Psy RDO: Official testing thread (version 0.6 out!)

[Dark Shikari]

Are you sure you are comparing the same type of frames? :pYes, both are P-frames. :)

[Razorholt]

So, does it mean that we "may" compare frames as long as they are the same type? ... I mean, in order to compare encoded files.

[Dark Shikari]

So, does it mean that we "may" compare frames as long as they are the same type? ... I mean, in order to compare encoded files.Comparing frames is usually fine, but you have to watch out for cases where it might be misleading (for example, if ratecontrol gave significantly different distributions of bits between two files).

By the way, the first image is latest x264 with maxed settings, b-adapt 2, and psyrd+psytrellis, and the second is x264 r614 with maxed settings. :p

[MasterNobody]

Dark Shikari
And can you tell the SSIM for this samples?

[smok3]

a little transcode test, besides some funny marks on some faces (posterization enhanced?) looks really good imho;

http://somestuff.org/flashAVC/flvplayer.php?moviename=movies/thespirit-x1280y544.mp4

p.s. useless encoding time.

[Atak_Snajpera]

Yes I agree! Excellent quality @ 1.8 MBps.

[Dark Shikari]

Dark Shikari
And can you tell the SSIM for this samples?I don't have the logs anymore; the SSIM was just about the same for both (negligable difference), around 0.95.
a little transcode test, besides some funny marks on some faces (posterization enhanced?) looks really good imho;

http://somestuff.org/flashAVC/flvplayer.php?moviename=movies/thespirit-x1280y544.mp4

p.s. useless encoding time.:eek:

[Sagittaire]

http://i38.tinypic.com/2edmtte.png http://i33.tinypic.com/312eb2e.png

PSNR 37.165 and 38.152 respectively, 300kbps.

Now stop talking about useless metrics.

Well it's paradoxal here because codec is simply mathematical algo. If you say that the first algo is better than the second algo for eyes you can by definition produce other mathematical algo (metric) for show that ... isn't it?

[jethro]

psy-rd is that 'mathematical algo (metric)', I think

[lexor]

a little transcode test, besides some funny marks on some faces (posterization enhanced?) looks really good imho;

http://somestuff.org/flashAVC/flvplayer.php?moviename=movies/thespirit-x1280y544.mp4

p.s. useless encoding time.

That is awesome (the encode I mean, the movie looks like it'll blow chunks). But did you use tv-pc colour conversion? The blacks don't look quite right, and for web playback the colours should always be converted to pc scale (since we can't have flash player post process in browser).

[lexor]

Well it's paradoxal here because codec is simply mathematical algo. If you say that the first algo is better than the second algo for eyes you can by definition produce other mathematical algo (metric) for show that ... isn't it?

No it isn't. What you want is not the metric that adapts to encoding algorithm, but the metric that simulates your eyes (with algo specific metric, you can't compare algorithms and settings, which is what you want to do). And there is no such metric in existence. Furthermore existence of mathematical objects/concepts is a fundamental problem it is certainly doesn't follow from some "definition" nor can you assume existence most of the time.

[smok3]

lexor, you are right, however i was trying to avoid any filters in the chain (especially if there is a chance of additional posterization by filter itself).
p.s. looks correct with coreavc where input&output are set to auto.

[lexor]

lexor, you are right, however i was trying to avoid any filters in the chain (especially if there is a change of additional posterization by filter itself).

Yeah, that stuff is always a trade off, the only reason I noticed is because wrong colour scale really shows up on my lcd. So I just have a habit for personal use to prioritize colour correction over everything else. The trailer still looks awesome though, I noticed it gets better as it goes to. In the intro, after tree with the birds disappears and the red line appears you can see a quick blocky flash in the bottom half of the frame, never happens again after that. I wonder if that's in the source?

[smok3]

lenox, check the 720p 'original';
http://www.apple.com/trailers/lions_gate/thespirit/

edit: another test, fast 2pass encoding, looks great
http://somestuff.org/flashAVC/flvplayer.php?moviename=movies/wall-e-clip-atwork-x1280y544.mp4

(trailer for the movie 'australia' seems to be a good partybreaker, when trying to encode at sub 2k bitrates, probably due to bunch of dissolves?)

q: if i use the compile 'x264.928.modified.02.exe' then --trellis 2 = psytrellis ?

