CQMs Compressibility TEST [Archive]

View Full Version : CQMs Compressibility TEST

SpikeSpiegel

25th April 2005, 11:39

I noticed that there's still a little bit confusion about CQMs and, as well, I've seen too many times some speeches like "hi/low bitrate matrix", etc..
I don't know if anybody has alredy done something like that, but I think that it's truly necessary:

I don't want to make a quality test (for, thanks to Soulhunter, there's no more need of it), but I want to make a clear, objective test about compressibility, in order to classify the most popular CQMs with this parameter.
...then, if it's absolutely unavoidable (for these terms are quite common even if they are nonsense), they can be categorized as "LOW - MEDIUM - HI BITRATE CQM".

HOW:
I suggest to encode a part of a popular movie with the same parameters / codec version, with a single pass at a fixed quant, then, the matrices can be classified according to the filesize. Settings could be:

Script:

Mpeg2Source("C:/.../vts.d2v")
SelectRangeEvery(140,14) --> =10% of the movie

XVID 1.1.0-Beta2 Settings:

Single Pass (Quant=2), Quarter Pxl, B-VOPs (default settings), MotionSearchPrecision=6, VHQ=Off, Chroma Motion, Trellis.

And the movie...well, I think that nothing is more suited for this test than...THE MATRIX!

I can test only a few matrices, for my CPU is very low, then I need your help!

Let me know what do you think about it.

Greetings!

Soulhunter

25th April 2005, 12:21

SpikeSpiegel

26th April 2005, 12:18

Originally posted by Soulhunter
Hi Spike, nice to see ya here again... :)

In my 1st CQM comparison I've already done something like this!

Unfortunately its no more up to date coz a lot new matrices were developed...
.
.
.
Maybe you can use it somehow for ya test... ^^

Bye

Thank you very much!

Your test on the "Lobby Shootout" chapter can be used as a base for this one; I've just done the same test with XviD 1.1b2 (all options disabled, default parameters, MotionSP=6, VHQ=1) and results are quite similiar:

XviD v.1 RC4 (your test) = 130.297.856 Bytes
XviD 1.1b2 (chroma motion enabled) = 129.046.528 Bytes
XviD 1.1b2 (chroma motion disabled) = 132.229.120 Bytes

Even if this is not a quality test, it would be nice to add the PSNR info too, but now I want to hear some opinions about it.

Which XviD setting can point out a CQM's "behaviour"?
Maybe all options can be disabled (BVOPs included) except QuarterPixel that could underline the precision of the HD ones and Trellis & Chroma Motion (that are generally used)....but what about VHQ?

Sharktooth

26th April 2005, 13:52

B-frames shouldnt be disabled coz they reflect the matrix inter-frame quantization.
VHQ, QPEL and all other options are not significative but should be the same for every encoding.

Didée

26th April 2005, 14:04

Originally posted by SpikeSpiegel
I've seen too many times some speeches like "hi/low bitrate matrix", etc. [...] then, if it's absolutely unavoidable (for these terms are quite common even if they are nonsense), they can be categorized as "LOW - MEDIUM - HI BITRATE CQM".
Hoho! and huhu!

Agreed that the terms hi-/lo-bitrate matrix might not be too well defined. But saying that these terms were nonsense is ... nonsense ;)

A high bitrate matrix will

- at any given quant, produce a clip with higher bitrate than when using a standard matrix at the same quant

and should

- at any given quantization, preserve more of high frequencies during DCT than a standard matrix would.

More or less vice versa for "low bitrate" matrices.

And then

but I want to make a clear, objective test about compressibility, in order to classify the most popular CQMs with this parameter
*That* is what I would call nonsense, if at all. A quantization matrix does not have such a thing like compressability. In particular, following your suggestion
I suggest to encode a part of a popular movie with [...] a single pass at a fixed quant, then, the matrices can be classified according to the filesize will tell nothing about "compressability", but only about the resulting size at a fixed quant. Which, again, allows to qualify the different matrices as high, medium or low bitrate.

Sorry for ranting, but it wasn't me who started out with the term of "nonsense".

Teegedeck

26th April 2005, 17:37

Well it would be nice if we could come up with a CQM naming-convention more apt than this 'high bitrate' and 'low bitrate' thing. Saying at least 'low compression' and 'high compression' instead would be a good start. And then there is the 'source-factor' of course. Sharp source, blurry source, noisy source. And then there is the 'taste' factor. Ahhh...

IMHO it would be very helpful if CQMs were described with attributes that indicate how to use a specific matrix to decent results, like (A) with what kind of sources preferably or with what kind of sources not at all, (B) at which levels of compression preferably or outside which range of compression use voids warranty , and (C) which characteristics - if any - of the original's look they might stress or alter (--> taste). It could perhaps facilitate the choice for most people.

That would result in some staggeringly horribly named matrices like 'Mr. X's clean-source-only, high-compression (downsize up to 30% of original!), smooth matrix' or 'Mr. Y's general-usage, medium-to-low-compression, neutral matrix' or whatever else. :D

Or perhaps authors of CQMS could at least, no: even better!, give some recommendation in brackets like '(clean src. only; 60-70% pref.; sharp-ish)'. Then anyone who fancies to do a matrix test would have some senseful lead as to 'how?' (rate by 'best subjective quality' - the irreplaceable subjective factor - at a given level of compression - the irreplaceable objective factor - and call the result 'efficient', perhaps) and 'what field of competitors?'.

