Jawor's 1CD Quantization Matrix [Archive] - Page 2

View Full Version : Jawor's 1CD Quantization Matrix

Pages : 1 [2]

Teegedeck

1st May 2004, 13:22

Friends, would you mind if I branch your fabulous DC-Tune discussion into a dedicated new thread? I'm sure lots of people don't even know this discussion is going on here, and also this used to be the thread about Jawor's 1 CD matrix, after all.

Edit: Done. Please continuue the DC-Tune discussion here (http://forum.doom9.org/showthread.php?&goto=lastpost&threadid=75416) .

rlin

6th May 2004, 20:57

Hello

Im all new to using a quantization matrix, and im not really sure what its about. But as far as i can understand this thread is about "inserting" the best settings for 1CD backups using custom settings for Xvid.

Today i encoded Brotherhood of the Wolf using your matrix and got fairly good results. The picture is however a bit to grainy at times. Should i use a deinterlace filter? Could that help your matrix produce better result? Another thing that went wrong was the target size, the file was about 100 mb to small. My target size was 717 mb and i got a 617 mb file? Ehh my Audio A says 117 mb in g-knot, could that be the problem? In the avi-overhead area is however no audio checked.

Btw, Brotherhood of the Wolf is 2h and 17m, and my screen-res is 640 x 272.

I find this matrix-stuff wery interesting and ill keep testing it, and in the end i might actually understand how it works.

/Skerno

DDogg

7th May 2004, 21:20

Jawor, seems to provide good compression when used for MPEG2. Compression was the only thing tested (http://forum.doom9.org/showthread.php?s=&threadid=75828).

Jawor

10th May 2004, 07:08

Originally posted by rlin
The picture is however a bit to grainy at times.
This matrix tends to keep grain indeed. Should i use a deinterlace filter?
Only if you can see interlacing artifacts on your source material.
Could that help your matrix produce better result?
To be frank, I have no idea. As a rule of thumb, always deinterlace/IVTC unless you want to keep the interlacing (using the Interlaced encoding option) for playback on an interlaced TV.
Another thing that went wrong was the target size, the file was about 100 mb to small. My target size was 717 mb and i got a 617 mb file?
My matrix is really compressible due to high quantizers.

Jawor

10th May 2004, 09:38

Here's another comparison I made for you, folks. This time I used a low bitrate on a high motion scene to create visible blocking artifacts. The test clip is The Matrix Reloaded (R2, PAL) chapter 20. I used XviD 1.0 RC4, VirtualDubMod 1.5.4.1, AviSynth 2.5.4.

AviSynth script:
Clip=MPEG2Source("D:\MATRIX_RELOADED_DISC_1\VIDEO_TS\reloaded.d2v")
Clip=Crop(Clip,0,76,-0,-76)
Clip=Deen(Clip,"c3d",1,4,6)
Clip=Crop(Clip,2,0,-0,-0)
Clip=UnDot(Clip)
Clip=LanczosResize(Clip,656,272)
Return Clip

XviD settings:
target bitrate 600 kbps
"fast" first pass
MSP=6, Chroma ME, VHQ=4
maximum I-frame interval 250
Quarter Pixel, GMC, Trellis ON
B-VOPs 2/1.00/1.00
Quantizers 2-31
Packed bitstream OFF
B-VOP Sensitivity 1

Quantizers for H.263:
average quantizer = 8.685
5 - 0.6 %
6 - 3.3 %
7 - 8.9 %
8 - 24.1 %
9 - 36.5 %
>9 - 26.6 %

Quantizers for MPEG:
average quantizer = 9.044
5 - 0.1 %
6 - 2.0 %
7 - 6.6 %
8 - 17.3 %
9 - 34.7 %
>9 - 39.4 %

Quantizers for Low Bitrate Matrix:
average quantizer = 9.012
5 - 0.1 %
6 - 1.7 %
7 - 6.9 %
8 - 18.6 %
9 - 35.1 %
>9 - 37.8 %

Quantizers for HVS - Good Picture:
average quantizer = 8.731
5 - 0.2 %
6 - 2.1 %
7 - 8.0 %
8 - 26.5 %
9 - 39.5 %
>9 - 23.7 %