[kemuri-_9]

q: if i use the compile 'x264.928.modified.02.exe' then --trellis 2 = psytrellis ?

if subme >= 6 & psy-rd > 0 as well (if unspecified, psy-rd is 1.0)

[Dark Shikari]

If psy_rd > 0, subme >= 6, and trellis > 0, psy trellis is on.

The final psy_rd commit will allow psy_trellis without subme >= 6.

[LoRd_MuldeR]

The final psy_rd commit will allow psy_trellis without subme >= 6.

Good news :D

[elguaxo]

Will there be a chance to use PsyRDO without Psy Trellis when commited? Old movies usually have lots of noise and/or grain and don't have many details (probably similar to 300). For the bitrates I'm using PsyRDO seems to be great on every kind of source, but PsyRDO+Psy Trellis is not so great on those old movies unless I use a more bitrate.

edit: I haven't tried using Psy Trellis with a 'lowered' PsyRDO and/or 'lowered' AQ

[Dark Shikari]

Will there be a chance to use PsyRDO without Psy Trellis when commited? Old movies usually have lots of noise and/or grain and don't have many details (probably similar to 300).Huh? Those movies are exactly why I created psy-trellis; because regular trellis sucks with that kind of grain. If even psy-trellis isn't good enough for you, try low deadzones or something.

[elguaxo]

I'll post some samples tomorrow!

[Atak_Snajpera]

edit: another test, fast 2pass encoding, looks great
http://somestuff.org/flashAVC/flvpla...-x1280y544.mp4
Once again you did correct levels!

[wyti]

it will be good if we are able to desactivate psy-trellis when we use psy-rdo.
Because psy-trellis isn't very usefull on anime. He try to conserve details that are invisible (or even not existing at all) and include some ringing (even at high bitrate).

[Dark Shikari]

it will be good if we are able to desactivate psy-trellis when we use psy-rdo.
Because psy-trellis isn't very usefull on anime. He try to conserve details that are invisible (or even not existing at all) and include some ringing (even at high bitrate).Is psy-RD useful on anime at all?

[wyti]

not very, but isn't that one of your goal that psy-rdo will be better on any type of sources ?

[lexor]

Is psy-RD useful on anime at all?
Not on most anime, but some of the newer ones that are broadcast in HD have artificial grain added to them like the movies shot all digital do (not quite as much though). So psy-rd is becoming useful for anime.

[Dark Shikari]

not very, but isn't that one of your goal that psy-rdo will be better on any type of sources ?Ideally, but I haven't found a solution for the "it makes edges in anime less sharp" problem. I can imagine its possible to make a psy-opt for anime (optimize for blurred edges instead of ringing, since the former looks better), but psy-rd isn't it.

I may switch --psy-rd as a parameter though from --psy-rd X to --psy-rd X:Y; this will allow you to tweak both psy-RD and psy-trellis.

[elguaxo]

I may switch --psy-rd as a parameter though from --psy-rd X to --psy-rd X:Y; this will allow you to tweak both psy-RD and psy-trellis.

That sounds perfect! :thanks:

[Ramir Gonzales]

Is there any sort of "Holey bible for the x264 settings) ?

Currently there are so many options, and many exceptions like "Never use option x with option y" that most not-so-knowledged users are completely loosing sight on how to correctly encode using x264 ? :helpful:

[Dark Shikari]

Is there any sort of "Holey bible for the x264 settings) ?

Currently there are so many options, and many exceptions like "Never use option x with option y" that most not-so-knowledged users are completely loosing sight on how to correctly encode using x264 ? :helpful:read this (http://ffmpeg.x264.googlepages.com/mapping)

[LoRd_MuldeR]

Currently there are so many options, and many exceptions like "Never use option x with option y" that most not-so-knowledged users are completely loosing sight on how to correctly encode using x264 ? :helpful:

That's why MeGUI provides various pre-defined profiles from "Fast" up to "Insane" ;)

[fachman]

The whole point of psy-rdo is that what ever IT IS, it SEEMS to be better than what it actually IS => Read : whatever the metrics (good or bad), it looks better.