And the other way round, CQM authors could protect their work from being judged unfairly; 'if you test my matrix at 50% compressibility with a noisy source, it's your own fault that you don't like the results - because it is way out of that matrix' specs!' or 'the name of my matrix clearly states 'smooth', so don't complain about that in your HDTV encoding test. Don't use it if you don't like it!' Nothing more bothering than unfair criticism, after all, don't you think so? :)

SpikeSpiegel

26th April 2005, 17:56

@ Sharktooth

Understood, I can't use Soulhunter's test...:(
Then let's use default settings of XviD 1.1b2 and quant=2, ok?
One chapter or SelectRangeEvery(X,Y)?

@ Didée

I've never seen so many misunderstandings in the same post:

1) Talking about CQMs, when I say "compressibility", I mean the compressibility of a determinate video that must be encoded with determinate CQM & other settings, IN THE SAME WAY AS, when everybody talk about "bitrate", they mean the average bitrate that is necessary to encode properly a determinate video with a determinate CQM & other settings or (better, but rare) the average bitrate of a file encoded at a fixed quant with these parameters.

2) Then, I'm not saying that "compressibility is a CQM feature", but I think that, compared to "bitrate", the term "compressibility" indicates more clearly the CQM's peculiarity that I want analyze in this test and creates less misunderstandings...LESS but not NONE, as I can see... ;-) .

3):
Originally posted by Didée
...will tell nothing about "compressability", but only about the resulting size at a fixed quant.
...are you sure? Compressibility is almost the percentile ratio between the "wanted" filesize and the size of the same video encoded with the same parameters @ a fixed quant: If this one is higher, the compressibility is lower and vice-versa.

@ Teegedeck

I agree, that's why I'm trying to test (at least) compression level (...I should have used this term...), but, maybe, this is the only objective parameter...the description (and measurement) of other ones (sharpness, definition, etc.) is quite harder

Sharktooth

26th April 2005, 18:15

@SpikeSpiegel: default settings and SelectRangeEvery(X,Y) are ok as long as the clip is long enaugh to draw conclusions on filesizes.

*.mp4 guy

27th April 2005, 09:15

Well there is a coralation between filesize@Q2 and compressability it is not so cut and dry as to be "matrix A > matrix B @Q2 so it must always be less compressable" depending on the source and the matrix Q2 can be much larger or smaller then "normal". For example take a "medium bitrate cqm" and a "high bitrate cqm" and use them on a noisy source, most likely the hi bitrate one will be much larger then it usually is, but if it was encoded at say Q4 it would have looked the same but be half as big.

The matrix classification seems like quite a good idea to me. Ill try and make an example.

Hybrid8AQ[sharp sources preferably with no oversharpening][medium bitrate][smooth, but hopfully still detailed][will not preserve film grain][not for blurry or low detail sources][most common artifacts are blurring and light ringing, low occurence of blocks]

SpikeSpiegel

28th April 2005, 22:09

Mpeg2Source("X:/.../vts.d2v")
Trim(0,186239)
SelectRangeEvery(776,25)
Crop(0,72,0,-72,align=true)

What do you think about this script for the test?
The clip is 4min long (no ending credits), it should be enough.

I've done a quick test with XviD 1.1b2, defaults/SinglePass-quant=2/CQM=H.263:

Size: 65.038.628 bytes
PSNR (min/average/max/overall): 41.3744/47.4218/106.0693/46.8505

There's only a little problem:
Unlike VirtualDud v.1.6.5, VirtalDubMod (last version/bugfix) cuts the last frame :confused: : does it happen only on my pc?

Sharktooth

29th April 2005, 00:26

It's a bug in Vdubmod.
However h.263 is not a matrix :)

SpikeSpiegel

29th April 2005, 12:41

Originally posted by Sharktooth
It's a bug in Vdubmod.
Ok, then I will use VD v.1.6.5
Originally posted by Sharktooth
However h.263 is not a matrix :)
Ok, but it's only a quick test to check parameters, if results are coherent (similiar to othe tests) and indicative, they will be used to test custom matrices.

By the way, the same test with MPEG:

Size: 75.128.874 bytes
PSNR (min/average/max/overall): 41.2397 / 47.1377 / 106.0693 / 46.6029

The size difference is smaller than the one found in the last Soulhunter's test (but this may depend on the source..), while average PSNR difference is almost the same (-0,1%)

pwh04

30th April 2005, 00:30

Thanks, look forward to an updated comparison!

