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

[NeD tHe OnE]

What does VHQ=4 exactly do?

more compression?

Right now I encoded a video MPEG1 format to mkv using ur Xvid settings ... I had aimed for 929 kbps and I got 528 kbps .. The quality wasnt bad nor too good,

So should check with VHQ=1 or 2? and what good will it do?

[NeD tHe OnE]

Thx foxyshadis ...

I got near .. i mean .. i got around the desired bitrate .. a bit oversized by a 2 mb

VHQ=1 worked

[NeD tHe OnE]

Should I try VHQ=2?

what is the difference between VHQ=2 & VHQ=1?

[foxyshadis]

Try it if you want, if the video is very clean it might help, but probably won't noticeably at that bitrate.

If it's oversized by 2 megs, try lowering the second pass bitrate by, say, 5-10 kbps.

[Sharktooth]

you can even keep min quant = 1 with ULR.

[jsquare]

Which one of these ULR matrices is compatible with SAP?

I tried V3ULR, which is my favorite but didn't work with my Philips DVP642.

[BigDid]

Which one of these ULR matrices is compatible with SAP?

I tried V3ULR, which is my favorite but didn't work with my Philips DVP642.
Hi,

This is related to your SAP chipset: ESS Vibrato.

Basically, you need to stick with the regular xvid matrix: H263 or MPEG and limit the max bitrate to 4000kbp/s by, IE, using "the home theatre" profile, PAL or NTSC;

Links:
http://forum.doom9.org/showthread.php?p=645982#post645982
http://forum.doom9.org/showthread.php?s=&threadid=82765
http://forum.doom9.org/showthread.php?t=94536&highlight=dvp642
http://forum.doom9.org/showthread.php?t=103972&highlight=dvp642
:search:

Did

[Sharktooth]

As BigDid said the chipset of your standalone player does not support custom quant matrices.
However check the first posts in this thread. There you'll find what matrices are compatible with SAPs which support custom quant matrices.

[jsquare]

As BigDid said the chipset of your standalone player does not support custom quant matrices.
However check the first posts in this thread. There you'll find what matrices are compatible with SAPs which support custom quant matrices.
Thanks guys, I was thinking of getting rid off the Philips(lack of support) and get another brand.

[BigDid]

Thanks guys, I was thinking of getting rid off the Philips(lack of support) and get another brand.
Hi,

If I may quote Sharktooth :) as sharktooth said "you'll find what matrices are compatible with SAPs which support custom quant matrices."

If you want a SAP with more compatible functions/items/possibilities and Cqm :) , the MTK 1389 chipsets are quite good. I have a H&B 3220 not aging fine (2 years old) but can play (nearly)whatever i give it except GMC or Ogg. I am waiting for the replacement that will be a Peekton PK 6006 same chip family, more compatible, firmware upgradable.

Links in the hardware section:
http://forum.doom9.org/showthread.php?t=98228
http://forum.doom9.org/showthread.php?s=&threadid=67028
and specially this one to compare different players in english or french:
http://www.divxtest.com/form/divxtest2_list.php?lang=eng

More on SAP subjects may be more suited in the hardware section. Enjoy.

Did

[Backflip]

Simply use (PAL or NTSC) 16:9 Pixel Aspect Ratio. That'll be enough.

Ok, I'll try that. It's just that to me it seemed incorrect because for example, I've cropped The Machinist down toe 720x432 (all black bar and messy pixels gone). Approximate aspect of the movie is 2.35:1 (DVD is R4 PAL), but when I use PAL 16:9 the aspect the resulting aspect is 2.42:1.

2.42:1 is totally correct?

ps - incidently the retail region 4 (PAL) DVD of The Machinist has 1.85:1 AR printed on the box :P Obvious typo.

[Terka]

which matrix is suitable for interlaced Xvid encoding (DV source), how high bitrate (or quantetizer) should i use?

[Backflip]

2.42:1 is totally correct?

Sorry to bring this up again, but now, just browsing the forums I've found this old post (http://forum.doom9.org/showthread.php?t=54677) where ChristianHJW encoded a sample anamorphically, but as you can see he suggested setting (display) aspect ratio as 2.35. Why in this example was PAL 16:9 PAR not used? Is it not the best option with regards to getting correct aspect?

I would really, really like to know if I correctly crop etc that simply setting the PAR as PAL 16:9 will definitively be the correct way to retain best aspect ratio. Can anyone confirm this so I can go along my merry way encoding without any doubts as to whether I'm doing things totally wrong or not.