And what's the matter with "someone else's" opinion you keep talking about : what about YOUR opinion ? What about YOUR eyes ? Forget the others, forget the numbers ... what do YOU see ? Which output do your eyes prefer ? You're putting the responsibility of what you percieve on "someone else", or "something else" (PSNR/SSIM). You seem to be saying "PSNR says it's better, so better it is". Am I wrong somewhere ? I'd be glad you correct me straight away !

Looking at an encode through metrics (which is what you apparently do) is like voluntarily watching a movie without the psy-enhancements. Do you get the idea ? It's voluntarily limiting your perception of the actual picture being displayed.

And yes, there was the ALWAYS word missing (post edited), I thought what I was getting at was obvious but I'll be careful next time :).

Ehhh. you simply do not get it. I would like to talk with you but you live in your own universe.
In this thread it was only one time when I presented my judgement and I have backed it up with the pictures. So please take a look at that post, and then make the proper judgement of my eyes then please stop this nonsence disscussion and focus on something what is real.

To Dark Shikari

I am here only for one reason. To help you improve the codec efficiency. I have done hundreds of encodes on various video files and I strongly belive my experience can be really helpful.
I know what I am talking about because I am talking about my OWN research and I can back it up with numbers and pictures. Please do not take personal for anything I am writing about your work down here, because my only intention is professional advice.
1. With all respect, but you have not used the max setting for r614 because you have not used my FASM technique, because you did not know about it and you still refuse to aknowledge the power of PBRATIO. I want to believe it is only because you are a busy man. If you do not want to make such tests on your own my computer is now computing some tests to back up my words, so please give me some time and I will give you reasonable explanation.
2. What you have presented us (two pictures) is the classic problemm with PSNR discribed fully here http://www.ece.uwaterloo.ca/~z70wang/research/ssim/. To bad you do not remember the exact SSIM values. So metrics are not useless they just need to be correctly interpreted and you can gain those knowledge by experimenting on various video files comparing what you see with what metrics tells you.
For example SSIM seems to be better, but it operates only on Luma so, setting Chroma QP_offset to -15 would provide better SSIM, but lowers PSNR and lowers perceptable picture quality. Because both metrics does not like your PSYRDO I would suggest another one with complexity comparision. I strongly belive that only when we properly balance those 3 metrics we can have something which would be reliable. After all those metrics were not created for anyones fun, but were created to compare quality!!!!
3. There is another thing we must remember here. We are working on video not on still pictures. X264 provides the PSNR and SSIM values which is average of those values of every frame. You have been so kind to correct me to do not use restrain to AQ, but when I use MAXQ=51 as everyone suggest I receive many scenes with so awful quality I can not even look at them, and only some with better quality(I am talking about low bitarate here). I also do receive lower PSNR and SSIM values....
4. If you would be so kind to tell me about your or someone elses experience with XVID and PBRATIO because I am really curious about it.

I am waiting for your response.

[Dark Shikari]

Ehhh. you simply do not get it. I would like to talk with you but you live in your own universe.
In this thread it was only one time when I presented my judgement and I have backed it up with the pictures. So please take a look at that post, and then make the proper judgement of my eyes then please stop this nonsence disscussion and focus on something what is real.If all you care about are metrics, you are wrong, since RDRC is guaranteed to give effectively optimal ratecontrol, and does not agree with your pbratio value.

If you don't care about metrics, and insist your eyes are right and everyone else's are wrong, I have news for you: there is no such thing as "golden eyes."
I know what I am talking about because I am talking about my OWN research and I can back it up with numbers and pictures. Please do not take personal for anything I am writing about your work down here, because my only intention is professional advice.You don't seem very professional to me.
2. What you have presented us (two pictures) is the classic problemm with PSNR discribed fully here http://www.ece.uwaterloo.ca/~z70wang/research/ssim/. To bad you do not remember the exact SSIM values. So metrics are not useless they just need to be correctly interpreted and you can gain those knowledge by experimenting on various video files comparing what you see with what metrics tells you.SSIM is just as useless; its PSNR weighted by variance.3. There is another thing we must remember here. We are working on video not on still pictures. X264 provides the PSNR and SSIM values which is average of those values of every frame.Which is the reason why pbratio=2.1 is a horrible idea. This is why (IMO) we should use Global SSIM and not Average, because Average misleads idiots.You have been so kind to correct me to do not use restrain to AQ, but when I use MAXQ=51 as everyone suggest I receive many scenes with so awful quality I can not even look at them, and only some with better quality(I am talking about low bitarate here). I also do receive lower PSNR and SSIM values....Surprise surprise--when you use broken ratecontrol options, you break ratecontrol!
4. If you would be so kind to tell me about your or someone elses experience with XVID and PBRATIO because I am really curious about it.B-frames in Xvid look awful. Simple as that.