SpikeSpiegel

1st May 2005, 11:12

CQM NAME / SIZE (bytes)

Acailas/113.844.640
Angel_BestLow/73.770.628
Angel_BestVeryLow/72.737.116
Audionut's/93.306.298
Audionut's_HQ/97.916.076
Bach1/66.762.710
Bulletproof's_Heavy_Comp/52.262.766
Bulletproof's_HQ_Matrix/72.008.568
CCE_Smooth_(CG_etc.)/87.668.248
CCE_Ultra_low_bitrate/67.167.056
CCE_Very_low_bitrate/72.784.918
Didees_SixOfNine/117.857.938
Didees_SixOfNine_HVS/115.641.710
eqm_autogk_sharp/82.901.156
eqm_v1/84.002.012
eqm_v3ehr/201.128.908
eqm_v3hr/88.475.138
eqm_v3lr/70.287.336
eqm_v3uhr/200.551.444
eqm_v3uhr_rev2/148.112.456
eqm_v3ulr/59.568.398
eqm_v3ulr_rev2/62.130.940
eqm_v3ulr_rev3/60.201.262
Fox_Home_Entertainment/178.773.270
HVS_Best/77.182.472
HVS_Better/69.515.122
HVS_Good/66.998.656
HVS-1.0/83.951.800
HVS-1.5/59.203.276
Hybrid8AQ/77.288.716
Hybrid-X3/73.238.542
Jawors_1CD/63.776.274
Jawors_2CD/78.921.422
Low_Bitrate_Matrix/73.735.486
mp4guy's_2nd_CQM/85.862.568
mp4guy's_mid_bitrate/125.383.076
Professional/129.731.406
Selur_High_datarate/131.456.952
Selur's_Matrix/129.868.976
Soulhunters_v3_(MPEG-EX)/93.180.908
Soulhunters_v5_(HVS1)/67.547.616
Soulhunters_v6_(HVS2)/77.701.870
Soulhunters_v7_(XHQ)/93.681.574
Soulhunters_v8_(HVS3)/66.237.912
Y.A.C.Q.M/97.317.974

These are the first results, if you want me to test other matrices, just let me know.
The layout (NAME/SIZE) allows you to import data easily on Excel and work on them.

The PSNR test in not really necessary, however I'd like to do it, only to give some info about how matrices work (READ: I don't want to give any judgement about video quality, for, IMHO, this is PARTIALLY related to PSNR).
Anyway, for this test I use a script like...

Original = Mpeg2Source("C:/../vts.d2v")...FILTERS(...)...ConvertToYUY2()
ClipAVI = Avisource("C:/..vts.avi").ConvertToYUY2()
Compare(ClipAVI, Original, "", "C:/vts.log")

...and then I load it on VD and play it; I have 2 questions about it:
1) Is the script OK?
2) Is there a way to "add to queue" a Preview on VirtualDub? Or, better, is there a PSNR calculator that allows something like that (READ: ..that doesn't need my presence to test sequentially all files)?
IIRC, PSNR Creator doesn't allow it, but I'm not sure...

@Sharktooth
eqm_AGKHR_b1-5....does it mean "Auto-Gordian-Knot-High-Rate"? Should I test them?

EDIT: 11 CQMs added

SpikeSpiegel

1st May 2005, 12:26

The "compression level" can be quantified with something like:

C.L. = (1/SIZE-in-bytes)*5*10^9 -- 5*10^9 is so that a 50MB file has a CL of 100, but I can use other values

This way we get:

Acailas 44
Angel_BestLow 68
Angel_BestVeryLow 69
Audionut's 54
Audionut's_HQ 51
Bach1 75
Bulletproof's_Heavy_Comp 96
Bulletproof's_HQ_Matrix 69
CCE_Smooth_(CG_etc.) 57
CCE_Ultra_low_bitrate 74
CCE_Very_low_bitrate 69
Didees_SixOfNine 42
Didees_SixOfNine_HVS 43
eqm_autogk_sharp 60
eqm_v1 60
eqm_v3ehr 25
eqm_v3hr 57
eqm_v3lr 71
eqm_v3uhr 25
eqm_v3uhr_rev2 34
eqm_v3ulr 84
eqm_v3ulr_rev2 80
eqm_v3ulr_rev3 83
Fox_Home_Entertainment 28
HVS_Best 65
HVS_Better 72
HVS_Good 75
HVS-1.0 60
HVS-1.5 84
Hybrid8AQ 65
Hybrid-X3 68
Jawors_1CD 78
Jawors_2CD 63
Low_Bitrate_Matrix 68
mp4guy's_2nd_CQM 58
mp4guy's_mid_bitrate 40
Professional 39
Selur_High_datarate 38
Selur's_Matrix 39
Soulhunters_v3_(MPEG-EX) 54
Soulhunters_v5_(HVS1) 74
Soulhunters_v6_(HVS2) 64
Soulhunters_v7_(XHQ) 53
Soulhunters_v8_(HVS3) 75
Y.A.C.Q.M 51

This time the layout allows you to go crazy after the 3rd matrix :devil:

EDIT: 11 CQMs added

unmei

1st May 2005, 17:17

This time the layout allows you to go crazy after the 3rd matrix
You know, you could use the code BBCode (#-Button) plus whitespace.. ;)

SpikeSpiegel

1st May 2005, 21:57

Originally posted by unmei
You know, you could use the code BBCode (#-Button) plus whitespace.. ;)
I'm still working on CQMs, it's too early to think about the layout.