Quantizers for HVS - Best Picture:
average quantizer = 9.176
5 - 0.1 %
6 - 1.6 %
7 - 5.7 %
8 - 14.6 %
9 - 34.1 %
>9 - 43.9 %

Quantizers for Jawor's 1CD Matrix:
average quantizer = 8.321
6 - 2.4 %
7 - 12.9 %
8 - 39.9 %
9 - 35.7 %
>9 - 9.0 %

And here are the screenshots (H.263, MPEG, Low Bitrate Matrix, HVS - Good Picture, HVS - Best Picture, Jawor's 1CD Matrix):

Series 1 - The Merovingian
Look at these interesting blocking patterns on Merv's face and on the wall

http://img7.imageshack.us/img7/8798/h263.png
http://img7.imageshack.us/img7/6057/mpeg.png
http://img7.imageshack.us/img7/907/lowbitrate.png
http://img7.imageshack.us/img7/2426/hvsgood.png
http://img7.imageshack.us/img7/5372/hvsbest.png
http://img7.imageshack.us/img7/9069/jawors.png

Series 2 - Debris
Blocking artifacts in the dust cloud and on parts of the marble statue flying around

http://img7.imageshack.us/img7/7519/h264.png
http://img7.imageshack.us/img7/1418/mpeg1.png
http://img7.imageshack.us/img7/5093/lowbitrate1.png
http://img7.imageshack.us/img7/7023/hvsgood1.png
http://img7.imageshack.us/img7/1302/hvsbest1.png
http://img7.imageshack.us/img7/3239/jawors1.png

Jawor

10th May 2004, 09:39

Series 3 - Anti-Gravity
Compare the blocking artifacts behind the guy's butt and above his sword

http://img7.imageshack.us/img7/4639/h265.png
http://img7.imageshack.us/img7/8527/mpeg2.png
http://img7.imageshack.us/img7/4420/lowbitrate2.png
http://img7.imageshack.us/img7/4616/hvsgood2.png
http://img7.imageshack.us/img7/3483/hvsbest2.png
http://img7.imageshack.us/img7/7177/jawors2.png

Series 4 - Neo vs. A Bad Guy
Blocks and mosquitos all over the frame

http://img7.imageshack.us/img7/8640/h266.png
http://img7.imageshack.us/img7/8954/mpeg3.png
http://img7.imageshack.us/img7/6818/lowbitrate3.png
http://img7.imageshack.us/img7/8555/hvsgood3.png
http://img7.imageshack.us/img7/6687/hvsbest3.png
http://img7.imageshack.us/img7/1300/jawors3.png

Jawor

10th May 2004, 09:44

Originally posted by DDogg
Jawor, seems to provide good compression when used for MPEG2. Compression was the only thing tested (http://forum.doom9.org/showthread.php?s=&threadid=75828).
No wonder, since I used high coefficients. I designed this matrix to provide good compressibility.

Thanks for testing my matrix :D

Jawor

10th May 2004, 14:02

Here's another series of screenshots from The Matrix Reloaded comparison (H.263, MPEG, Low Bitrate Matrix, HVS Good Picture, HVS Best Picture, Jawor's 1CD Matrix):

Series 5 - Cutting Edge
Even more blocky than the previous series
http://img4.imageshack.us/img4/8798/h263.png
http://img4.imageshack.us/img4/6057/mpeg.png
http://img4.imageshack.us/img4/907/lowbitrate.png
http://img4.imageshack.us/img4/2426/hvsgood.png
http://img4.imageshack.us/img4/5372/hvsbest.png
http://img4.imageshack.us/img4/9069/jawors.png

Soulhunter

11th May 2004, 03:04

@ Jawor

I wanna make sure you dont miss this (http://forum.doom9.org/showthread.php?s=&threadid=76021) !!!

Just coz your matrix got some real good results... ;)

Bye

Jawor

12th May 2004, 07:10

Here's another test: Star Trek: First Contact (R2, PAL) chapter 4. This chapter has some static scenes with panning and a scene full of smoke and shaky camera movements during the space battle. I was particularly curiuos how do the matrices cope with the smoke, so I took just one series of screenshots.