I'm generally open to new ideas, but I don't like people who insist that they are right on absolutely everything, don't listen to a word that anyone else says, and refuse to provide any theoretical backing for any of their claims. As I have stated before, I explicitly refuse to consider any idea which has no theoretical basis; I don't deal in cargo-cult encoding. If you want me to consider your ideas, create a solid theoretical basis for them.

[Ramir Gonzales]

read this (http://ffmpeg.x264.googlepages.com/mapping)
Thanks for the link DS, unfortunately nothing about Psy-RDO and other new options to find there...

[martino]

Thanks for the link DS, unfortunately nothing about Psy-RDO and other new options to find there...
Because those aren't in the source yet... most likely.

[elguaxo]

Ok, here goes my sample. I always like PsyRDO and I like PsyRDO+PsyTrellis, but if you don't have enough bitrate and the source has noise and/or grain and no details then I see very often things like this:

PsyRDO
http://img384.imageshack.us/img384/5642/nopsytrellistn5.png

PsyRDO+PsyTrellis
http://img384.imageshack.us/img384/6405/withpsytrellispk5.png

if I use more bitrate then PsyRDO+PsyTrellis wins hands down.

Here are the encoded samples. Both encoded at 1200kbps (a bit low for this sample) and I used this (http://forum.doom9.org/showpost.php?p=1165193&postcount=615) build for PsyRDO only and this (http://forum.doom9.org/showpost.php?p=1169563&postcount=668) for PsyRDO+PsyTrellis:

http://omploader.org/vb2Fz/x264.928.modified.no.psytrellis.mkv
http://www.mediafire.com/?meok8mwoewk

http://omploader.org/vb2Fx/x264.934.modified.with.psytrellis.mkv
http://www.mediafire.com/?gfuppl5ie1g

[Dark Shikari]

Use a better host like Mediafire;your downloads terminate repeatedly after a couple minutes because the server terminates the connection.

Edit: Nevermind, was able to curl the first bit of them.

I've looked through a few dozen random frames in the first few thousand frames of the video; psy-trellis looks consistently better to me... though I do notice psy-trellis seems to result in slightly smaller B-frames (which could result in a few of them looking worse); this is probably due to the higher quantizer, and thus the harsher skip threshold.

[smok3]

some more examples, targeting 1800 kbps;
http://somestuff.org/flashAVC/flvplayer.php?moviename=movies/terminatorsalvation-x1056y448.mp4 - downscaled, no level correction
http://somestuff.org/flashAVC/flvplayer.php?moviename=movies/eagleeye-x1056y448.mp4 - downscaled, no level correction
http://somestuff.org/flashAVC/flvplayer.php?moviename=movies/theexpress-x1056y448.mp4 (problematic regarding constant quality, probably source not very good) - downscaled, no level correction
http://somestuff.org/flashAVC/flvplayer.php?moviename=movies/australia-x1280y544.mp4 (problematic dissolves?, source no good) - no resizing, no level correction

edit1: Problems seems to have a lot to do with the original as well, since this are all transcodes.
edit2: will try some downscaled 1080p sources, comparing Lanczos and bilinear downscale.

edit3: 1080p source, downscaled, almost perfect (compared to source)
http://somestuff.org/flashAVC/flvplayer.php?moviename=movies/deathrace-LanczosDownscale-x1024y576.mp4

p.s. don't blame me for the dumb movies they made this days....

[IgorC]

Dear IgorC

Although I have no idea how Komisar did his tests, for the future I would consider think twice before you will say something...
?


Also I could bet you do not live with Komisar or Nico to know how they did their tests
It's not secret. http://forum.mediatory.ru/viewtopic.php?f=15&t=3780&start=3800

[Razorholt]

I've heard somewhere that PsyRDO+PsyTrellis were acting in first pass and that Trellis 2 was the primary choice over deadzones.
If this is true, why using Trellis 0 on first pass by default?