Maybe, the very last results will be posted in another thread...and they will be surely more comprehensible ;)

darkavatar1470

2nd May 2005, 03:43

@ SpikeSpiegel
It came to me the other day 1/(bit per pixel) might be a nice unit for "compressibility" , its just pixels per bit , higher value means more data saved for given amount of space, which is pretty much "compressibility" to me. Another reason for this unit is that full quality DVDs have a bit/pixel around 1 ,resulting a starting pixel/bit value of one, and a clip that has pixel/bit of 10 would mean a ten times compression compared to the original DVD.

i think its easier to understand & to make an useful meaning than the original bit/pixel.

yaz

2nd May 2005, 11:02

@spikespiegel
xtreme ! thx a lot ! anyway, i would love to see the 'standard' matrices too (h263, mpeg) ... or is it quite outta date to use such things these days ? ;)

thx
y

Sharktooth

2nd May 2005, 14:26

Originally posted by SpikeSpiegel
@Sharktooth
eqm_AGKHR_b1-5....does it mean "Auto-Gordian-Knot-High-Rate"? Should I test them?
They were test matrices for autogk. If you want to test them you're free to do it, but they are almost somewhere in between from v3hr to v2

SpikeSpiegel

3rd May 2005, 01:04

@ darkavatar1470
Well, honestly I don't think it to be more indicative and clear than 1/SIZE...anyway, are you sure that DVDs b/p ratio is almost = 1?

@ yaz
H.263 & MPEG have been already tested (respectively, the size is 65.038.628 and 75.128.874 bytes), look at my 4th and 5th posts: they aren't in the last ones for they're not CQMs.

@ Sharktooth
Ok, I will do the size test on these CQMs anyway, this is very fast, unlike PSNR test...apropos:
is there a program that can do it on multiple files (almost) automatically? I can use Avisynth's "compare" and VD, but it may take much time for all 47 (ehm, 52) matrices...

Sharktooth

3rd May 2005, 04:04

Sorry, i dont know about programs for automating psnr tests...

yaz

3rd May 2005, 09:09

Originally posted by SpikeSpiegel
@ yaz
H.263 & MPEG have been already tested (respectively, the size is 65.038.628 and 75.128.874 bytes), look at my 4th and 5th posts: they aren't in the last ones for they're not CQMs.aha ... if it's a reson for neglecting ... ;)
Originally posted by SpikeSpiegel
is there a program that can do it on multiple files (almost) automatically? I can use Avisynth's "compare" and VD, but it may take much time for all 47 (ehm, 52) matrices... try psnrlab (http://forum.doom9.org/showthread.php?s=&threadid=58102) not exactly what you want but it's easy to adapt.
the bests
y

SpikeSpiegel

3rd May 2005, 20:45

These are the last results:

SIZE:

H263/65.038.628
MPEG/75.128.874

Acailas/113.844.640
Andreas 78er Matrix/99.013.574
Andreas doppelte 99er Matrix/98.966.822
Andreas einfache 99er Matrix/98.981.706
Andreas Matrix fuer DivX Quelle/144.213.928
Angel_BestLow/73.770.628
Angel_BestVeryLow/72.737.116
Audionut's/93.306.298
Audionut's_HQ/97.916.076
Bach1/66.762.710
Bulletproof's_Heavy_Comp/52.262.766
Bulletproof's_HQ_Matrix/72.008.568
CCE_Smooth_(CG_etc.)/87.668.248
CCE_Ultra_low_bitrate/67.167.056
CCE_Very_low_bitrate/72.784.918
Didees_SixOfNine/117.857.938
Didees_SixOfNine_HVS/115.641.710
eqm_AGKHR_b1/87.991.612
eqm_AGKHR_b2/86.088.822
eqm_AGKHR_b3/82.889.034
eqm_AGKHR_b4/85.338.030
eqm_AGKHR_b5/80.692.202
eqm_autogk_sharp/82.901.156
eqm_v1/84.002.012
eqm_v3ehr/201.128.908
eqm_v3hr/88.475.138
eqm_v3lr/70.287.336
eqm_v3uhr/200.551.444
eqm_v3uhr_rev2/148.112.456
eqm_v3ulr/59.568.398
eqm_v3ulr_rev2/62.130.940
eqm_v3ulr_rev3/60.201.262
Fox_Home_Entertainment/178.773.270
HVS_Best/77.182.472
HVS_Better/69.515.122
HVS_Good/66.998.656
HVS-1.0/83.951.800
HVS-1.5/59.203.276
Hybrid8AQ/77.288.716
Hybrid-X3/73.238.542
Jawors_1CD/63.776.274
Jawors_2CD/78.921.422
Low_Bitrate_Matrix/73.735.486
mp4guy's_2nd_CQM/85.862.568
mp4guy's_mid_bitrate/125.383.076
Professional/129.731.406
Selur_High_datarate/131.456.952
Selur's_Matrix/129.868.976
Soulhunters_v3_(MPEG-EX)/93.180.908
Soulhunters_v5_(HVS1)/67.547.616
Soulhunters_v6_(HVS2)/77.701.870
Soulhunters_v7_(XHQ)/93.681.574
Soulhunters_v8_(HVS3)/66.237.912
Y.A.C.Q.M/97.317.974

