Variance AQ Megathread (AQ v0.48 update--defaults changed)

[Yoshiyuki Blade]

Answer: stop adjusting your sensitivity. It looks just fine at sensitivity 13. Kthx. :p

Yeah, I think upping the strength to 0.6 is a fairly decent number for some tests I've done. The results seem change rather drastically moving up in increments of 0.1 strength at a time (where 1.0 is horrible). At this point, the difference in quality is slight in comparison to the older AQ, so I'm probably not gonna sweat on the whole AQ v0.47+ vs v0.46 issue for anime.

[bkman]

Dark, I can't remember if you already have or not (this thread is rather large :P) but can you explain the exact meaning of the strength and sensitivity settings within your AQ method? And what is the expected result from changing each?

I'd like to be able to optimise the settings more intelligently than just trial and error.

[Dark Shikari]

Dark, I can't remember if you already have or not (this thread is rather large :P) but can you explain the exact meaning of the strength and sensitivity settings within your AQ method? And what is the expected result from changing each?

I'd like to be able to optimise the settings more intelligently than just trial and error.Sensitivity should not be changed, stop touching it. It will be removed when AQ hits SVN.

Strength just affects how wide the range of quantizers used is.

[Blue_MiSfit]

I'm very pleased with the quality of AQ. I have enabled it for my HD-DVD Planet Earth backups, and things look good at ~10mbit. I'm backing up to fit 2 episodes on a DVD9, and it's hard for me to tell a difference from the source.

Anyone with normal eyes wouldn't stand a chance.

The AQ really helps improve detailed areas, and also evens out the big flat areas which are so problematic.

I add a tiny bit of noise in ffdshow (10-15 luma) for 1080p content, and it's a very visually pleasing image to me!

~MiSfit

[burfadel]

I add a tiny bit of noise in ffdshow (10-15 luma) for 1080p content, and it's a very visually pleasing image to me!
~MiSfit

Adding a bit of noise through ffdshow does make things look better! However, the default settings aren't optimal (in my opinion). Chroma noise makes things look worse I think, and a setting for luma of 30 by default is too high as well!

The other problem with the noise function is that its different for each resolution, so what looks good at 1080 looks quite grainy at low res.

Anyways, AQ 0,48 is great now, I can raise the CRF a point or two (or more) without it looking worse than the lower CRF with AQ off - in other words its visually much more bit efficient. Keep up the good work!

[Blue_MiSfit]

I definitely alter the ffdshow settings considerably :) No chroma, much less luma, and I switch the noise mode to something other than the defaults (not sure what those are offhand though)

/OT

I still do 2 pass, but I find overall visual quality improves overall at the same bitrate, so that's all I care about!

~MiSfit

[Sagittaire]

For high quality encoding (1080p at more than 6 Mbps for example) the most difficult part are dark area. Why not make different AQ level with different mask (DarkMasking, ComplexityMasking ... ect ect) exactly like libavcodec.

[CruNcher]

Sagittaire Dark Areas are problematic indeed (alot is hidden in the luminance) but correctly calibrated it shouldn't be that visible @ all under normal viewing conditions for most people. Tough noise that gets really blocky can get even visible then to trained eyes (and that's where the new AQ does perfect correctly balanced and imho Akus first released version was the most efficient balanced compared to the latest), definitely non (wrong) calibrated it gets visible to everyone, tough for really low bitrate and a complete 1 pass scenario the old AQ seemed better balanceing everything and don't left any scenes without enough bitrate to be visual mega pleasing (for trained eyes correctly calibrated) in such low luminance parts but therfore it didn't enhance the detail preservation (frame wise) like Darks AQ does it, the most efficient of both worlds combined would be the ultimate for a 1 pass (lookahead) scenario i think :).