AviSynth script:
Clip=MPEG2Source("D:\DVD_VIDEO\VIDEO_TS\first_contact.d2v")
Clip=Crop(Clip,2,74,716,432,align=true)
Clip=Tweak(Clip,sat=0.75,bright=15)
Clip=Deen(Clip,"c3d",1,4,6)
Clip=UnDot(Clip)
Clip=LanczosResize(Clip,640,272)
Clip=UnDot(Clip)
Return Clip

XviD settings:
target bitrate 500 kbps
"fast" first pass
Qpel, GMC, Closed GOV, Trellis ON
B-VOPs 2 / 1.50 / 1.00
B-VOP Sensitivity 2
AQ, interlaced encoding, packed bitstream, Turbo, chroma optimizer, greyscale OFF
MSP=6, ChromaME, VHQ=1
Quantizers 2-31
maximum I-frame interval 250

The first five screenshots (H.263, MPEG, HVS Good Picture, HVS Best Picture, Jawor's 1CD Matrix):

Look at the blocky cloud on the left and notice that the rest of the frame looks much better.
http://img8.imageshack.us/img8/8798/h263.png
http://img8.imageshack.us/img8/8954/mpeg3.png
http://img8.imageshack.us/img8/8555/hvsgood3.png
http://img8.imageshack.us/img8/6687/hvsbest3.png
http://img8.imageshack.us/img8/1300/jawors3.png

Be patient, some other matrices are currently being tested with this clip.

Jawor

12th May 2004, 18:31

Some more screenshots from the Star Trek: First Contact comparison (Andreas 78er, Andreas einfache 99er, Andreas doppelte 99er, Bulletproof's Heavy Compression, Bulletproof's High Quality, Angel BestLow):
http://img5.imageshack.us/img5/220/Andreas78.png
http://img5.imageshack.us/img5/1443/Andreasein99.png
http://img5.imageshack.us/img5/8299/Andreasdop99.png
http://img5.imageshack.us/img5/1578/BulletproofHeavy.png
http://img5.imageshack.us/img5/5067/BulletproofHigh.png
http://img5.imageshack.us/img5/2617/AngelBestLow.png

Jawor

13th May 2004, 11:57

Even more screenshots from Star Trek: First Contact (Angel BestVeryLow, Low Bitrate Matrix, Very Low Bitrate Matrix, Ultra Low Bitrate Matrix, Selur's Matrix, Ultimate Matrix, EthanoliX FourRooms):
http://img16.imageshack.us/img16/6807/AngelBestVeryLow.png
http://img16.imageshack.us/img16/159/LowBitrate.png
http://img16.imageshack.us/img16/245/VeryLowBitrate.png
http://img16.imageshack.us/img16/5589/UltraLowBitrate.png
http://img16.imageshack.us/img16/7/Selurs.png
http://img16.imageshack.us/img16/2436/Ultimate.png
http://img16.imageshack.us/img16/8190/EthanoliXFourRooms.png

odysseus

22nd May 2004, 22:34

Is there a suggested quantization for mostly high motion?

Not movies with occasional high motion but lots of motion all the time?

Like handheld cams with lots of global motion, but also besides the global motion, lots of motion in the scene itself.

Like for example following soccer players and not recording during breaks (so all the low motion parts are gone & the video is all high motion, with rare low-motion sequences).

Things like Matrix have lots and lots of low motion scenes, to get the best compression you need settings that compress both low and high motion well.

What I'm looking for is, I don't care if the settings do badly on low motion but I'd love them to do well on high motion.

Jawor

24th May 2004, 07:04

All that I can tell you is that my matrix copes quite well with high motion. Look at the screenshots from The Matrix Reloaded.

iago

18th June 2004, 12:57

Additionally I can say that it copes with low motion quite well too. Finally I had the time to give Jawor's matrix a try (together with his latest XviD_15062004_p3 build).

I used a resolution of 640*352 with LanczosResize accompanied by undot only and headed for a 1CD encode of a low-medium motion, mainly dark source with XviD settings: AQ - Trellis - VHQ1 - cm - b-frames 2/1.50/1.00.