Compression Level
1/SIZE * 5*10^9
H263 77
MPEG 67

Acailas 44
Andreas 78er Matrix 50
Andreas doppelte 99er Matrix 51
Andreas einfache 99er Matrix 51
Andreas Matrix fuer DivX Quelle 35
Angel_BestLow 68
Angel_BestVeryLow 69
Audionut's 54
Audionut's_HQ 51
Bach1 75
Bulletproof's_Heavy_Comp 96
Bulletproof's_HQ_Matrix 69
CCE_Smooth_(CG_etc.) 57
CCE_Ultra_low_bitrate 74
CCE_Very_low_bitrate 69
Didees_SixOfNine 42
Didees_SixOfNine_HVS 43
eqm_AGKHR_b1 57
eqm_AGKHR_b2 58
eqm_AGKHR_b3 60
eqm_AGKHR_b4 59
eqm_AGKHR_b5 62
eqm_autogk_sharp 60
eqm_v1 60
eqm_v3ehr 25
eqm_v3hr 57
eqm_v3lr 71
eqm_v3uhr 25
eqm_v3uhr_rev2 34
eqm_v3ulr 84
eqm_v3ulr_rev2 80
eqm_v3ulr_rev3 83
Fox_Home_Entertainment 28
HVS_Best 65
HVS_Better 72
HVS_Good 75
HVS-1.0 60
HVS-1.5 84
Hybrid8AQ 65
Hybrid-X3 68
Jawors_1CD 78
Jawors_2CD 63
Low_Bitrate_Matrix 68
mp4guy's_2nd_CQM 58
mp4guy's_mid_bitrate 40
Professional 39
Selur_High_datarate 38
Selur's_Matrix 39
Soulhunters_v3_(MPEG-EX) 54
Soulhunters_v5_(HVS1) 74
Soulhunters_v6_(HVS2) 64
Soulhunters_v7_(XHQ) 53
Soulhunters_v8_(HVS3) 75
Y.A.C.Q.M 51

SpikeSpiegel

3rd May 2005, 20:54

Next (and maybe last) round should be:

1. Max GE 1900 kbps
2. Max LT 1900 kbps
3. mb1 DV 720x576 min 2000 avg 4000 max 5000
4. mb1 interlaced DV
5. Packed DVD
6. SxGz (DVD v3)
7. SxGz (HB v3)
8. SxGz (LB v4)
9. SxGz (MB v7)
10. SxGz (SVCD v3)

There are other popular matrices, but, till I can't get any info (purpose, etc.), I can't test them:

1. 28DaysLater
2. Andreas SVCD
3. Angel SVCD 1CD 2.35 avg800 max1500
4. Angel SVCD 4-3 SVCD Interlaced 34
5. Angel SVCD 4-3 SVCD Interlaced 99
6. AVAMAT6-AutoQ1
7. EthanoliX Four Rooms
8. Flat 8-16
9. Holy cow batman
10. Kika Trickfilm High
11. Kika Trickfilm Low
12. KR S-curve
13. KVCD 320x432 CQ65
14. KVCD 448x320 CQ65
15. KVCD Notch
16. QLB v1.2
17. Semi-insane
18. TMPGEnc
19. Ultimate
20. Warner

@ yaz
Yes, yes, the only thing I'm neglecting is your inattention! This time I'll override it, but the next I'll be merciless! :p
P.S. Seriously, H.263 is one of my favorites!

Originally posted by yaz
try psnrlab (http://forum.doom9.org/showthread.php?s=&threadid=58102) not exactly what you want but it's easy to adapt.

Thanks, but I can't use it for the test and...well, maybe there's no more (http://forum.doom9.org/showthread.php?s=&postid=649192#post649192) need of a PSNR test..

Soulhunter

3rd May 2005, 23:55

Dunno if everything is right, but...

01. ???
02. LBR MPEG2
03. LBR MPEG2
04. MBR Interlaced MPEG2
05. MBR Interlaced MPEG2
06. LBR
07. ???
08. HBR
09. ???
10. Cartoon LBR
11. Cartoon HBR
12. ???
13. LBR MPEG1
14. LBR MPEG1
15. LBR MPEG1
16. ???
17. HBR
18. ???
19. ???
20. ???

Bye

darkavatar1470

4th May 2005, 14:56

Originally posted by SpikeSpiegel
@ darkavatar1470
Well, honestly I don't think it to be more indicative and clear than 1/SIZE...anyway, are you sure that DVDs b/p ratio is almost = 1? Uh... thats because most of my anime DVD collection is around 8Mbs:
(8 * 1000 * 1000) / (720 * 480 * 24) = 0.964506173
and i used full DVD bitrate of 10.4Mbs for my first guess..
(10.4 * 1000 * 1000) / (720 * 480 * 30) = 1.00308642
i kinda forgot that normal action film movies don't get that much...

1/SIZE is sort of like what i was talking about, but using which unit doesn't really matter anyway.... scale the highest vaule to 100% is enough to make a comparison i think.

ps: how odd that PAL gets the same b/p:
(10.4 * 1000 * 1000) / (720 * 576 * 25) = 1.00308642

SpikeSpiegel

4th May 2005, 17:24

Originally posted by Soulhunter
Dunno if everything is right, but...

01. ???
02. LBR MPEG2
.
.
19. ???
20. ???