[DeathTheSheep]

Answer: stop adjusting your sensitivity. It looks just fine at sensitivity 13. Kthx. :p

Sensitivity 13? Strength 0.6? "Just fine?" Are you guys absolutely kidding me? If so, well... please don't. Kthx. :p

0.9726494 SSIM, at very junky quality (I'd honestly say a non-AQ encode looks better). It's a ringing and artifact melee, with a lot of dark edges utterly botched. No, increasing deblocking doesn't help, obviously. How does this compare to the sharp, crisp, clean effect of 0.46? Well, for a limited time only, you can Take a look(TM)!

x264.dark.aq.0.47.exe -o horrible.264 -q28 --aq-strength 0.6 --aq-sensitivity 13
--level 1.3 --no-cabac -m6 --thread-input --me umh --sar 1:1 --keyint 1500 script.avs --progress

Output here (http://gabext.com/samples/Shikari/strength06.264). Is this really what you meant, or am I getting my settings wrong here, or...?

[Dark Shikari]

Sensitivity 13? Strength 0.6? "Just fine?" Are you guys absolutely kidding me? If so, well... please don't. Kthx. :p

0.9726494 SSIM, at very junky quality (I'd honestly say a non-AQ encode looks better). It's a ringing and artifact melee, with a lot of dark edges utterly botched. No, increasing deblocking doesn't help, obviously. How does this compare to the sharp, crisp, clean effect of 0.46? Well, for a limited time only, you can Take a look(TM)!

x264.dark.aq.0.47.exe -o horrible.264 -q28 --aq-strength 0.6 --aq-sensitivity 13
--level 1.3 --no-cabac -m6 --thread-input --me umh --sar 1:1 --keyint 1500 script.avs --progress

Output here (http://gabext.com/samples/Shikari/strength06.264). Is this really what you meant, or am I getting my settings wrong here, or...?Looks perfectly fine to me... :rolleyes:

I don't know what kind of stuff you're on, but I want some of it.

PROTIP: If you're expecting perfect quality at QP28 in anime, you're crazy. And if you expect to get good efficiency encoding anime without B-frames, and without CABAC, you're an idiot. Your sample is using none of the settings that give x264 its efficiency--and the source has absurdly high motion--yet you're complaining about low quality at your low bitrate.

In my own tests, AQ 0.47 at default settings looked better than no AQ at all. I'm not responsible for other peoples' incompetence at encoding, or insisting that their quality should be perfect while using such an absurdly high quantizer.

Not to say that 450kbps isn't enough for good quality--with proper settings and encoding, you can fit pretty good quality in that. Baseline Profile H.264 is pretty damn weak though.

[DeathTheSheep]

DeathTheSheep munches dead grass, I'll have you know.

Of course no anime can look perfect at Q28. By definition of lossy encoding, nothing at all can except perhaps a blank screen or static edges that align perfectly with macroblock borders or some such artificial contrivance.

And of course baseline (as per Zune, QT, iPod, PSP, all decoder compatible) requires the absence of CABAC and B-frames and such--also by definition, obviously.

What's wrong with the clip, you ask? The edges, man, the edges are destroyed. All right, I'll prepare some more images... :rolleyes:

[Dark Shikari]

The edges, man, the edges are destroyed.

Q28

You keep answering your own question.

Moreso, you're doing your comparisons at a specific quantizer, rather than a specific bitrate, which is meaningless because AQ does not keep bitrate constant.

Looking over your encode again, the ringing isn't bad at all even at such a high quantizer; its only really visible when you zoom in, and you can't zoom in on an iPod. Ringing on low-resolution sources when one zooms in is expected.

[DeathTheSheep]

Wrong. Look at 0.46. It looks more than half decent. Look at 47. It doesn't. No fool questions why his high QP encodes don't look perfect; that's utterly ridiculous. When it's clear 0.46 produces better results than 0.47 on this clip, why do you keep refuting this? I thought we were interested in explaining (or bugfixing) rather than baselessly shooting people down or trying to debunk results. Sensitivity 13, strength 0.6 (or strength 1.0) is also worse than my previously chosen threshold in terms of quality, no matter how algorithmically "intentional" the QP distribution. You don't need any zoom to see this on the aforementioned devices/applications, but many forum members here are on a high-res, small-size monitor, which I've accounted for in using bilinear interpolation to achieve an easier viewing size.

