I'd like to know more about the following points :

- Does it mostly depends on personal preferences since it's psychovisual, or is there any broader recommandations (excluding anime values recommandations and not going beyond 1.0) ?

- Does lowering psy-rd have some advantages as well ? It seems at least better on flat areas and constrats from what I read here and there and from some tests I did.

- I noticed that some psy-rd values appears to be less "stable", particularly between 0.6 and 0.9 : it causes some kinds of artificats that are more noticeable and bothering than those of 1.0 or 0.4. Could it be for a reason ?

Thanks.
andiandi is offline   Reply With Quote
Old 16th September 2022, 18:01   #2  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
There's more concrete to it than just personal preferences, but it certainly is somewhat content dependent. Generally higher values can be used with very low noise natural image content. And it interacts with a variety of other parameters, so even with the same content, the optimal value can vary some.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 17th September 2022, 18:48   #3  |  Link
andiandi
Registered User
 
Join Date: Aug 2017
Posts: 58
Quote:
Originally Posted by benwaggoner View Post
There's more concrete to it than just personal preferences, but it certainly is somewhat content dependent. Generally higher values can be used with very low noise natural image content. And it interacts with a variety of other parameters, so even with the same content, the optimal value can vary some.
Thanks

I was barely focused on psy-rd but even though I have little knowledge about it, I realized it's not easy to determine a "nearly optimal" value.

Currently, I find that values such as 0.8 or 0.7 are more or less better on most of my encodings, but the visual issues that come with those values made me wonder... (otherwise I'd just have slightly lowered the value to stay in a "safe range" and that's it. Whereas in this case, I have to lower it to something like 0.5 to have a sufficient visual balance on most of sources I tested)


It seems even more weird when lower values are supposed to mean less visual issues, but it doesn't look that simple. Actually, what I was wondering is if it's generally considered "safe" to lower psy-RD at all ? (whatever the values are, excluding the impact on details and sharpness of course), and I'm also wondering if it's related to AQ (I may be completely wrong, but as far as I understand, Psy-RD works the opposite way of AQ, so I'm guessing their default values were intended to balance each other, which would then be something to consider).

Anyway, I hope i'm not digressing too much and that it's less tricky than I think...

Last edited by andiandi; 17th September 2022 at 18:52.
andiandi is offline   Reply With Quote
Old 17th September 2022, 21:45   #4  |  Link
microchip8
ffx264/ffhevc author
 
microchip8's Avatar
 
Join Date: May 2007
Location: /dev/video0
Posts: 1,843
If you want to preserve as much as the original (inc. noise/grain), high values of both psy-rd and psy-rdoq can really help. Personally I use --psy-rd 4 and --psy-rdoq 15, the latter really helps eliminating banding.
__________________
ffx264 || ffhevc || ffxvid || microenc
microchip8 is offline   Reply With Quote
Old 18th September 2022, 00:12   #5  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by andiandi View Post
I was barely focused on psy-rd but even though I have little knowledge about it, I realized it's not easy to determine a "nearly optimal" value.

Currently, I find that values such as 0.8 or 0.7 are more or less better on most of my encodings, but the visual issues that come with those values made me wonder... (otherwise I'd just have slightly lowered the value to stay in a "safe range" and that's it. Whereas in this case, I have to lower it to something like 0.5 to have a sufficient visual balance on most of sources I tested)


It seems even more weird when lower values are supposed to mean less visual issues, but it doesn't look that simple. Actually, what I was wondering is if it's generally considered "safe" to lower psy-RD at all ? (whatever the values are, excluding the impact on details and sharpness of course), and I'm also wondering if it's related to AQ (I may be completely wrong, but as far as I understand, Psy-RD works the opposite way of AQ, so I'm guessing their default values were intended to balance each other, which would then be something to consider).
Psy-rd is a psychovisual optimization. Lower values will give a more basic encode focused on mathematically more than psychovisual accuracy. At high bitrates, it doesn't matter much since everything looks good. It's at lower bitrates where psy-rd helps you tune to retain the more desired details when there aren't enough bits to preserve all the details.

It's a tool to best test using a 2-pass ABR encode, so you can compare the impact on the settings at the same file size. Testing with CRF you get both quality and bitrate changes, so it's hard to tell if you're getting better psychovisual compression efficiency.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 18th September 2022, 19:02   #6  |  Link
andiandi
Registered User
 
Join Date: Aug 2017
Posts: 58
Quote:
Originally Posted by microchip8 View Post
If you want to preserve as much as the original (inc. noise/grain), high values of both psy-rd and psy-rdoq can really help. Personally I use --psy-rd 4 and --psy-rdoq 15, the latter really helps eliminating banding.
Hi,
I'm not using x265 for now (I would have liked to use it for archive purposes with high bitrates, but apparently x264 is better for this - the question about Psy-RD is for my regular encodings)


Quote:
Originally Posted by benwaggoner View Post
Psy-rd is a psychovisual optimization. Lower values will give a more basic encode focused on mathematically more than psychovisual accuracy. At high bitrates, it doesn't matter much since everything looks good. It's at lower bitrates where psy-rd helps you tune to retain the more desired details when there aren't enough bits to preserve all the details.

It's a tool to best test using a 2-pass ABR encode, so you can compare the impact on the settings at the same file size. Testing with CRF you get both quality and bitrate changes, so it's hard to tell if you're getting better psychovisual compression efficiency.
Yes, I know it's purpose but I'm not sure about some points (particularly if it takes bits on flat areas), and I think that the default value is sometimes a bit too strong, but it's way more noticeable with subme above 8 (and a bit with Trellis 2). Maybe that's because I use too low bitrates for my tests (in order to see what helps the most), but given that I have weird results when I lower it just a little bit, I was wondering if there was specific reasons for this or no, because I only expected a slight loss of details (which doesn't bother me)

Last edited by andiandi; 18th September 2022 at 19:08.
andiandi 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 22:47.


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