Currently I've got Star Wars episode III encoded, setting PAR as 16:9 PAL gives me an aspect of 2:46:1 on playback (704x416 => 1024x416)

Again I apolagise for bringing up the same question over and over again, but it really takes a long time to encode stuff and I'd like to be certain that it's right instead of coming back to it in the future with the knowledge that it's infact wrong :(

Thanks very much :)

[dado023]

hi

i have to encode 48min long video(interlaced source) @910kbps
encoded size is cca 355mb

is v3ulr matrix good for this task and what is more important will it work on yamada 6600 divx player, coz i wanna watch it on a TV....


tnx in advance

[Sharktooth]

which matrix is suitable for interlaced Xvid encoding (DV source), how high bitrate (or quantetizer) should i use?
hi

i have to encode 48min long video(interlaced source) @910kbps
encoded size is cca 355mb

is v3ulr matrix good for this task and what is more important will it work on yamada 6600 divx player, coz i wanna watch it on a TV....


tnx in advance

:readguid: :search:

[Sharktooth]

Sorry to bring this up again, but now, just browsing the forums I've found this old post (http://forum.doom9.org/showthread.php?t=54677) where ChristianHJW encoded a sample anamorphically, but as you can see he suggested setting (display) aspect ratio as 2.35. Why in this example was PAL 16:9 PAR not used? Is it not the best option with regards to getting correct aspect?

I would really, really like to know if I correctly crop etc that simply setting the PAR as PAL 16:9 will definitively be the correct way to retain best aspect ratio. Can anyone confirm this so I can go along my merry way encoding without any doubts as to whether I'm doing things totally wrong or not.

Currently I've got Star Wars episode III encoded, setting PAR as 16:9 PAL gives me an aspect of 2:46:1 on playback (704x416 => 1024x416)

Again I apolagise for bringing up the same question over and over again, but it really takes a long time to encode stuff and I'd like to be certain that it's right instead of coming back to it in the future with the knowledge that it's infact wrong :(

Thanks very much :)
this thread is about my custom matrices not anamorphic encoding...

[CrashUK]

hi
is v3ulr matrix good for this task and what is more important will it work on yamada 6600 divx player, coz i wanna watch it on a TV....


tnx in advance

also tested the matrix on a lot of divx dvd players and they all work.. if you know what your doing xvid can be better for dvd/divx hardware then divx6.

[dado023]

:thanks:
@CrashUK
i was waiting for days for some1 to reply me, this was my home page coz i was waitin for answer.....a million thanks

@sharktooth
i always search and read the guides since the time of divx3 sbc.......but if missed something does that mean i have to read 10 times or more.....
i apologise for this but i have to say your attitude is so geek....

best regards

[Sharktooth]

well, that's not my attitude... it's answering almost the same questions 1.000.000 of times that hurts...
however, there are rules on this forum... and guess what's the first rule?
1) Read up before asking! There are tons of guides, FAQs and each forum has a dedicated Q&A thread with additional info.
1a) Use the search function before posting. Chances are your question has already been answered.

[fight2win]

for a 2-hr. movie, 704X384 resize, 1466 mb ac3 2-ch xvid rip, which matrix would u guys advise?

[Sharktooth]

exactly how many atoms of materia are in the air your breathing?
if you can answer that i will answer your question...

[fight2win]

exactly how many atoms of materia are in the air your breathing?
if you can answer that i will answer your question...

i am not demanding an answer, i just wanted advice, i thought forums did that, but this looks like a special case!
Anyways,:thanks:

[Sharktooth]

without info on the compressibility of the source i cant give you any answer.

[BigDid]

for a 2-hr. movie, 704X384 resize, 1466 mb ac3 2-ch xvid rip, which matrix would u guys advise?
Hi,

No precise answer, cause Sharktooth comment is true, but some pointers in this post (http://forum.doom9.org/showthread.php?p=799006#post799006) from 1 page behind for low to mid bitrates.
For Mid to mid-high bitrates see comments from Sharktooth the post right above concerning LR, V2 and HR.

This thread is only 14 pages long and lots of interesting things discussed, may I suggest you take the time to read it?

Did

Edit, just posted on a similar subject here (http://forum.doom9.org/showthread.php?p=825536#post825536)

[fight2win]

i tried both eqm_v3lr and hr, got lots of blockiness and artifacts, guess i'll go with mpeg matrix

[Sharktooth]

that means you need a lower bitrate matrix.
mpeg wont help...

[Livesms]

that means you need a lower bitrate matrix.
mpeg wont help...

Tell me please - which matric should i use for realtime video 720*576 capturing with Xvid with Q=1-2 (bitrates 3000 optimum-5000 max)?