What are you talking about, implying that I'm not keeping bitrate constant? Of course I am. All tests must end up within 1KB of each other for them to be valid to me. I thought you saw the files yourself, so what are you saying? Heck, letting x264 do a 2-pass encode with these settings doesn't even get the bitrates as close as I do with QP (1KB or less, in most cases, out of a ~4.5MB file).

Now the pictures. I don't even need an explanation. 0.47 sensitivity 13, compared to the same 0.46 I used in the big test. Filesizes ~1KB of one another. Here we go.

Here's 0.47:
http://gabext.com/samples/Shikari/7-2.PNG
and 0.46:
http://gabext.com/samples/Shikari/6-2.PNG
Less artifacting, smearing, tons more detail--Look at the character's arm! Also look at the edges of his head. Look at the sky--free of blotching and color trails. Crisper, sharper, objects look more solid, lines clearly straighter.

Again. Here's 0.47:
http://gabext.com/samples/Shikari/7-3.PNG
and .46:
http://gabext.com/samples/Shikari/6-3.PNG
Look at the reflection in the sword hilt. 0.46 is sharper, clearer, less artifacts. Left side of face is now intact, not chipped out. And look at that ugly smear in 0.47 to the left of the sword hilt. What is that? It's hideous, especially in motion.

Again! 0.47:
http://gabext.com/samples/Shikari/7-4.PNG
...and 0.46:
http://gabext.com/samples/Shikari/6-4.PNG
Look at left character's leg edges. Speaks for itself. Look at right character's mouth. Where is his mouth in 0.47?

Now for a real dark scene, where AQ supposedly helps most. Here's 0.47:
http://gabext.com/samples/Shikari/7-5.PNG
And 0.46:
http://gabext.com/samples/Shikari/6-5.PNG
Does this really need explanation? Very crisp, clean, sharp in 0.46 like expected. And here's the kingpin: *Less blocking in flat areas.*

And I just stopped here. It's not just on a smattering of scenes. It's really on just about all of them. In motion, it's even more apparent. Much, much more so, especially on the handheld devices it's made for. Need I reiterate the much higher SSIM of the 0.46 encode, either?

[edit]And on an encode of strength of 0.7 or higher (same sensitivity=13), it's much worse for the poor edges.

No, this isn't a matter of making something perfect at *any* bitrate. It's a question of which algorithm presents the better answer in these cases.

[CruNcher]

@DeathTheSheep
Sometimes i'm also suprised about Darks Visual Cortex ;)
and im with you and your 0.46 vs 0.47 results even as a non Anime guy i agree with them, Visual Quality decreased for low bitrate at least and it's hard to belive that people don't see that :)
could you please try to shift the AQ results of 0.47 useing a cqm best bet here is the prestige matrix (and no i don't either understand what people have against the prestige matrix).

[Dark Shikari]

Um... DtS... you're making no sense.

The purpose of AQ is to improve background areas.

It does that, perfectly, in every picture you've posted. 0.47 is better in all of them at what its supposed to do. The reason it sometimes looks worse in the background is because there's banding in your source, and lowering the quality actually helps improve that a bit.

You're complaining it makes edges less sharp.

That's what its supposed to do. If you don't like it, don't use it.

(Moreso: you're zooming into 320x240 images and reporting quality problems. This is a bad method of comparison because it creates aliasing and such that was not in the source.)

