QLB matrix update - 1.2

[dragongodz]

this is QLB matrix version 1.1. a minor update. find the 1.2 update further down the page.

for those who dont know the Quenc Lower Bitrate(QLB) matrix was designed to retain better quality with mid to lower bitrate encoding with the QuEnc mpeg1/2 encoder.

i would appreciate any feedback to any quality differences you can see compared to the original QLB matrix. thanks.

Intra
8 16 19 22 26 27 29 34
16 16 22 24 27 29 34 37
19 22 26 27 29 34 37 39
22 22 26 27 29 34 38 45
22 26 27 29 32 36 42 52
26 27 29 32 36 40 52 63
26 27 29 35 40 50 61 77
27 29 35 40 50 61 75 89

Non-Intra
16 17 18 19 20 21 22 23
17 18 19 20 21 22 23 25
18 19 20 21 22 23 24 26
19 20 21 22 23 24 26 30
20 21 22 23 25 26 28 31
21 22 23 24 26 28 31 35
22 23 24 26 28 29 33 36
23 24 25 28 29 33 34 40

[danpos]

@dragongodz

I'll do some tests with the up-to-date QLB matrix and let you know about them ...

THX,

[dragongodz]

thanks danpos. i just basically need to know if there is any noticable quality loss with the new version and if there is then how bad and how hard is it to pick(need to examine still frames carefully or seeable playing ? that type of thing).

[SiXXGuNNZ]

updated, thanks

[Ebobtron]

@dragongodz

Do I understand correctly that QLB stands for quality low bit-rate? If I am correct; what bit rates are you talking about? 1500 2000 Also, is there a higher bit rate at which you wouldn’t expect any improvement.

Thanks,

[dragongodz]

Do I understand correctly that QLB stands for quality low bit-rate?
no i said what it stands for in the first post
Quenc Lower Bitrate(QLB)
and its aim
designed to retain better quality with mid to lower bitrate encoding with the QuEnc mpeg1/2 encoder.
meaning it should try to not block or smooth or whatever as much as using no custom matrix would using QuEnc with lower bitrates.

what bit rates are you talking about? 1500 2000 Also, is there a higher bit rate at which you wouldn’t expect any improvement.
that i cant really answer. the reason being the myriad of different types of content. clean sources do not compress the same as noisy ones, same with high motion compared to still motion etc.
for the sake of arguement lets say that once you start to go under 4000kbps for dvd resolution you MAY start to see a difference depending on the source, but dont hold me to that. :D

[Ebobtron]

@dragongodz
Wow, I am embarrassed. Thank you for the answer.
Can you bear with me for another question concerning matrixes?

I can not see any difference between QuEnc’s standard matrix and QLB over 1800 frames no matter how I setup QnEnc. I can measure the difference but the difference is tiny. If the result is a little smoother or sharper I can’t tell.

Can we remove visual perception from the discussion and focus on encoding accuracy instead?

Which will a matrix affect more, compressibility or encoding accuracy?

I can foresee where a matrix, ill suited for a particular source might trash the accuracy and improve compressibility. Therefore, am I answering my question? Is it strictly a compromise like the rest of the encoding process?

To use a common vernacular; I’m trying to learn to drive the car. I don’t really care what is under the hood. It’s more like: what does that button do?

Thanks again,

[dragongodz]

Can we remove visual perception from the discussion and focus on encoding accuracy instead?
not really since this is the most important factor. i already know what QLB does compressability wise compared to no custom matrix. so whether you can see any difference or quality loss or gain(compared to no custom matrix at lower bitrates) is what i need to know. for this case its really comparing the update QLB to old QLB since i know the old QLB could lighten very fine detail sometimes. you usually have to freeze the frames and compare though to see it which is good. :)

Which will a matrix affect more, compressibility or encoding accuracy?
depends on the matrix. you will find that most low or lower bitrate matrices try to increase the compressability. QLB tries to do this while trying to sacrafice only a tiny amount of detail, which should be hard for you to notice if all is working well.

I can not see any difference between QuEnc’s standard matrix and QLB over 1800 frames no matter how I setup QnEnc. I can measure the difference but the difference is tiny. If the result is a little smoother or sharper I can’t tell.
well to see things easiest you have to be encoding at the point where you can see QuEncs output starting to show not the best results such as blocks or artifacts etc. the differences should be easier to see then as i said. so you may want to lower your tartget bitrate more.


anyone else have any impressions ? especially compared to old QLB ? i would like to know before any new versions of encoders are released whether its ok for the updated version to be used or if i should kill this update off.

[Ebobtron]

@dragongodz

Thanks,

[freelock7]

Flat parts of the picture are too smooth with some details lost.
I prefer the older.

[dragongodz]

Flat parts of the picture are too smooth with some details lost.
thanks for the response. i will do a new matrix tonight which should be a midish point between the 2. really closer to QLB 1.0 but with a tiny tweak or 2.

cant at the moment as some bastard decided to steal our car ,yes on good friday eve. now the cops have found it run up a concrete edging with the windows smashed with bricks, roof stomped in, tyres slashed etc. looks like a right-off. great frelling easter huh ?