[Sharktooth]

http://forum.doom9.org/showthread.php?p=825468#post825468

[Livesms]

http://forum.doom9.org/showthread.php?p=825468#post825468

This case is for 2-pass encoding and I'm capturing from TV tuner and only 1-pass Const bitrate or Const Quality can be used.

2-hr. for 1466 mb is 1500-1600 and i can afford 3000 or even 5000 in some case - 1 hour for 2048mb.

[BigDid]

i tried both eqm_v3lr and hr, got lots of blockiness and artifacts, guess i'll go with mpeg matrix

Is my english so bad? :(

From my previous post: some pointers in this post from 1 page behind for low to mid bitrates
The quoted post says this: 1/ULR3 More compressible, good sharp overall look, some ringing sometimes strong, sometimes soft.
2/Jawor1cd, 5.3% less compressible than ULR3 (on that clip), good overall look, softer than ULR3, nearly no ringing.
3/H263, 14.5% less compressible than ULR3, good overall look, softer than ULR3, nearly no ringing.
More: Edit, just posted on a similar subject here

The quoted post says this: So from more compressible/less details to less compressible/more details(usually):
V3ULR -> more details than Jawor but more ringing
Jawor 1CD -> less compressible than V3ULR(+-5%), less details, less ringing
H263<=>V3LR (quite same compressibility but not same usage )
V2 aka AGK sharpmatrix
V3HR

Thank you for your attention, or should I say please,pay attention :rolleyes:

Did

[Sharktooth]

There's also a description for every of my matrices matrices on their own post...

[Livesms]

Do any of this matrices speed up compression? Which of them will you suggest for real-time capturing from TV tuner with max quality?

[Sharktooth]

No speedup. Also you're really started annoying me with thos "what should i use...".
The answer is :search: &:readguid:
Once you start understanding how the codec works you'll know what to choose.

[Heini011]

i modded eqm_v3hr for nearly the same / slightly smaller filesize and simultaneously a bit better quality (more homogeneous quality, less bluring and less blocky) based on experience. the new matrices uses mainly less inter-frame compression and slighty less hf compression, cause that should be done by filtering in avisynth before. therefore i increased only intra-frame lf compression a bit.

heini_mr.xcm

8, 10, 11, 11, 12, 13, 14, 16,
10, 11, 11, 12, 13, 14, 16, 17,
11, 11, 12, 13, 14, 16, 17, 19,
11, 12, 13, 14, 16, 18, 20, 23,
12, 13, 14, 16, 18, 21, 24, 30,
13, 14, 16, 18, 21, 25, 32, 40,
14, 16, 17, 20, 24, 32, 43, 53,
16, 17, 19, 23, 30, 40, 53, 70

13, 14, 15, 15, 16, 18, 20, 22,
14, 15, 15, 16, 18, 20, 22, 24,
15, 15, 16, 18, 20, 22, 24, 27,
15, 16, 18, 20, 22, 25, 28, 32,
16, 18, 20, 22, 25, 29, 33, 42,
18, 20, 22, 25, 29, 34, 45, 56,
20, 22, 24, 28, 33, 45, 60, 74,
22, 24, 27, 32, 42, 56, 74, 98


and for lower bitrate encodings a similar hvs tuned "Six-Of-Nine" principle matrix for min quant = 3.

heini_lr_6o9.xcm

8, 8, 8, 8, 8, 9, 10, 12,
8, 8, 8, 8, 9, 10, 12, 13,
8, 8, 8, 9, 10, 12, 13, 14,
8, 8, 9, 10, 12, 13, 14, 17,
8, 9, 10, 12, 13, 15, 18, 22,
9, 10, 12, 13, 15, 19, 24, 29,
10, 12, 13, 14, 18, 24, 31, 38,
12, 13, 14, 17, 22, 29, 38, 49

10, 11, 11, 11, 11, 12, 13, 16,
11, 11, 11, 11, 12, 13, 16, 17,
11, 11, 11, 12, 13, 16, 17, 19,
11, 11, 12, 13, 16, 18, 20, 23,
11, 12, 13, 16, 18, 21, 24, 30,
12, 13, 16, 18, 21, 25, 32, 39,
13, 16, 17, 20, 24, 32, 41, 51,
16, 17, 19, 23, 30, 39, 51, 66

[BigDid]

... and for lower bitrate encodings a similar hvs tuned "Six-Of-Nine" principle matrix for min quant = 3...
Hi,

Do you mean lower bitrate than the original 6of9Hvs ?
In my (small) tests, at fixed Q=2 V3Hr had given me 2491Kbp/s and 6of9Hvs 3104kbp/s (Original 6of9 was 3164)
So I suppose your tuned 6of9 is somehow between 2500 to 3100 kbp/s...

