Recommended psy_rdo & aq-strength settings?

[Forteen88]

Hi,
I got a question: What x264 psy_rdo & aq-strength settings are recommended (for HD, 1080p or 720p) normal film (not cartoon/anime)?

I guess different settings only depends on the amount of grain on the source?!

I use psy_rdo 1.2:0.0 and aq-strength 0.8, but I don't know if it's that great as it could be.

And should 3D-animated movies be encoded with same setting as normal film?

EDIT: Different settings also depends on the bitrate.

[LoRd_MuldeR]

Before you try anything else, you may want to give Psy Trellis a try...

[Forteen88]

Before you try anything else, you may want to give Psy Trellis a try...Thanks, but I saw some frames in a video-comparison, and Psy-trellis was creating artifacts (maybe because of its sharpening?):

Original:
http://www.imagebam.com/image/13edc723204198

crf 20, w/ --psy-rd 1.3:0.0 --aq 0.9
http://hdimage.org/images/jch6ko9pr94ix5zxb9kf_crf201.png

crf 20, w/ --psy-rd 1.3:0.1 --aq 0.9
http://hdimage.org/images/n6qq6jke46y88lnzubn4_crf202.png

[Selur]

might be me, but could you post a picture where you highlight the artifacts,... (I see some noise differences between each of the three, but nothing that jumps my eye as artifact)

[burfadel]

try --psy-rd 1.0:1.0

Psy-rd 1.3 is fairly strong, and psy-trellis 0.1 is very weak. I can't imagine such a weak setting creating harmful artifacts over being disabled. Try 1.0:1.0 and see how things turn out.

[Forteen88]

might be me, but could you post a picture where you highlight the artifacts,... (I see some noise differences between each of the three, but nothing that jumps my eye as artifact)Personally it's mainly the lips and forehead of the guy (Agent Smith) that don't look the same as the source, but other complained about psy_trellis artifacts in this picture also (didn't ask them why, unfortunately that's an old thread now).

@burfadel, wow, that's strong psy_trellis, but I suppose it might be good on some sources, or especially when doing bitrate-starved encodes, but for 1080p I'd rather increase bitrate than get artifacts psy_trellis creates.

[Adub]

@burfadel, wow, that's strong psy_trellis, but I suppose it might be good on some sources.

Actually its not. It's is often the base psy-trellis value that many people start with and then vary a couple of tenths of a decimal in either direction.

[Dark Shikari]

Actually its not. It's is often the base psy-trellis value that many people start withAre you nuts? 1.0 is generally quite overkill (as I discovered soon after committing the patch)... 0.2-0.4 seems to be more than sufficient for most situations.

[Chengbin]

Are you nuts? 1.0 is generally quite overkill (as I discovered soon after committing the patch)... 0.2-0.4 seems to be more than sufficient for most situations.

Are you saying the default value 1 is overkill???

[Dark Shikari]

Are you saying the default value 1 is overkill???The default value of psy-trellis is zero.

[LoRd_MuldeR]

Are you saying the default value 1 is overkill???

He talks about Psy Trellis, not Psy RDO. The default value for Psy Trellis is 0.0 ;)

@Dark Shikari:
If Psy Trellis 1.0 usually is overkill, wouldn't it be a good idea to adjust the scale of Psy Trellis then? Like making 0.4 the new 1.0?

[Dark Shikari]

He talks about Psy Trellis, not Psy RDO. The default value for Psy Trellis is 0.0 ;)

@Dark Shikari:
If Psy Trellis 1.0 usually is overkill, wouldn't it be a good idea to adjust the scale of Psy Trellis then? Like making 0.4 the new 1.0?That would be a nice way to confuse everyone used to using it.

I'll do that when I fix psy-trellis and psy-RDO to be better.

[LoRd_MuldeR]

That would be a nice way to confuse everyone used to using it.

Hehe, indeed. But everybody not following the x264 git log will be confused anyway - sooner or later :p

I'll do that when I fix psy-trellis and psy-RDO to be better.

Is there anything specifically in the making yet? Maybe related to this (http://forum.doom9.org/showthread.php?t=146507) thread? :)

[Dark Shikari]

Is there anything specifically in the making yet? Maybe related to this (http://forum.doom9.org/showthread.php?t=146507) thread? :)Mostly me just bruteforcing a bunch of psy metrics with QNS.

It's only about 40 times slower than subme 9+trellis 2.

[Chengbin]

DS, I'm curious, what if you use an overkill value for psy, for example a value of 2 for psy-rd, what will happen to the video?

[Dark Shikari]

DS, I'm curious, what if you use an overkill value for psy, for example a value of 2 for psy-rd, what will happen to the video?Psy-RD strength represents the relative importance of detail versus quality.

Too high psy RD will bias too much in favor of detail, creating artifacts.

[IgorC]

As I understood Psy RD is smart adjustment of lowpass filter. Higher values of Psy RD increase cutoff frequency (ouptut video has more frequency content) at cost of lower data rate per frequency band.

[Dark Shikari]

As I understood Psy RD is smart adjustment of lowpass filter. Higher values of Psy RD increase cutoff frequency (ouptut video has more frequency content) at cost of lower data rate per frequency band.x264 has no lowpass filter, and psy-RD actually in practice works in reverse; it allows low frequencies to take the place of high frequencies in terms of energy.

This is IMO a bad property but doesn't cause any issues currently; it only crops up when you throw QNS into the mix.