Thanks,
- Dan

[Ranguvar]

I am 99% sure that Psy RDO and Psy Trellis are only useful in the second pass.

[Ranguvar]

So since you were already busy DS, i tried tackling this myself and the result is:
x264_psy_rdo.0.5+psy_trellis_01_r929_mod.diff (http://kemuri9.net/dev/x264/patches/x264_psy_rdo.0.5+psy_trellis_01_r929_mod.diff)

added a new --psy-trellis option to enable and control code usage for psy trellis
allowing it to be on/off similar to psy rdo (within its dependencies of trellis > 0 and psy-rd > 0)

of course this needs your review when you have the time to make sure i didn't fubar anything.
I'll be adding this to my compiles. Quoting because it was buried by the metrics debate, in hope that someone will verify it doesn't b0rk anything :)

Thanks!

[Dark Shikari]

I'll be adding this to my compiles. Quoting because it was buried by the metrics debate, in hope that someone will verify it doesn't b0rk anything :)

Thanks!Try this instead (http://pastebin.com/f1685f9f7)...

[gigah72]

Try this instead (http://pastebin.com/f1685f9f7)...

patch failed :confused:

Administrator@SDFBF-9RX23LR8A /c/x264
$ patch -p1 < x264_psyrd_14082008.diff
patching file `common/common.c'
patching file `common/common.h'
patching file `common/dct.h'
patching file `encoder/analyse.c'
Hunk #2 succeeded at 1069 (offset 1 line).
Hunk #4 succeeded at 2121 (offset 1 line).
Hunk #6 succeeded at 2649 (offset 1 line).
patching file `encoder/encoder.c'
Hunk #1 FAILED at 411.
Hunk #2 succeeded at 484 (offset 1 line).
1 out of 2 hunks FAILED -- saving rejects to encoder/encoder.c.rej
patching file `encoder/macroblock.c'
patching file `encoder/macroblock.h'
patching file `encoder/rdo.c'
Hunk #2 succeeded at 326 (offset 2 lines).
Hunk #4 succeeded at 487 (offset 2 lines).
patching file `x264.c'
patching file `x264.h'

[Razorholt]

I am 99% sure that Psy RDO and Psy Trellis are only useful in the second pass.


You're right. That's the b-frames decisions that are made in first pass :)


Thanks Ranguvar.

- Dan

[Dark Shikari]

patch failed :confused:

Administrator@SDFBF-9RX23LR8A /c/x264
$ patch -p1 < x264_psyrd_14082008.diff
patching file `common/common.c'
patching file `common/common.h'
patching file `common/dct.h'
patching file `encoder/analyse.c'
Hunk #2 succeeded at 1069 (offset 1 line).
Hunk #4 succeeded at 2121 (offset 1 line).
Hunk #6 succeeded at 2649 (offset 1 line).
patching file `encoder/encoder.c'
Hunk #1 FAILED at 411.
Hunk #2 succeeded at 484 (offset 1 line).
1 out of 2 hunks FAILED -- saving rejects to encoder/encoder.c.rej
patching file `encoder/macroblock.c'
patching file `encoder/macroblock.h'
patching file `encoder/rdo.c'
Hunk #2 succeeded at 326 (offset 2 lines).
Hunk #4 succeeded at 487 (offset 2 lines).
patching file `x264.c'
patching file `x264.h'Bah, I even have to rebase my patches for you...

Linkage (http://pastebin.com/m3fb0df)

[kemuri-_9]

ha ha ha, the r929 snafu came up again.

hmm... so now the accepted formats for --psy-rd are as follows:
psy-rd strength & psy trellis strength at separate strengths,
float,float (i.e. 1.0,1.0)
float:float (i.e. 1.0:1.0)

both at same strengths
float (i.e. 1.0)

first one taking psy-rd strength, and second for psy trellis strength.

or did i make a mistake somewhere?

[Dark Shikari]

ha ha ha, the r929 snafu came up again.

hmm... so now the accepted formats for --psy-rd are as follows:
psy-rd strength & psy trellis strength at separate strengths,
float,float (i.e. 1.0,1.0)
float:float (i.e. 1.0:1.0)

both at same strengths
float (i.e. 1.0)

first one taking psy-rd strength, and second for psy trellis strength.

or did i make a mistake somewhere?Correct, same as --deblock.

