New Quantization Matrices - EQM V3 series (last update: 04/02/2005 - V3ULR)

[Sharktooth]

Description:
This is a HIGH BITRATE matrix designed to work at full anamorphic DVD res (720x432) with bitrates over 1600kbits (for an average motion movie) and works pretty well even with Q2 encodings or extremely High bitrates. But for that target there will be an Ultra-high bitrate matrix (EMQ V3UHR).

EQM V3HR:

8 10 10 10 11 11 13 15
10 10 10 10 11 12 14 16
10 10 11 11 13 14 16 18
10 10 11 13 15 17 19 23
11 11 13 15 19 22 26 29
11 12 14 17 22 28 34 41
13 14 16 19 26 34 44 55
15 16 18 23 29 41 55 72

15 15 15 15 16 17 19 22
15 15 15 15 16 18 20 23
15 15 16 17 19 20 23 27
15 15 17 19 22 25 29 33
16 16 19 22 28 32 38 43
17 18 20 25 32 41 50 60
19 20 23 29 38 50 66 81
22 23 27 33 43 60 81 106


Here's the direct link: EQM V3HR (http://www.webalice.it/f.corriga/CQMs/eqm_v3hr.xcm)

Notes: Unlike V1 and V2, EQM V3 set of matrices were generated using a matrix built starting from a HVS study on luminance by IEEE. The same used by ReferenceDivX to build his HVS matrices.The differences with V2 are minimal (coz V2 is HVS tuned too, but in a different way) but the new coefficient scaling ensures V3 is full IEEE HVS compliant.
The HR matrix is a mid-high/high bitrate matrix.A low, ultra-low and ultra-high bitrate counterparts are coming...
Have fun with it and post your impressions.
Thanx in advance.

WARNING: It may not work with some standalones (expecially the ones with ESS chipsets).

IMPORTANT NOTE:
- FIXED -
There's a recently discovered bug in Xvid that affects all 1.xx versions with Trellis Quantization.
Some matrices trigger an overflow in trellis causing visible blocking at certain quants. One of them is EQM V3HR (but other matrices where reported as well).
There are 2 workarounds (at least i found those 2 only):
disable Trellis Quantization OR limit the MinQ and MaxQ to 2-31 respectively.
- FIXED -

EDIT: Updated on 01/10/2004

[Sharktooth]

Description:
This is a LOW BITRATE matrix designed to work at full anamorphic DVD res (720x432) with bitrates lower than 1500kbits (for an average motion movie) and works pretty well even with high motion sources (using higher bitrates) keeping details where other matrices produce artifacts.

EQM V3LR:

8 14 14 14 15 15 18 21
14 14 14 14 15 16 19 22
14 14 15 15 17 19 22 25
14 14 15 18 21 23 27 31
15 15 17 21 26 30 35 40
15 16 19 23 30 38 46 56
18 19 22 27 35 46 61 76
21 22 25 31 40 56 76 99

19 19 19 19 20 21 24 28
19 19 19 19 20 22 26 29
19 19 20 21 23 26 29 34
19 19 21 24 28 31 36 42
20 20 23 28 35 41 48 55
21 22 26 31 41 51 63 76
24 26 29 36 48 63 83 103
28 29 34 42 55 76 103 134


Here's the direct dowload link: EQM V3LR (http://www.webalice.it/f.corriga/CQMs/eqm_v3lr.xcm)
The LR matrix was created in the same way of HR matrix but tuned for low bitrate encodes.

[Sharktooth]

Description:
This is an ULTRA-LOW BITRATE soft matrix (re)designed for 2 reasons:
- Reducing artifacts as low as possible while keeping a good overall picture quality on low bitrate encodes.
- Producing the best matrix for 1 CD backup.


EQM V3ULR (revision 3)

8 14 20 28 38 45 56 65
14 20 26 37 44 56 64 70
20 26 36 43 55 64 70 75
28 37 43 54 63 70 76 79
38 44 55 63 70 77 80 83
45 56 64 70 77 81 84 88
56 64 70 76 80 84 89 95
65 70 75 79 83 88 95 102

18 19 20 22 24 26 28 30
19 20 22 24 26 28 30 34
20 22 24 26 29 30 34 40
22 24 26 30 32 36 44 50
24 26 29 32 38 48 55 60
26 28 30 36 48 58 74 80
28 30 34 44 55 74 92 104
30 34 40 50 60 80 104 128


Here's the direct download link:EQM V3ULR (http://www.webalice.it/f.corriga/CQMs/eqm_v3ulr_rev3.xcm)

Notes: This CQM is not HVS tuned.

EDIT: Matrix updated on 04/02/2005. Gained some compression, should avoid smaller artifacts.