Bye
Thanks!
Anyway, now, a terrible doubt haunts me: "Holy cow batman" is referred to a (very strange) Batman's parody or it's a nickname? :p

Now almost all popular CQMs have been tested (results will arrive soon)...all but these:

KVCD Notch
QLB v1.2
SxGz (DVD v3)
SxGz (HB v3)
SxGz (LB v4)
SxGz (MB v7)
SxGz (SVCD v3)
Warner

No, I don't wanna discriminate 'em, I just can't find .cqm/.qmatrix/.xcm files (only .txt that can't be directly imported...I'm too lazy, I know :o ).

@ darkavatar1470
Ok, anyway values have been already scaled so that MAX=~100% (I used a little approssimation)

darkavatar1470

4th May 2005, 17:25

arg, you beat my edit...i tend to edit them alot....

pwh04

4th May 2005, 19:18

Spike - thx again for your effort here. I love testing matrices and this really helps - very efficient. Note- I would have used Compression Results order instead of/or in addition to alphabetically :-) but I'm just being very picky!

Thanks!!!

SpikeSpiegel

6th May 2005, 13:58

Compression Level - Size (MB) - CQM Name (alphabetical order)

77 62,03 H263
67 71,65 MPEG

29 164,85 28DaysLater
44 108,57 Acailas
50 94,43 Andreas_78er_Matrix
51 94,38 Andreas_doppelte_99er_Matrix
51 94,4 Andreas_einfache_99er_Matrix
35 137,53 Andreas_Matrix_fuer_DivX_Quelle
68 70,27 Andreas_SVCD_Matrix
68 70,35 Angel_BestLow
69 69,37 Angel_BestVeryLow
51 94,3 Angel_SVCD_1CD_2.35_avg800_max1500
48 99,74 Angel_SVCD_4-3_SVCD_Interlaced_34
48 99,72 Angel_SVCD_4-3_SVCD_Interlaced_99
54 88,98 Audionut's
51 93,38 Audionut's_HQ
82 57,96 AVAMAT6
75 63,67 Bach1
96 49,84 Bulletproof's_Heavy_Comp
69 68,67 Bulletproof's_HQ_Matrix
57 83,61 CCE_Smooth_(CG_etc.)
74 64,06 CCE_Ultra_low_bitrate
69 69,41 CCE_Very_low_bitrate
42 112,4 Didees_SixOfNine
43 110,28 Didees_SixOfNine_HVS
57 83,92 eqm_AGKHR_b1
58 82,1 eqm_AGKHR_b2
60 79,05 eqm_AGKHR_b3
59 81,38 eqm_AGKHR_b4
62 76,95 eqm_AGKHR_b5
60 79,06 eqm_autogk_sharp
60 80,11 eqm_v1
25 191,81 eqm_v3ehr
57 84,38 eqm_v3hr
71 67,03 eqm_v3lr
25 191,26 eqm_v3uhr
34 141,25 eqm_v3uhr_rev2
84 56,81 eqm_v3ulr
80 59,25 eqm_v3ulr_rev2
83 57,41 eqm_v3ulr_rev3
67 70,79 EthanoliX_Four_Rooms
46 104,34 Flat_8-16
28 170,49 Fox_Home_Entertainment
40 119,33 Holy_cow_batman
65 73,61 HVS_Best
72 66,29 HVS_Better
75 63,89 HVS_Good
60 80,06 HVS-1.0
84 56,46 HVS-1.5
65 73,71 Hybrid8AQ
68 69,85 Hybrid-X3
78 60,82 Jawors_1CD
63 75,27 Jawors_2CD
57 84,28 Kika_Trickfilm_High
34 138,89 Kika_Trickfilm_Low
42 112,6 KR_S-curve
69 69,12 KVCD_320x432_CQ65
69 69,18 KVCD_448x320_CQ65
68 70,32 Low_Bitrate_Matrix
68 70,38 Max_GE_1900kbps
68 70,22 Max_LT_1900_kbps
69 69,37 mb1_DV_720x576_min2k-avg4k-max5k
48 99,71 mb1_interlaced_DV
58 81,88 mp4guy's_2nd_CQM
40 119,57 mp4guy's_mid_bitrate
75 63,72 Packed_DVD_by_Fl0ppy
39 123,72 Professional
38 125,37 Selur_High_datarate
39 123,85 Selur's_Matrix
30 161,14 Semi_Insane
54 88,86 Soulhunters_v3_(MPEG-EX)
74 64,42 Soulhunters_v5_(HVS1)
64 74,1 Soulhunters_v6_(HVS2)
53 89,34 Soulhunters_v7_(XHQ)
75 63,17 Soulhunters_v8_(HVS3)
58 81,77 TMPEGEnc_CG
90 52,7 Ultimate_Matrix
74 64,06 Ultra_Low_Bitrate_Matrix
69 69,41 Very_Low_Bitrate_Matrix
51 92,81 Y.A.C.Q.M

Compression Level - Size (MB) - CQM Name (size-C.L. order)