[fachman]

If all you care about are metrics, you are wrong, since RDRC is guaranteed to give effectively optimal ratecontrol, and does not agree with your pbratio value.

I do not care ALL about metrics, I just disagree with you that metrics are useless.
PSNR and SSIM have been created as a tool for comparing quality. The people which created it have made many tests and in especially of SSIm compared their metrisc to Perceptual Quality which was all described in the article.
Let me give you the example. The knife is a tool used to cut things. It works pretty well in most cases but you have found that it can not cut metal and insist on that example to saying it is USELESS.

If you don't care about metrics, and insist your eyes are right and everyone else's are wrong, I have news for you: there is no such thing as "golden eyes."

As far this nonsence disscussion goes we are going deeper into imagination not into what is real. At the begining of mine conversation I have presented two pictures comparing PSYRDO and PSYRDO+TRELLIS claiming that PSYRDO+TRELLIS prodcued in this particular case worse results. And you know what??? So far noone here commented their quality, no one dares to say that PSYRDO+PSYTRELLIS produced better result, but instead I was blamed for faulty vision. Maybe I have faulty vision, so I need to ask you which one in your personal opinion from those two pictures I have presented, produced better quality????

You don't seem very professional to me.

Well because I did not show you anything yet :)

SSIM is just as useless; its PSNR weighted by variance.Which is the reason why pbratio=2.1 is a horrible idea. This is why (IMO) we should use Global SSIM and not Average, because Average misleads idiots.

It is good to see that you start to take back your words regarding metrics. :)
I have no idea what kind of SSIM produces x264, because there is only one, but Global PSNR is also a form of "average" metrics just computed with different math.

Surprise surprise--when you use broken ratecontrol options, you break ratecontrol!B-frames in Xvid look awful. Simple as that.

I have no idea what do you mean by broken rate control. If you mean that the bitrate is different that it should be, this is not true and I have posted statistics to prove it. If you mean that there is huge variance of bitrate between the frames, I will say that is the reason ABR exists.
Well B-frames in X264 with PBRATIO=2.1 look really nice and the whole video look MUCH NICER than with PBRATIO=1.3 without AQ and in case of AQ it is a long story. This is the statistics for PBRATIO=1.3