[Sharktooth]

Description:
This is an ULTRA-HIGH BITRATE matrix designed to work at full anamorphic DVD res (720x432) at very very high bitrates. The saturation point is 30% higher than Didée's 6of9 matrix but at the cost of a lower compressibility.

EQM V3UHR (revision 2):

8 8 8 8 8 8 10 11
8 8 8 8 8 9 10 12
8 8 8 8 9 10 12 14
8 8 8 10 11 13 15 17
8 8 9 11 14 16 19 22
8 9 10 13 16 21 25 31
10 10 12 15 19 25 33 41
11 12 14 17 22 31 41 54

10 10 10 10 10 10 10 11
10 10 10 10 10 10 10 12
10 10 10 10 10 10 12 14
10 10 10 10 11 13 15 17
10 10 10 11 14 16 19 22
10 10 10 13 16 21 25 31
10 10 12 15 19 25 33 41
11 12 14 17 22 31 41 54


Here's the direct dowload link: EQM V3UHR (http://www.webalice.it/f.corriga/CQMs/eqm_v3uhr_rev2.xcm)

WARNING: It may not work with some standalones (expecially the ones with ESS chipsets).

UPDATED 17/12/2004

[JarrettH]

Thanks, I'll have to try them next movie I do.
Any comments regarding these matrices versus others?

[Sharktooth]

Sure. All those matrices are HVS tuned, that means they are tuned in some way to spare bits from unvisible details and use them for the visible details.
UHR is a real monster when used at constant Q3 (or for DVD-R size backups) expecially with high quality sources.
For a general use i would reccomend LR or HR.
ULR is not properly an ultra-low bitrate matrix but it's very suitable for full DVD resolution 1CD backups (i dont even take into consideration bitrates lower than 700kbps cause V3 series matrices are aimed for movie backups).
As i said ULR has the same application of Jawor's 1CD matrix.
LR and HR are suitable for 2 or 3 CD backups, you can use them in place of HVS Better/Best or similar matrices.
I'm still testing them on different sources and i need some feedback on compressibility results.
I will do some fine tuning after the test results.

[spolja]

Hi,
Is Ultra High Bitrate matrix suitable for HD content?

[Sharktooth]

It is for sure, but remember that HD content needs more bitrate due to the higher resolution.

[Sharktooth]

Description:
This is an EXTREMELY HIGH BITRATE matrix designed for constant quality encodings ONLY. Compressibility is "ridiculus" in comparison with other high-bitrate matrices but there is a reason...
It's application is pretty obvious: it's for backing up original material or HD material without using lossless codecs.

EQM V3EHR:

8 8 9 11 13 13 14 17
8 8 11 12 13 14 17 18
9 11 13 13 14 17 17 16
11 11 13 13 13 17 18 20
11 13 13 13 16 17 20 24
13 13 13 16 17 20 24 29
13 12 13 17 19 23 28 34
12 13 17 19 23 28 34 41

8 8 8 8 8 8 8 8
8 8 8 8 8 8 8 8
8 8 8 8 8 8 8 8
8 8 8 8 8 8 8 8
8 8 8 8 8 8 8 8
8 8 8 8 8 8 8 8
8 8 8 8 8 8 8 8
8 8 8 8 8 8 8 8


Here's the direct download link: EQM V3EHR (http://www.webalice.it/f.corriga/CQMs/eqm_v3ehr.xcm)

Notes: This CQM is not HVS tuned.

WARNING: It may not work with some standalones (expecially the ones with ESS chipsets).

EDIT: Matrix updated (uhm... completely redesigned) on 05/10/2004.

[olavitur]

Shartooth: I tested your new EQM V3UHR quantization matrix with two movies (The Bourne Identity and Kill Bill Vol. 2) and the quality I got was simply amazing. Though I had to use bitrates around 3000kbps to get there but it was worth every bit.
Both of the movies were in PAL 720x576 format so I decided to use LanczosResize to resize then to 720x480 so that I wouldn't have to use even higher bitrates (~3000kbps is high enough for me). First pass (Q2) bitrates were around 6500kbps for both of the movies.

Currently I'm making some more tests and comparing EQM V3UHR with Semi-Insane (by the way, SixOfNine didn't have a change against V3UHR).

[Sharktooth]

I've spent a lot of time for "internal" tests and I'm really happy to receive some positive feedback.
There is also an "experimental" version of EQM V3EHR for insanely high bitrates (6000+) or Q2 encodings.
If you have some free time i suggest you to try EHR too.
Have a nice day and thanks again for the feedback.