Did

[Heini011]

Hi Did,

>Do you mean lower bitrate than the original 6of9Hvs ?

No. heini_lr_6o9.xcm is a matrix, which is similar hvs tuned as heini_mr.xcm, which is a slightly modded version of Sharktooth eqm-v3 hr matrix. from "6of9" is only the idea to increase min quant to 3 for a more consistent quantizer distibution.

i was never happy with the original six-of-nine, cause the encodings are blocky on motion. this is not the case with my matrix, but the price is ~ 7-8% lower compressability. but the compressabilty at q=3 is still ~ 24% better than heini_mr.xcm at q=2.

for high quality / high bitrate encodings we need min quant = 2 (cause all coefficients have to be >=8, dc closed)

greetings

[Teegedeck]

Interesting; I hope to find the time to test this eventually. :)

[BigDid]

i modded eqm_v3hr for nearly the same / slightly smaller filesize and simultaneously a bit better quality (more homogeneous quality, less bluring and less blocky) based on experience. the new matrices uses mainly less inter-frame compression and slighty less hf compression, cause that should be done by filtering in avisynth before. therefore i increased only intra-frame lf compression a bit.

heini_mr.xcm...
Hi,

Days are passing and still have no time to share an extensive review :(

A few words to let you know I tested your 2 Cqm.

The 6o9hvs_tuned is not compressible enough for me (too much bitrate needed)
The Hr_tuned (heini_mr) is promising.
I compared it more with the V2 than the original V3HR that your Cqm definitely outperform (for me)
The V2 is still very comparable and a little more compressible.

So, I will continue to use the V2 unless codec is saturating in which case I will switch to your heini_mr.
:thanks:

Did

[Heini011]

Hi Did,

i wrote it already: heini_lr_6o9.xcm is meant for minimum quantizer = 3. so you have to set quantizer = 3 in zones for first pass only and then 3 x Min Quant.. = 3 at more \ quantization. then it gives the desired compressibility.

eqm v3 is almost the same as v2, but has less inter-frame compression. heini_mr is a modded version from v3 and should be compared with that matrix, cause is gives round about the same compressability.

greetings.

[Plamen234]

Please, I need a piece of advice to choose a CQM in this particular case?
I'm asking this because from my preliminary figures I have to use Shraktooth's v3 ULR CQM @2300kbps which confuses me.

Details:
movie: Lord of the rings - The Two Towers SEE
target size: dvd-r (=3455MB for the video only)
resolution: 704x416 (cropping only) anamorphic 16:9
average bitrate: 2350kbps

Enc size prediction results (q=2):
mpeg -> 5314MB (and the target size 3455MB is 65% )
Shraktooth's v3 HR -> 6062MB (...=57%)
6of9 nonHVS -> 8015MB (...=43%)

StatsReader says from the 1st-pass file that:
Shraktooth's v3 ULR -> 4211MB (...=82%)
Shraktooth's v3 LR -> 4761MB (...=73%)

According to the firgures I should use v3ULR because it would be the least constrained choice. But Sharktooth has posted before that such bitrates are too high for ULR. So I should use the next bitrate-hungry CQM - v3LR.

Is this the rigth decision?
Any help is quite welcome.

p.s. xvid settings ASP5, B-frames 2-1-1, qpel, chroma optimiser ON, MSP=4, vhq=3, chroma motion ON, trellis on, turbo off, quants=2-31

[Teegedeck]

I'm not commenting on the comressibility of LOTR here - although I think it is quite normal - because I have yet to try and encode the whole movie (I only use selected scenes from the trilogy for testing), but just to say that what you found there is nothing extraordinary.

Actually you cannot say 'this matrix is meant for a bitrate of xyz kbps' at all. A 'high bitrate matrix' might produce a 'low' bitrate* and a 'low bitrate matrix' might produce a 'high' bitrate, depending on the source. The only thing you can tell for sure is that on the same source, the high bitrate matrix will always produce a higher bitrate than the low bitrate matrix.

* Take for example your typical Hollywood drama/romantic comedy; you'll easily be able to compress such a movie with a very-high-bitrate matrix at quantizer 3 to a filesize that you usually achieve with a low bitrate matrix at quantizer 3 for your 'average' movie.

[Plamen234]

So in this particular case, is LR the right choice for cqm? If no, which one would you recommend? Please.