[DeathTheSheep]

@DeathTheSheep
Sometimes i'm also suprised about Darks Visual Cortex ;)
and im with you and your 0.46 vs 0.47 results even as a non Anime guy i agree with them, Visual Quality decreased for low bitrate at least and it's hard to belive that people don't see that :)

...DS, you...really can't see that difference? You're...kidding...right? Please tell me you're kidding? Okay, somebody tell me how to do a decent anigif...

could you please try to shift the AQ results useing a cqm best bet here is the prestige matrix (and no i don't either understand what people have against the prestige matrix).
Ah, the matrix of course helps a ton, and so does CABAC, trellis, 16 refs, and at least some B-frames! Unfortunately, my encoding is only for the devices/decoders I mentioned, which only support baseline profile. Sadly, CQMs and baseline don't tend to mix. :( That's why

[Dark Shikari]

OK, I just did my own encode and I still can't see what in the world you're talking about.

Top is encode, bottom is source, same bitrate as yours.

http://i29.tinypic.com/29w0qd5.png

Seriously, it looks fine. What are you complaining about? You're not going to get flawless detail retention at 450kbit.

[DeathTheSheep]

That's a keyframe. ^_^'

[Dark Shikari]

That's a keyframe. ^_^'Yes, a 1500 byte keyframe.

[CruNcher]

I see DeathTheSheep maybe posting pics of the source aside can make people realize the problems we are talking about ;)

[DeathTheSheep]

Keyframes always look almost the same regardless of what AQ you use.

So you found the one frame in the entire clip that comes close to 0.46...and it's a keyframe. I think I speak for more than myself when I say there's something suspicious about that.

...Almost as if your results are rigged. ;)

You want the source? I uploaded it, too. But in the source, the artifacts I mentioned obviously weren't present, and the characters did have mouths, believe it or not. The lines weren't hideously jagged. Color wasn't smeared everywhere.

It's a keyframe. It's not a predicted frame, where artifacts, motion trails, smearing, gradient motion blocking, etc really have a chance to occur. The problems I mention (and indeed, pretty much any possible problems at CQP) arise after the I-frames. :rolleyes:

All of them look fine, so does a non-AQ, so do old AQ (aku's, haali, et al). What are you trying to prove here by showing a frame that looks, for all intents and purposes, the same in almost any situation?

[Dark Shikari]

Keyframes always look almost the same regardless of what AQ you use.

So you found the one frame in the entire clip that comes close to 0.46...and it's a keyframe. I think I speak for more than myself when I say there's something suspicious about that.

...Almost like your results are rigged. ;)Its one of the smallest frames in the entire section; in fact, I'd say its far too small and the ratecontrol should be giving it more.

Anyways, I'm running a test of my own with 0.46 vs 0.48 at the same settings.

[Dark Shikari]

Oh wow. Now this really surprised me.

The reason AQ 0.46 looks better in background areas, even though AQ 0.48 lowers the quantizers there more, is because the deblocking filter is stronger at higher quantizers.

Literally, areas with lower quantizers look worse in 0.48 than 0.46 where the quantizers were higher--and that's the only explanation I can possibly think of. AQ 0.48 is actually backfiring despite doing an overall better job.

Indeed this seems to apply to a lot of things; the gradient in the anime clips is extremely smooth, and as such the deblocking filter tends to do a very good job at approximating it.

[DeathTheSheep]

Settings are not bizzare at all. They're actually what AVC is most used for these days--mobile encoding.

Mobile uses baseline.

Baseline cannot use B-frames or CABAC.

To "ignore" this obviousness is to choose "ignor"ance, quite literally. Anybody who has to pick between the 2 clips I linked to for download (with full commandline, nothing dubious here) will obviously choose 0.46, as even CruNcher admits.

If you don't like the keyframes in 0.46, use --ipratio 1.5. Then they'll get a boost.

This isn't dubious. This is, on the contrary, obvious. :rolleyes:

[DeathTheSheep]

Woah, just got your post. I gotta say...wow!