[olavitur]

Originally posted by Sharktooth
I've spent a lot of time for "internal" tests and I'm really happy to receive some positive feedback.
There is also an "experimental" version of EQM V3EHR for insanely high bitrates (6000+) or Q2 encodings.
If you have some free time i suggest you to try EHR too.
Have a nice day and thanks again for the feedback.

Thanks, but no thanks... I'm not that insane :)
I think your EQM V3UHR is good enough for most of us high quality freaks. Not many of us wants to encode xvid clips which are the same size as original dvd...

I managed to do some comparing between V3UHR and Semi-Insane (SI). This time it was really tough to find any real differences between these two matrixes as they produced about equal picture quality. But it seemes that V3UHR handles noisy parts of the movie a little bit better than SI does. With SI when there was panning in the video noise tend to float on the screen a little. With V3UHR I didn't notice this so much. Maybe SI works better on movies that have minimal amount of noise.

[Sharktooth]

UHR should produce a smaller file than semi insane @constant Q2.
Tnx again for your feedback :)

[mateo4x4]

Hi Sharktooth !
For the last two days I've been testing EQM V3ULR. The original DVD movie 110 min long, with average motion. AR=1.88 (cropping). I decided to make 1 CD bacup using XviD 1.0.2 with 2-pass encoding. Audio vbr mp3 96 kbps. I tried to make a backup without resizing and also did some tests with different resize values (608 and 544). Encoder settings:
QP OFF
GMC OFF
B-VOPs 2/1.5/1
CG ON
MSP 6
CM ON
VHQ 2
MIfi 300
Q I/P/B 2-31
Trellis OFF
The compressibility tests were between 40-68% (depending on the resolution) and they were always about 12% higher than for Jawor's matrix. The problem is that I always got oversized files up to 760 MB. When useing Jawor's matrix, I got proper sizes of files. I don't quite understand what is the reason.

[Sharktooth]

Weird. Have you tried it with different movies?
However try playing with the overflow treatment options.

[JarrettH]

Going to use any of these for your comparison Soulhunter?:confused:

[mateo4x4]

Have you tried it with different movies?
I've just done 4 encodings for the same movie. I didn't want to use a piece of Vob doing this kind of tests, because I think they are not thruthful enough.

[Sharktooth]

Originally posted by JarrettH
Going to use any of these for your comparison Soulhunter?:confused:
Yes :)

[Sharktooth]

EQM V3EHR is no longer "beta" and matrices for "streaming bitrates" are coming ...

@mateo4x4: Any news?

EDIT: There won't be matrices for streaming bitrates at this time...

[mateo4x4]

Hi !
I had some problems with network connection and I was also quite busy by writting one article. But today I'm going to compress another 2 movies and use EQM V3ULR.

[Sharktooth]

Tnx for your effort. I decided to investigate a bit more on the V3ULR matrix. I'm not 100% convinced it's doing good...

[mateo4x4]

Another two movies compressed.
Both long films (118 and 124 mins).
I decided to choose one 4:3 and one 16:9.
This time I did not resize at all.
Average bitrate for video 750 kb/s
XviD settings the same as in my first post.
This time both files got ideal sizes (700 and 701 MB).
So, the reason of previous oversizing is still enigmatic. I'm not sure, but it's quite possible that I've set min. quantisers to 1 for my first encoding. But it shouldn't be the reason. I think I'll try to make same more conversions. I like this kind of matrices, because most often I make 1 CD backup.

[Sharktooth]

The oversize was probably due to the source.
In very particular situations the codec rate control simply fails.
The easiest solutions for ovesized encodings are 2: limit the quantizer to 2-31 or play with overflow treatment settings.
BTW, a new V3ULR is coming. A new softer matrix will completely replace it.
I've noticed too much blocking due to the coefficient distribution.
The new V3ULR won't be HVS tuned but aimed to reduce blocking and artifacts.

EDIT: V3ULR has been just updated.

[mateo4x4]

BTW, a new V3ULR is coming. A new softer matrix will completely replace it.
I've noticed too much blocking due to the coefficient distribution.
The new V3ULR won't be HVS tuned but aimed to reduce blocking and artifacts.
Grand news for me :)

[Sharktooth]

Just updated.
Dont worry about the compressibility tests... do the whole encode and look at the final quality:)

[mateo4x4]

new V3ULR - 4 encodings and super results. Really good matrix for 1 CD compression and low bitrate :)

[Sharktooth]

tnx for your feedback! There were 2 versions of V3ULR and i wasnt able to choose which one should be the official ULR. Theoretically speaking the first one should provide a better quality so it was my first choice.
After some negative feedbacks i immediately nuked it coz of the artifacts.
HVS matrices are useless at the lowest bitrates coz there are not enaugh details...
However can you please post your xvid settings and some screenshots of your encodes with V3ULR?