[Teegedeck]

These percentage figures mean very little to me, sorry. I would aim for an average quantizer 3 and use the matrix which allows for that. I would prefer the LR matrix at an average quantizer 3 over the ULR matrix at an average quantizer of 2. But it doesn't seem you have that choice.

You could of course mildly denoise to improve compressibility. Search a bit and I think you will find ample discussions about how best to encode LOTR.

[Plamen234]

Thank you. I'll do it your way.

[G_M_C]

I do use custom matrices. Up till now ive used some of the custom matrices that have come with DVD-Rebuilder's Matrix editor (for CCE). They seem to do a good job, but I also want to try the matrices found in this thread.

Well i've read through most posts but I'm a bit at a loss. And it's quite a simple question: What would we consider Very High Bitrate or just HIGH or LOW Bitrate for that matter.

The question is not that simple to anwer offcourse, because 3600 Kbps @ standard PAL Anamorphic (720x576) could be considered very high bitrate i quess, but the same 3600 Kbps @ 1080p would be considered very low bitrate. The bitrate itself could better be mesured in bits per pixel I think; For instance; 720 x 576 = 414720 pixels, bpp = 3.600.000 / 414720 ~~ 8,68.

So getting bach to the question; Could someone post a "rule of thumb" on the in this thread mentioned bitrates; How many bpp would be considered EXTREMELY HIGH BITRATE, VERY HIGH BITRATE etc. With this "rule of thumb", it might become more easy to choose the proper Matrix for everybody, form novices to advanced user.

[foxyshadis]

You can see Teegedeck's presets thread for Enc compressibility levels and the "rules of thumb" he derives from them. Not to mention being much more accurate than bpf. Such a test would also probably be the best way to choose between these, although I usually just pick one based on some rough idea of what I want.

[G_M_C]

[...]although I usually just pick one based on some rough idea of what I want.

I do too, but it is quite possible to, for instance, "undersetimate" the possible settings.

In the previous example 3600kbps @ PAL anamorpic; I've always treated this as Very high bitrate, and chose a matrix that fitted that discription. But somehow "ive got the feeling" that I underestimeted this bitrate, and could have used the "EXTREMELY high bitrate matrix" found in this topic (and thus allmost encoded this "lossless").

A short overview or rule of thumb as I mentioned above, would have made shure i chose te right one (eliminating guessing like you and me do now ;) )


BTW; Short explination maybe required. I came to this high bitrate because I can get a documentary consisting of 3 eps @ 50 mins fitting on 1 DVDR ;) And i want as high as possible quality bacause the original is in 1080P format (and i dont have a HDTV, PAL-SD is max for me :'( )

[708145]

In the previous example 3600kbps @ PAL anamorpic; I've always treated this as Very high bitrate, and chose a matrix that fitted that discription. But somehow "ive got the feeling" that I underestimeted this bitrate, and could have used the "EXTREMELY high bitrate matrix" found in this topic (and thus allmost encoded this "lossless").


Is there any interest in an application that encodes to xvid q2 and automatically picks the CQM (or interpolates between existing ones) to hit the requested final size?

I'm willing to code this if there is interest.

bis besser,
T0B1A5

[G_M_C]

Is there any interest in an application that encodes to xvid q2 and automatically picks the CQM (or interpolates between existing ones) to hit the requested final size?

I'm willing to code this if there is interest.

bis besser,
T0B1A5

Offcourse a tool would be a nice thing. But i think there would be more "to do" as you might think.

For instance; I encode especially for "DivX Certified" players, mind you NOT DivX 6.x but the 5.x versions.

this means i cannot use stuff like Qpel, packed bitstream, GMC or more than 1 consecutive b-frame. I simply comes down to the matrix i choose, because that is allmost the only significant improvement i can make. And the other thing I'm interested in is QUALITY, speed of the encode doesnt matter at all to me (using some AviSynth filters slows things down anyway, so XviD encoding speed doesnt matter anymore ;) )

The reason I do this is that many home DivX players will not play the XviD's with the mentiones "advanced" settings; not even mine (a Denon !).

So when you make a tool like you purposed, you need to make different profiles too (like DivX certified-compliant, or pure XviD-compliant etc. etc.) to account for different players/capabillities. But if you can make such a thing, i think it will be a great help (for me at least).

But a simple list with some numbers (bpp, as i described) would also be good enough for me ;)

[foxyshadis]

EHR is more 10Mbps range. On a test video (11 Mbps EHR q2), using 3600 kbps results in video that's significantly more ringy and blocky than UHR, very high quants; it's really made for processing, not hvs, and shouldn't be used in final encodes unless you have the space for archival quality. There's a good chance it won't work on standalones too, the dividers are so low.