67 71,65 MPEG
77 62,03 H263

25 191,81 eqm_v3ehr
25 191,26 eqm_v3uhr
28 170,49 Fox_Home_Entertainment
29 164,85 28DaysLater
30 161,14 Semi_Insane
34 141,25 eqm_v3uhr_rev2
34 138,89 Kika_Trickfilm_Low
35 137,53 Andreas_Matrix_fuer_DivX_Quelle
38 125,37 Selur_High_datarate
39 123,72 Professional
39 123,85 Selur's_Matrix
40 119,33 Holy_cow_batman
40 119,57 mp4guy's_mid_bitrate
42 112,4 Didees_SixOfNine
42 112,6 KR_S-curve
43 110,28 Didees_SixOfNine_HVS
44 108,57 Acailas
46 104,34 Flat_8-16
48 99,74 Angel_SVCD_4-3_SVCD_Interlaced_34
48 99,72 Angel_SVCD_4-3_SVCD_Interlaced_99
48 99,71 mb1_interlaced_DV
50 94,43 Andreas_78er_Matrix
51 94,38 Andreas_doppelte_99er_Matrix
51 94,4 Andreas_einfache_99er_Matrix
51 94,3 Angel_SVCD_1CD_2.35_avg800_max1500
51 93,38 Audionut's_HQ
51 92,81 Y.A.C.Q.M
53 89,34 Soulhunters_v7_(XHQ)
54 88,98 Audionut's
54 88,86 Soulhunters_v3_(MPEG-EX)
57 83,61 CCE_Smooth_(CG_etc.)
57 83,92 eqm_AGKHR_b1
57 84,38 eqm_v3hr
57 84,28 Kika_Trickfilm_High
58 82,1 eqm_AGKHR_b2
58 81,88 mp4guy's_2nd_CQM
58 81,77 TMPEGEnc_CG
59 81,38 eqm_AGKHR_b4
60 79,05 eqm_AGKHR_b3
60 79,06 eqm_autogk_sharp
60 80,11 eqm_v1
60 80,06 HVS-1.0
62 76,95 eqm_AGKHR_b5
63 75,27 Jawors_2CD
64 74,1 Soulhunters_v6_(HVS2)
65 73,61 HVS_Best
65 73,71 Hybrid8AQ
67 70,79 EthanoliX_Four_Rooms
68 70,27 Andreas_SVCD_Matrix
68 70,35 Angel_BestLow
68 69,85 Hybrid-X3
68 70,32 Low_Bitrate_Matrix
68 70,38 Max_GE_1900kbps
68 70,22 Max_LT_1900_kbps
69 69,37 Angel_BestVeryLow
69 68,67 Bulletproof's_HQ_Matrix
69 69,41 CCE_Very_low_bitrate
69 69,12 KVCD_320x432_CQ65
69 69,18 KVCD_448x320_CQ65
69 69,37 mb1_DV_720x576_min2k-avg4k-max5k
69 69,41 Very_Low_Bitrate_Matrix
71 67,03 eqm_v3lr
72 66,29 HVS_Better
74 64,06 CCE_Ultra_low_bitrate
74 64,42 Soulhunters_v5_(HVS1)
74 64,06 Ultra_Low_Bitrate_Matrix
75 63,67 Bach1
75 63,89 HVS_Good
75 63,72 Packed_DVD_by_Fl0ppy
75 63,17 Soulhunters_v8_(HVS3)
78 60,82 Jawors_1CD
80 59,25 eqm_v3ulr_rev2
82 57,96 AVAMAT6
83 57,41 eqm_v3ulr_rev3
84 56,81 eqm_v3ulr
84 56,46 HVS-1.5
90 52,7 Ultimate_Matrix
96 49,84 Bulletproof's_Heavy_Comp

CQM Name / Size (in bytes)