And I'm really impressed by one of the most visually pleasing results I've got so far. At least I can say that Jawor's matrix is actually fine-tuned for 'my' eyes :). Ah, also for compressibility I guess, but without losing the good details and the natural look especially on faces.

Good work. Thanks.
iago

CruNcher

19th June 2004, 07:26

but without losing the good details and the natural look especially on faces.

you should better watch a 2 time :P

Jawor

19th June 2004, 11:52

Originally posted by iago
At least I can say that Jawor's matrix is actually fine-tuned for 'my' eyes :).
That makes two of us :D
Originally posted by iago
Ah, also for compressibility I guess, but without losing the good details and the natural look especially on faces.

To be frank, I was surprised when I saw, that the matrix keeps so much detail. First I thought it would be sort of HVS - Worse Picture due to quite high coefficients I used.

@CruNcher: Maybe CruNcher's Visual System is different than iago's Visual System and Jawor's Visual System :D

@iago: Thanks for testing both the matrix and the VBV build.

Teegedeck

22nd June 2004, 19:12

I don't really encode at those bitrates but still: Damned fine work; good man! That makes for my new recommendation for low-bitrate-matrices.

Jawor

23rd June 2004, 11:03

Originally posted by Teegedeck
I don't really encode at those bitrates but still: Damned fine work; good man! That makes for my new recommendation for low-bitrate-matrices.
I'm just trying to do my best :D

Alxemi

26th June 2004, 16:19

What about a Jawor´s 2CD quantization matrix? :)

Jawor

28th June 2004, 13:28

Originally posted by Alxemi
What about a Jawor´s 2CD quantization matrix? :)
I'm not designing any quantization matrix at the moment. Of course you can use this 1CD matrix for 2CD encodes :D

But seriously, I wouldn't recommend this.

spyto

29th June 2004, 12:47

Originally posted by Alxemi
What about a Jawor´s 2CD quantization matrix? :)

2CD, you may use HVS Matrix ;)

Jawor´s 1CD is a best at low/medium bitrate.

srk3461

22nd March 2011, 16:50

Soul hunter the link is dead.... !
can u provide a nw link for the The Evil Matrix - v0.1 ... ? thnx in advnce !!!

Soulhunter

22nd March 2011, 19:16

Soul hunter the link is dead.... !
can u provide a nw link for the The Evil Matrix - v0.1 ... ? thnx in advnce !!!
Which link? >.>

srk3461

23rd March 2011, 05:21

hey thnx fo th reply.............
i need downld link fo evil matrix of urs its gud,,jus want try it out..... ! u posted th downld link fo th evil matrix in th first page of this discussn bt its dead...! can u post it agin ??

Didée

23rd March 2011, 12:41

Found it in the guts of an old HDD. Evil.v0.1.qmatrix (http://www.mediafire.com/?m3iprnpu5f35u4g)

Just FYI, this is evil v0.1:

http://img813.imageshack.us/img813/2364/evilv01qmatrix.png (http://img813.imageshack.us/i/evilv01qmatrix.png/)

srk3461

24th March 2011, 15:44

hey Didée thnx fo th link :D !
dnt wrry i'll keep it safe...... xD !

orion44

18th April 2017, 15:10

hartford

25th July 2017, 20:23

This is all that I have; Hope it helps.

evil v0.1

Intra:
8,12,12,12,20,60,255,255
14,10,12,12,14,34,255,255
12,12,20,22,30,255,255,255
12,12,20,46,255,255,255,255
16,12,28,255,255,255,255,255
56,32,255,255,255,255,255,255
255,255,255,255,255,255,255,255
255,255,255,255,255,255,255,255

Inter:
12,18,18,20,38,255,255,255
18,14,18,18,26,255,255,255
18,18,30,42,82,255,255,255
18,18,36,255,255,255,255,255
30,24,80,255,255,255,255,255
255,255,255,255,255,255,255,255
255,255,255,255,255,255,255,255
255,255,255,255,255,255,255,255

orion44

29th July 2017, 04:43

Brazil2

8th August 2017, 16:28

I've found it in my archives.