[onesoul]

Originally posted by dragongodz
cant at the moment as some bastard decided to steal our car ,yes on good friday eve. now the cops have found it run up a concrete edging with the windows smashed with bricks, roof stomped in, tyres slashed etc. looks like a right-off. great frelling easter huh ? Sorry for hearing that :(

[Mug Funky]

that sucks, man (about the car i mean...). had that happen tp my old shitbox a couple years ago. hated the car, but i had all my CDs in there plus a nice camera flash i haven't found a good replacement for yet :(

well, have a nice easter anyhoo. eat lots of chocolate :)

btw, what sources are you wanting to test this on? my machine's too slow to do much denoising, but i believe an encoder should know what to do with noise anyway. i can also run a super-clean, but field-blended anime through it (azumanga 5... 6 will be out in may, just FYI :))

the main thing i use QuEnc for is noisy interlaced DV of random bands playing, just for home projects. to be honest, with that sort of crappy material to work with, the matrix isn't going to make much difference (QuEnc does a superb job on this stuff anyway - for bitrates < 4.5mbps it kicks the bollocks off pro hardware encoders, too).

[ernstblaauw]

Is this matrix worth trying with HC or CCE? Or can you say that this matrix is heavily tuned for QuEnc and thus not worth testing on HC and CCE?

[dragongodz]

ok heres QLB matrix 1.2(beta?) so people can test. its only a tiny mod to the original so should only maybe help a tiny bit more with real low bitrates but in general be practically the same as the original QLB. :)

INTRA
8 16 19 22 26 27 29 34
16 16 22 24 27 29 34 37
19 22 26 27 29 34 37 39
22 22 26 27 29 34 38 42
22 26 27 29 32 36 40 50
26 27 29 32 36 40 50 61
26 27 29 35 40 50 59 75
27 29 35 40 50 59 75 89

NON-INTRA
16 17 18 19 20 21 22 23
17 18 19 20 21 22 23 25
18 19 20 21 22 23 24 26
19 20 21 22 23 24 26 28
20 21 22 23 25 26 28 29
21 22 23 24 26 28 29 31
22 23 24 26 28 29 31 34
23 24 25 28 29 31 34 38

what sources are you wanting to test this on?
basically anything that you start to see a bit of blocking etc start to happen. though this update and the original QLB should look very similar in 99% of cases. :)

Is this matrix worth trying with HC or CCE?
HC already has the original QLB matrix and others have said QLB works fine with CCE aswell.

What part of Australia are you in?
Hobart, Tas
cops said 5 other cars were also taken from the area in the same night. 2 ended in the river, 2 near the river in bushes, 1 close to a different river(ours) and 1 driven in to a picnic hut.
I wish i had insurance since i wont be replacing it real quick on the dole... :/
we have insurance but since its an older car(my wife has had it for 18 years) we will get bugger all(payout based on market value) even though we have had the motor done not 12 months ago, 4 new tyres less than 6 months ago, completely serviced(new sparks, fan belt etc) 3 weeks ago, etc. so what we will get back would basically buy us a sh*tbox. instead we are going to get a loan for a half decent(not top of the line) second hand car(late 90's model) and be in more debt for the next 7 or 8 years. :(

[Fishman0919]

@dragongodz:

I'm just starting to play around with the 2nd part of my "Blind Test". I've got three TV's at work setup next to each other. I'll try to do a test between the old, 1.1 and 1.2 and see what happens.


PS - This with some better good timing in your part, I just started and was gonna do some encoding with QuEnc tonite

[dragongodz]

I'm just starting to play around with the 2nd part of my "Blind Test". I've got three TV's at work setup next to each other. I'll try to do a test between the old, 1.1 and 1.2 and see what happens.
nice. the first test you did with encoder versus transcoder was interesting reading but left people wanting to know more(including me). so i cant wait to see how your new tests go. :D

[dragongodz]

anyone, anyone, anyone ????? :confused:

[freelock7]

Encoded a clip at 2200 kbs, VBR2Pass with CCE2.67, HC0.12 and QuEnc059b2.
This one is much better a little bit smoother than original QLB but details remains good.
CCE and especially HC take advantage of this matrix: grain is removed in part.
QuEnc picture seems better with the older.

Have a nice easter although.

[dragongodz]

freelock7 - how noticable was the difference with QuEnc ? that is was it easy to see playing or hard or did you have to look at still frames etc ?

[freelock7]

It's a little bit blurred (almost imperceptible).
The QuEnc problem at very low bitrate comes from macroblocs in fast scenes not from the picture quality-witch remains out of the ordinary very good (no grain).
I don't know if a particular matrix could correct it (?).

[ernstblaauw]

Originally posted by freelock7
Encoded a clip at 2200 kbs, VBR2Pass with CCE2.67, HC0.12 and QuEnc059b2.
This one is much better a little bit smoother than original QLB but details remains good.
CCE and especially HC take advantage of this matrix: grain is removed in part.
QuEnc picture seems better with the older.