--[NoImage] Job commandline: "C:\Program Files (x86)\megui\tools\x264\x264.exe" --pass 2 --bitrate 655 --stats "05.stats" --level 3.1 --keyint 99999 --ref 4 --mixed-refs --no-fast-pskip --bframes 16 --b-rdo --bime --weightb --direct auto --subme 7 --trellis 2 --partitions p8x8,b8x8,i4x4,i8x8 --8x8dct --qpmin 24 --qpmax 35 --qpstep 13 --ratetol 100 --qcomp 0 --cplxblur 2 --scenecut 100 --me umh --merange 64 --threads auto --thread-input --progress --output
--[Information] [2008-08-11 06:48:44] Encoding started
--[NoImage] Standard output stream:
--[NoImage] Standard error stream
---[NoImage] avis [info]: 1280x720 @ 25.00 fps (183160 frames)
---[NoImage] x264 [info]: using cpu capabilities: MMX2 SSE2Fast SSSE3 Cache64
---[NoImage] x264 [info]: cabac=1 ref=4 deblock=1:0:0 analyse=0x3:0x113 me=umh subme=7 psy_rd=1.000000 brdo=1 mixed_ref=1 me_range=64 chroma_me=1 trellis=2 8x8dct=1 cqm=0 deadzone=21,11 chroma_qp_offset=0 threads=6 nr=0 decimate=1 mbaff=0 bframes=16 b_pyramid=0 b_adapt=1 b_bias=0 direct=3 wpredb=1 bime=1 keyint=99999 keyint_min=25 scenecut=100(pre) rc=2pass bitrate=655 ratetol=100.0 rceq='blurCplx^(1-qComp)' qcomp=0.00 qpmin=24 qpmax=35 qpstep=13 cplxblur=2.0 qblur=0.5 ip_ratio=1.40 pb_ratio=1.30 aq=3:11:1.00
---[NoImage] mp4 [info]: initial delay 1 (scale 25)
---[NoImage]
---[NoImage] x264 [info]: slice I:319 Avg QP:26.58 size: 13058 PSNR Mean Y:45.86 U:49.41 V:49.79 Avg:46.72 Global:45.48
---[NoImage] x264 [info]: slice P:82918 Avg QP:28.85 size: 5284 PSNR Mean Y:43.59 U:47.41 V:47.75 Avg:44.48 Global:43.20
---[NoImage] x264 [info]: slice B:99923 Avg QP:30.37 size: 1530 PSNR Mean Y:42.64 U:46.99 V:47.33 Avg:43.62 Global:42.49
---[NoImage] x264 [info]: consecutive B-frames: 19.4% 25.2% 12.9% 17.5% 18.4% 5.4% 0.4% 0.0% 0.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.6%
---[NoImage] x264 [info]: mb I I16..4: 53.8% 44.0% 2.2%
---[NoImage] x264 [info]: mb P I16..4: 9.0% 4.7% 0.1% P16..4: 38.2% 2.7% 3.1% 0.0% 0.0% skip:42.1%
---[NoImage] x264 [info]: mb B I16..4: 0.1% 0.2% 0.0% B16..8: 25.4% 0.2% 0.3% direct: 0.4% skip:73.4% L0:31.0% L1:67.6% BI: 1.4%
---[NoImage] x264 [info]: final AQ sensitivity: 8.0000
---[NoImage] x264 [info]: 8x8 transform intra:34.9% inter:90.3%
---[NoImage] x264 [info]: direct mvs spatial:95.6% temporal:4.4%
---[NoImage] x264 [info]: ref P L0 69.9% 16.5% 10.0% 3.6%
---[NoImage] x264 [info]: ref B L0 90.4% 6.0% 3.6%
---[NoImage] x264 [info]: SSIM Mean Y:0.9795424
---[NoImage] x264 [info]: PSNR Mean Y:43.077 U:47.183 V:47.522 Avg:44.011 Global:42.801 kb/s:649.91
---[NoImage] encoded 183160 frames, 12.25 fps, 649.93 kb/s

Take a good look what happened with SSIM.

I'm generally open to new ideas, but I don't like people who insist that they are right on absolutely everything, don't listen to a word that anyone else says, and refuse to provide any theoretical backing for any of their claims. As I have stated before, I explicitly refuse to consider any idea which has no theoretical basis; I don't deal in cargo-cult encoding. If you want me to consider your ideas, create a solid theoretical basis for them.

Dark Shikari, I have given you theorethical backing of my claims in the very begining. If it is not enough for you, I will do my best to create better.
Also thank you very much for honesty. I also will be honest with you. I do not like people who put the words into someone elses mouth which they did not say.
So far the disscussion on this forum is at scandalic unprofessional level. I have found something and instead of trying to find a reasonable explanation I am bullied here.
Also I do not belive in majority opinions in every case because as the reality shows (this thread also), many many people rely more on their emotions than on the reality. If you do not belive me then ask yourself question why majority people in your country elected George W Bush for two terms??? (I will not continue this topic so far because it is not a proper place for this. I just wanted to explain my beliefs)

[Dark Shikari]

So far the disscussion on this forum is at scandalic unprofessional level.And this is why I am not going to read any more of your posts.

Also, "it works" or "it increases quality" is not a theoretical basis; its a circular argument. For example, let me give you a theoretical explanation for your --pbratio 2.1, since you don't seem to be capable of (or care to) come up with one on your own:

"Biprediction allows considerably better quality even with no residual, so a much higher quantizer doesn't hurt quality much and allows considerably more bits to be saved by increasing the number of skip macroblocks."
(Reason it fails: too high a quantizer means not even enough bits for motion vectors, killing the effectiveness of biprediction, too)

or

"If you drop quality every other frame, the human eye won't notice it much, especially if with biprediction blurring the difference between the two neighboring frames."
(Reason it fails: you're using more than 1 B-frame)

Of course, neither of these are valid explanations for such a high pbratio (they work rather well for explaining the current ~1.4 pbratio), but they're a good start.

[nerdpunk]

Bah, I even have to rebase my patches for you...

Linkage (http://pastebin.com/m3fb0df)

for us all ;) thanx, find them very useful... but... is there a public repository where we can follow the devel process?