See, there IS an explanation!! But try this for even more excitement: use --ipratio 1.5 while maintaining bitrate. You might be quite surprised. SSIM goes even higher with 0.46.

[Dark Shikari]

To "ignore" this obviousness is to choose "ignor"ance, quite literally. Anybody who has to pick between the 2 clips I linked to for download (with full commandline, nothing dubious here) will obviously choose 0.46, as even CruNcher admits. There's a reason I suggested not using AQ with anime.

After inspection, I have come to the conclusion that the reason 0.46 looks better than 0.48 is because it is not as powerful. If you removed AQ altogether, it would likely look even better.

Anime is all about minimizing ringing. AQ increases ringing.

[DeathTheSheep]

There's a reason I suggested not using AQ with anime.

The reason 0.46 looks better than 0.48 is because it is not as powerful. If you removed AQ altogether, it would likely look even better.

Absolutely positively not. (See, I'm 100% sure now).

SSIM goes so far down it hurts, and things get so blocky and smeary I actually warn against eye bleeding. For our sake and yours, don't even suggest that--people might get the wrong idea, and seriously end up in the ER.

[Dark Shikari]

Absolutely positively not. (See, I'm 100% sure now).

SSIM goes so far down it hurts, and things get so blocky and smeary I warn against eye bleeding. For our sake and yours, don't even suggest that--people might get the wrong idea, seriously.Well I'm encoding a clip without AQ, so we'll see ;)

Edit: WOW. Without AQ its not even a contest, it absolutely shreds the AQ encode at this low resolution. Zooming in only makes it more obvious.

In that dark scene you pointed out earlier, AQ does better, but only because it adds more bits to it. The exact same effect could be gotten by reducing cplxblur and qcomp without AQ.

The non-AQ encode at the same CRF turned out higher quality and lower bitrate. At this point I think its safe to conclude that at these low resolutions, AQ is useless for anime.

[DeathTheSheep]

Spare me!!! Oh, the horror... Want some more screenshots? (you might wanna be in peak health before looking).


Here we go.

NO AQ:
http://gabext.com/samples/Shikari/noAQ.PNG
0.46:
http://gabext.com/samples/Shikari/6-5.PNG

I'd say noAQ is almost as bad as 0.47 (if not worse).

[DeathTheSheep]

Again with the riggin? I don't use Qcomp. This is CQP... *sigh*

I already mentioned how Qcomp destroys clips at this bitrate--in high motion scenes, there's nothing but slop.

No, for anime, AQ is a much better RC algorithm than standard Qcomp crf, as already verified here by different users.

And no, it's not for one measly scene, it's for the vast majority of the clip. Even bright edge details are brought out nicely, and if your theory about 0.46 is correct, the loopfilter keeps ringing to a minimum elsewhere. I dare you (seriously) to give me more optimal results than AQ.46. You can use any combination of qcomp and complexity reduction settings and AQ version, keeping all other options the same. Go for it, I'm open. :)

[Dark Shikari]

Spare me!!! Oh, the horror... Want some more screenshots? (you might wanna be in peak health before looking).Please read my post before responding. Almost the entire rest of the video looks better without AQ, and that one scene would look just fine without AQ if you lowered your qcomp.

You insist on using QP instead of CRF, even though CRF does a far better job in that scene yet looks drastically better than AQ everywhere else.

[CruNcher]

I just wana might add that i tested with Real Film Footage ;) with very fine details no Anime involved im my Visual tests and i got the same results especialy ringing got to problematic with 0.47 that it becomes a real Visual problem especialy for trained eyes.

[Dark Shikari]

Come on, you have to be kidding me.

http://i29.tinypic.com/2nle07a.png

http://i26.tinypic.com/einuwh.jpg

One of these is AQ 0.46, the other isn't (not saying which). The AQ 0.46 one is 20% higher bitrate than the other. Which one is it?