Have a nice easter although.
Do you prefer this new QLB also abobe the standard matrix with HC?

[freelock7]

Yes indeed. At low bitrate, this matrix is useful for HC.

[dragongodz]

It's a little bit blurred (almost imperceptible).
i assume you mean you can see this while playing and not just looking at single frames ?
almost imperceptible sounds good though becuase that should mean people would not see any difference if just watching the movie. :)

I don't know if a particular matrix could correct it (?).
no macroblocks with fast scenes could be fixed by making that as compressable as possible(blur the crap out of it in other words :)). then of course when you went to a still scene if you used the same matrix it would lose all detail and look terrible. the only way you could do it is to use different matrices for different scenes. the encoder would have to then decide which it thinks is needed and use it. even then you may not agree with its decisions. ;)
so no, no 1 matrix will be perfect for all cases.

anyone else done any testing ? i GREATLY appreciate freelock7 posting what he sees but i would like some more results aswell to get a broader idea.

[freelock7]

dc decision=8,GOP=12 reduce considerably macroblocs in QuEnc (all my tests before used DC=9 and GOP=15)
It is not so perceptible in HC.

[Fishman0919]

Well I did abit of testing with QLB 1.0, 1.1 and 1.2 at work. It's been a little busy over the last couple of days and I couldn't keep the 3 TV's next to each other to long so I couldn't play to much.

The test was done with 3 clips from the first test,

Matrix Revolutions - 1:41:48 - 1:42:08
Star Wars - Phantom Menace - 2:01:14 - 2:01:30
Titan AE - 5:52 - 6:11

Each clip was encoded with Quenc .59b2 @ 2000,2500 and 3000 kbits/sec with each different matrix. Then I made a DVD for each matrix with the 9 clips.

Viewing Equipment:

TV: 3 Mitsubishi WS55315 55" HD Projection TV set next to each other
DVD Player: 3 Denon DVD1910 DVD Player
Cables: 3 1m M1000CV Monster Cable

Results:
I only got 5 for each one, I'll try to get more soon but it's hard right now.

Matrix:

2000 kbit/sec
-------------
1.0 - 1.1 - 1.2
3 - 1 - 2
3 - 1 - 2
3 - 2 - 1
3 - 1 - 2
3 - 2 - 1

2500 kbits/sec
--------------
1.0 - 1.1 - 1.2
2 - 3 - 1
2 - 3 - 1
2 - 3 - 1
1 - 3 - 2
2 - 3 - 1

3000 kbits/sec
--------------
1.0 - 1.1 - 1.2
1 - 3 - 2
2 - 3 - 1
2 - 3 - 1
1 - 2 - 2
2 - 3 - 1

Star Wars:

2000 kbit/sec
-------------
1.0 - 1.1 - 1.2
3 - 2 - 1
3 - 1 - 2
2 - 1 - 3
3 - 1 - 2
3 - 2 - 1

2500 kbits/sec
--------------
1.0 - 1.1 - 1.2
1 - 3 - 2
2 - 3 - 1
2 - 3 - 1
3 - 2 - 1
1 - 3 - 2
2 - 3 - 1

3000 kbits/sec
--------------
1.0 - 1.1 - 1.2
2 - 3 - 1
2 - 3 - 1
1 - 3 - 2
2 - 3 - 1
2 - 3 - 1

Titan AE:

2000 kbit/sec
-------------
1.0 - 1.1 - 1.2
3 - 1 - 2
3 - 1 - 2
2 - 1 - 3
3 - 1 - 2
3 - 1 - 2

2500 kbits/sec
--------------
1.0 - 1.1 - 1.2
3 - 1 - 2
1 - 2 - 3
3 - 1 - 2
2 - 3 - 1
3 - 1 - 2

3000 kbits/sec
--------------
1.0 - 1.1 - 1.2
3 - 1 - 2
2 - 1 - 3
3 - 2 - 1
3 - 1 - 2
3 - 1 - 2

Edit: link to first test

http://forum.doom9.org/showthread.php?s=&threadid=91418

[onesoul]

-Fishman0919

Interesting test. While QLB 1.1 seems to be the weaker at Matrix and Star Wars tests (except at the 2000kb/s bitrate), it outcomes a clear winner at Titan movie.
Analyzing your results, I would stick with QLB 1.2 which seems the better compromise.

[Fishman0919]

QLB 1.1 was smoother then the other two, most people said it looked "Too plain". I think that help with the 2000 kbits/sec and with Titan AE being that it is CG.

[dragongodz]

Fishman0919 - thanks, interesting results. QLB 1.2 does indeed seem a good compromise. gaining better general scores for 2000Kbit/s and going toe to toe at the higher bitrates against 1.0.

so i think i will make QLB 1.2 the new official version. :D

BIG thanks to both freelock7 and Fishman0919 for testing these out.

[Nic]

Cool dgz, haven't had a chance to answer your PM. Will include in new QuEnc soon. Very busy at present :( But have updated QuEnc, because I changed the way aspect ratio is done, so will release soon.
(Try encoding a clip with 768x576 at 4:3 aspect and you'll see what I mean)

-Nic