[mateo4x4]

No problem. I've just started another encodig. I tried to reach the compressibility test over 66% and then set all quantizers to 1-3.
I decided to go fishing today so I will try to post all results after I come back from the lake :)

[Soulhunter]

ShamelessPlug*

Our CQM comparison V3 (http://forum.doom9.org/showthread.php?s=&threadid=84276) started... :D

* The ShamelessPlug is copyrighted by mf


Bye

[Didée]

Is there an error in the forum's database? I can access the "CQM comparison V3" thread through above link. But it doesn't show up for me in the thread list of the XviD forum :confused:

[Soulhunter]

Originally posted by Didée

Is there an error in the forum's database? I can access the "CQM comparison V3" thread through above link. But it doesn't show up for me in the thread list of the XviD forum :confused:

You cant see this... :confused:

http://img14.exs.cx/img14/3817/336.th.png (http://img14.exs.cx/my.php?loc=img14&image=336.png)

Should I send a pm to a mod or so ???


Bye

[Mnl]

I can't see the thread either :confused:

[unmei]

true, i see thread downto changed 18th oct on the first page - but no "CQM comparison V3" (22th oct) anywhere :/

[Teegedeck]

The thread is there; make sure you're not viewing the page from your browser's cache. Click the 'reload'-button. Also make sure you're not using a proxy. Another option may be that one of this site's proxies lags a bit behind. If you don't succeed with the above proposed, I'll pass it on to our tech department.

[Soulhunter]

Hmm, is it normal that my thread has no icon @ the left side... :confused:

http://img14.exs.cx/img14/3817/336.png

Btw, I cant see it @ my own "Subscribed Threads" page... :eek:

Maybe you can see it coz you are a mod or so ???

Or have I enabled the "hide thread" option... :D


@ All

Sorry for all this off-topic stuff !!!


Bye

[Teegedeck]

Uh-oh. There is a problem indeed. I cannot reply in that thread! 'Invalid thread specified'.

[Soulhunter]

If you are able to delete it, I could simply re-post it !!!


Bye

[Teegedeck]

That doesn't work either. Please be patient until someone more into this PHP stuff than me or Koepi has had a look at it. Thanks.

Edit: Oh, what the heck! Why don't you post it again, now? :)

[Soulhunter]

Is it possible to post a thread with the same title as a "existing" one... :confused:

Hmm, I will simply try n' see !!!


EDIT: Yay, seems to work ... :)


Tia n' Bye

[Teegedeck]

Well, it worked. :D The other thread will either be eventually deleted or simply wander off into oblivion.

[mateo4x4]

EQM V3 ULR

The results of my tests:

Compressibility: about 16% higher than HVS Best and 0,2% higher than
Jawor's 1CD QM.
4 different DVD sources compressed.
All XviD settings: default
Some scenes were encoded with resolution of 720x* and some resized to
640x*
Av. bitrate 850-950 kb/s
No additional AVS filters used.

In my opinion it works really fine and is much better than previous ULR matrix. The picture is much cleaner.

Download screens (http://www.dreamland.neostrada.pl/Screens_updated.zip)- updated

[Sharktooth]

This version is performing really better than the previous.
Thanx for the screenshots (the .zip is broken) and for your feedback.
Once the Soulhunter's Custom Quantization Matrices test will end i could "retouch" some of the V3 matrices (expecially the high bitrate ones).

[mateo4x4]

(the .zip is broken)
You mean it doesn't work ? Any problems with unpacking this file ?

[Sharktooth]

i had to repair it to see the screenshots but 1 is corrupted.

[mateo4x4]

Hmmm.. I really don't know what is the reason :confused:
I tried to download, unpack .zip and all of the screenshots are fine.
In 10 min time I'll try to add same more screens and make a new .zip file under that link.

[Ark]

Originally posted by mateo4x4
Hmmm.. I really don't know what is the reason :confused:
I tried to download, unpack .zip and all of the screenshots are fine.
In 10 min time I'll try to add same more screens and make a new .zip file under that link.

Err.. the new .zip is corrupted too.. :(

What're you using to create the .zip file? (Windows i guess)

[Sharktooth]

Corrupted :(
I fixed it with winrar.

[mateo4x4]

What're you using to create the .zip file? (Windows i guess)
No, I use Winrar 3.30.
Really strange. I have no idea what is the reason :confused:

[708145]

I was able to download both correctly, although it took several times with the first one.

Look here (http://www.funknmary.de/data)

bis besser,
T0B1A5

[Sharktooth]

Maybe it's the server. Who knows?

@Tobias: I will PM you tomorrow with a LONG explanation... :)