[CruNcher]

the top one hurts the bottom one looks ok but compareing 1 still frame is far from what i call a subjective quality test :D

[Dark Shikari]

the top one hurts the bottom one looks ok but compareing 1 still frame is far from what i call a subjective quality test :DIt looks nearly the same on all frames. AQ causes massive ringing, regardless of whether its 0.47 or 0.46 or whatever. The quality gain is near zero, and the quality loss is huge. I inspected dozens of frames throughout the video, and the effect is the same on all of them.

At this point, I trust my eyes. Any claim that AQ is good for this source (especially based on that single dark scene, while the entire rest of the video suffers horribly) is laughable and deserves to be mocked.

It is not my job to deal with people who are so in love with their own pet algorithm that they didn't even make (AQ 0.46) that they ignore that AQ as a whole causes a problem for the source. I stick to my original statement that AQ is probably not a good idea for anime in general.

Complaining about dark scenes while at the same time refusing to use qcomp, which is designed to improve quality in dark/flat scenes, is absolutely stupid. I do not argue with stupidity. I am done with this conversation.

[DeathTheSheep]

Top is AQ, bottom isn't. Give me your result clip. :)

I'd trade in good dark scenes for this kind of negligible difference any day. You can always reduce the strength, too, if you don't like it. 0.46 outperforms .47 at lower strengths too, you see. ;)

[CruNcher]

http://forum.doom9.org/showpost.php?p=1095104&postcount=663 <- Dark and whats about this no Anime here same Visual Problems
compare this with http://forum.doom9.org/showpost.php?p=1091711&postcount=514

[Dark Shikari]

Strange ringing

AQ strength 1.0

:rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes::rolleyes:

[DeathTheSheep]

But according to him, any less than 1.0 and results are sub-optimal for 0.47 as well (he posted about this before, remember?).

So yes, there is a problem there, too, no getting around it (unless you use a good matrix and such to mitigate losses).

It all comes down to how much you're willing to sacrifice in terms of high-contrast areas (and the strengths/thresholds you use to do so).

[CruNcher]

yes but to low and the noise in the black area gets to blocky no way to compensate that with 0.47 0.46 didn't need that as you can see if you watch all areas closely.

[Dark Shikari]

AQ increases ringing and decreases blocking/blurring.

If you cannot decrease blocking without causing ringing problems, you need more bitrate.

How many times do I have to explain this?

If people cannot accept simple limitations of the algorithm like this, I'm going to simply delete this thread and continue any further development in private.

[DeathTheSheep]

CruNcher: So you're saying .47 doesn't even work with your matrix fix?

DS: Of course we know this. We know what AQ does by now. We're just saying 0.46 may be the better of the two, after all.

[IgorC]

What about public blind test with something like at least 18 samples and 20 viewers? Would be reasonable. Perception isn't the same for all persons.Each one has its own idea about subjective quality, visual deficiencies etc.

[Sharktooth]

D_S: that's why i was so harsh when ppl asked if AQ was going to be commited to SVN...
ppl never learn, ppl think they know it all even if they dont understand what they say is their subjective opinion...
so, dont listen to ppl when they become silly and go ahead.

[Dark Shikari]

CruNcher: So you're saying .47 doesn't even work with your matrix fix?

DS: Of course we know this. We know what AQ does by now. We're just saying 0.46 may be the better of the two, after all.The only thing that I have seen of any comparison is that 0.46 is better because it is weaker, and it is only better in the case in which no AQ is better than AQ.

:rolleyes:

Sorry, at this point, if you want to improve AQ in a way you like it, do it yourself.

[DeathTheSheep]

Who's being silly here--isn't that your own subjective opinion? Did you even see the results of the tests I've posted, Sharktooth?

It's always debatable whether AQ or noAQ looks better in certain scenes--it's all a matter of tradeoffs--ringing for better dark scenes.

But between the two AQs themselves (0.46 and 0.48), that's a matter that does need looking into before blindly "going ahead."