H263/65.038.628
MPEG/75.128.874

28DaysLater/172.861.180
Acailas/113.844.640
Andreas_78er_Matrix/99.013.574
Andreas_doppelte_99er_Matrix/98.966.822
Andreas_einfache_99er_Matrix/98.981.706
Andreas_Matrix_fuer_DivX_Quelle/144.213.928
Andreas_SVCD_Matrix/73.684.942
Angel_BestLow/73.770.628
Angel_BestVeryLow/72.737.116
Angel_SVCD_1CDa2.35avg800max1500/98.885.778
Angel_SVCD_4-3_SVCD_Interlaced_34/104.582.530
Angel_SVCD_4-3_SVCD_Interlaced_99/104.562.432
Audionut's/93.306.298
Audionut's_HQ/97.916.076
AVAMAT6/60.774.594
Bach1/66.762.710
Bulletproof's_Heavy_Comp/52.262.766
Bulletproof's_HQ_Matrix/72.008.568
CCE_Smooth_(CG_etc.)/87.668.248
CCE_Ultra_low_bitrate/67.167.056
CCE_Very_low_bitrate/72.784.918
Didees_SixOfNine/117.857.938
Didees_SixOfNine_HVS/115.641.710
eqm_AGKHR_b1/87.991.612
eqm_AGKHR_b2/86.088.822
eqm_AGKHR_b3/82.889.034
eqm_AGKHR_b4/85.338.030
eqm_AGKHR_b5/80.692.202
eqm_autogk_sharp/82.901.156
eqm_v1/84.002.012
eqm_v3ehr/201.128.908
eqm_v3hr/88.475.138
eqm_v3lr/70.287.336
eqm_v3uhr/200.551.444
eqm_v3uhr_rev2/148.112.456
eqm_v3ulr/59.568.398
eqm_v3ulr_rev2/62.130.940
eqm_v3ulr_rev3/60.201.262
EthanoliX_Four_Rooms/74.233.400
Flat_8-16/109.407.046
Fox_Home_Entertainment/178.773.270
Holy_cow_batman/125.125.766
HVS_Best/77.182.472
HVS_Better/69.515.122
HVS_Good/66.998.656
HVS-1.0/83.951.800
HVS-1.5/59.203.276
Hybrid8AQ/77.288.716
Hybrid-X3/73.238.542
Jawors_1CD/63.776.274
Jawors_2CD/78.921.422
Kika_Trickfilm_High/88.372.446
Kika_Trickfilm_Low/145.638.360
KR_S-curve/118.074.594
KVCD_320x432_CQ65/72.479.036
KVCD_448x320_CQ65/72.543.032
Low_Bitrate_Matrix/73.735.486
Max_GE_1900kbps/73.796.070
Max_LT_1900_kbps/73.630.902
mb1_DV_720x576_min2k-avg4k-max5k/72.739.046
mb1_interlaced_DV/104.556.378
mp4guy's_2nd_CQM/85.862.568
mp4guy's_mid_bitrate/125.383.076
Packed_DVD_by_Fl0ppy/66.812.564
Professional/129.731.406
Selur_High_datarate/131.456.952
Selur's_Matrix/129.868.976
Semi_Insane/168.962.942
Soulhunters_v3_(MPEG-EX)/93.180.908
Soulhunters_v5_(HVS1)/67.547.616
Soulhunters_v6_(HVS2)/77.701.870
Soulhunters_v7_(XHQ)/93.681.574
Soulhunters_v8_(HVS3)/66.237.912
TMPEGEnc_CG/85.745.562
Ultimate_Matrix/55.255.652
Ultra_Low_Bitrate_Matrix/67.167.056
Very_Low_Bitrate_Matrix/72.784.918
Y.A.C.Q.M/97.317.974

The test is almost finished, now it's your turn!

You should decide how to classify these CQMs;
I suggest to use the EQMs acronyms (for most people is already familiar with 'em) and "compression level" as parameter, for example:
99 > CL > 80 --> ULR ("Ultra Low bitRate")
80 > CL > 65 --> LR ("Low bitRate")
.....

EDIT1: Obviously CL is less important than "CQM purpose"; I mean, first of all, CQMs should be classified considering their purpose (DVD/MPEG1/DV encoding, VHS/TV capture, etc.), then, once we have assigned 'em to these categories, we can classify CQMs in each "group" using CL as parameter.

Then, as Soulhunter has already done, if you know which is the purpose of the "strangest" CQMs, let me know, please!

EDIT2: fixed a typo (or freudian lapsus...): DV->TV
EDIT3: ...idem: EMQs -> EQMs

Poutnik

6th May 2005, 22:31

Originally posted by Teegedeck
IMHO it would be very helpful if CQMs were described with attributes that indicate how to use a specific matrix to decent results, like (A) with what kind of sources preferably or with what kind of sources not at all, (B) at which levels of compression preferably or outside which range of compression use voids warranty , and (C) which characteristics - if any - of the original's look they might stress or alter (--> taste). It could perhaps facilitate the choice for most people. I think some forum authority ( as you, Didee, Soulhunter, Sharktooth ... :-) ) could recommend some naming scheme like: Author-FriendlyName-AxByCz.cqm,
e.g.: Didee-SixOfNine-A(1-..)B(1-..)C(1-..).cqm

A1 - sharp noiseless source
B2 - low compression, high BPP ratio ( > 0,3 ), 2-3CD backups
C3 - HiFi matrix, keeps all details/noise

CQM author should respect such convention....

MacAddict

6th May 2005, 22:38

Excellent thread and information here. Many thanks!

I agree, some sort of classification for each CQM would be ideal...at the very least a common naming convention.

SpikeSpiegel

8th May 2005, 18:53

@ Poutnik

I agree with what you wrote...except this:
Originally posted by Poutnik
I think some forum authority (as you ,...)
:confused: :p

Seriously, I think that it's quite difficult that:
Originally posted by Poutnik
CQM author should respect such convention....
Anyway this thread may be a "guide" to these matrices or may be used to add some info in the sites were they can be downloaded.
...and, as I've already said answering the Teegedeck's post, the only info we can give is about compression level, because the "(C)" characteristics (overall quality, sharpness, blocks&noise presence, etc.) are quite subjective and hard to test as well.

About the "(A)" characteristics (purpose), we can classify 'em this way:
- General purpose / DVD encoding
All but:

- ANIME:
Kika_Trickfilm_High
Kika_Trickfilm_Low

- SVCD-MPEG2 (that may be integrated in "DVD encoding"):
Andreas_SVCD_Matrix
Angel_SVCD_1CDa2.35avg800max1500
Angel_SVCD_4-3_SVCD_Interlaced_34
Angel_SVCD_4-3_SVCD_Interlaced_99

- VCD-MPEG1
KVCD_320x432_CQ65
KVCD_448x320_CQ65

- DV
mb1_DV_720x576_min2k-avg4k-max5k
mb1_interlaced_DV
"General puropse / dvd enc." can be split on different categories (considering CL, or char. of the source)
Let me know if there are errors or a CQM missing (sure).