IgorC's blind test is an excellent idea, but if that doesn't satisfy people, why not include both? It's a matter of using one rounding method or another, no more, no less. A matter of a one-line if statement to chose which to invoke.

I'm just worried (and for good reason) that 0.46, which to my eyes and some others' is the superior algorithm, will simply get the shaft before it's committed.

[CruNcher]

CruNcher: So you're saying .47 doesn't even work with your matrix fix?

DS: Of course we know this. We know what AQ does by now. We're just saying 0.46 may be the better of the two, after all.

It improves it it's shifting it and so the ringing @ the edges is not that visible anymore it almost looks like 0.46 then again.
And yeah DS and me are talking about what's balanced and what isn't or overdone and will hurt at lower bitrate scenarios and highering the bitrate is no option here but anyway i found a visual aceptable workaround for this so im not scared if it goes like this into SVN @ all now.
And yeah i know that im @ the edge here im allways @ it and so i might see what others don't but every visual improvement for low bitrate is important jesus i wish people would think a little bit more like the Ateme Devs (not so metric obsesed) in this specific situation we just talking about what's more efficiently balanced and if people that do low bitrate (very high quant encoding) have to suffer from this im not sure if it should go like this into SVN, if thats the case as it wouldn't be balanced in my eyes then.
Quality should scale with bitrate and if you can prevent ringing in low bitrate situations you should do it @ all costs (that's what H.264 stands for compared to ASP) as the rest (more details) come with a bitrate increase @ low bitrates Picture stability should go over Detail Preservation in any case.

[Dark Shikari]

I'm just worried (and for good reason) that 0.46, which to my eyes and some others' is the superior algorithm, will simply get the shaft before it's committed.From an unbiased perspective here (ignoring which AQ is better), there is a snowball's chance in hell that you will get 0.46 committed. Pengvado will never, ever accept patches with arbitrary rounding bugs in them, even if they have a positive effect.

[IgorC]

D_S: that's why i was so harsh when ppl asked if AQ was going to be commited to SVN...
ppl never learn, ppl think they know it all even if they dont understand what they say is their subjective opinion...
so, dont listen to ppl when they become silly and go ahead.

I think there is another problem that people don't realize what they see (or heard). Same things happened for example during LAME delevopment. When people say they can heard weird artifacts on mp3 files. But blind test (with hidden original) can't prove that.
That why all statements about audio quality should based on blind test in hydrogenaudio forum. http://www.hydrogenaudio.org/forums/index.php?showtopic=16295
If there is no blind test it's violation of the rules.
Don't listen one person, than another and another. It's easy way to lose yourself in statements of others.
Sorry if it seems like hypocrisy.

[DeathTheSheep]

No, it's a great idea, IgorC.

The only thing that I have seen of any comparison is that 0.46 is better because it is weaker, and it is only better in the case in which no AQ is better than AQ.
That's not the whole story. If it was simply a matter of adjusting strengths, one would just lower 0.47's strength to get the better result of the two. 0.47 does not distribute QPs in a visually pleasing way in the scenes that need it most, whereas 0.46 does, at the same bitrate. Using no AQ doesn't benefit the scenes of interest at all, which are the worst scenes in the encode. We're sacrificing quality in some places to retain it in others, for a constant quality effect. Keep in mind, for dark/flat scenes (which is all that matters at this bitrate--who cares about a little ringing here and there when you can't even see what's happening in darker frames?), 0.46 is better than 0.47, which is the purpose of the AQ. SSIM is higher, which you certainly tote as a key point, too. In the high-contrast scenes, there is little difference between the two algorithms anyway.

All that's left is to do the double blind. I tried to convey the expression in my test post containing screenshots of the worst-looking scenes. I thought, people who see that the horrible quality in these areas (which is why we're using AQ even at its potential ringing cost), can get a good idea of which is better. The people I've shown agreed with me here, so I made the post.

We just need IgorC's test to prove it.