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

[XiOUS]

I made a DVD -> XViD copy of one of my DVD-9 R2 (PAL) movies yesterday following the guide located here (http://www.videohelp.com/forum/viewtopic.php?t=221896)
Using DGIndex (http://neuron2.net/fixd2v/decodefix.html) (instead of DVD2AVI) VitrualDub 1.5.10 and VirtualDubMod 1.5.10.1 (for muxing AVI with AC3), I used latest (to my knowledge) XviD-1.0.2-29082004.

As "Quantization type" I used MPEG (as it was recommended for higher bitrate), I guess I should have used your V3HR or VRUHR matrices now when I found out about them, right? I used bitrate 2400kbps (shows up as 2100 kb/s in GSpot v2.21). Which matrices would u recommend using with that kind of bitrate? And do you have any recommendations other then default settings when making XViD with that high bitrate?

My other question is that my source was 720x576@25fps (PAL), in DGIndex when using the cropping feauture (to take away small black bars at top and bottom) I could crop 16x2 (16 at top and 16 at buttom) so it showed up as 720x544 (576-16*2=544). I guess the source is anamorphic DVD, so it uses 544 horizontal lines (and 720 vertical lines)? But when following the guide to keep original Aspect Ration of 1.85:1 (says on DVD-box) I would make 720/1.85 = 389 horizontal lines (after cropping), which is far less then original 544. To have a height that is a multiple of 8 I decided upon 392 horizontal lines (720/392 = 1.84:1). It all looks alright, but my question is that when encoding DVD to XViD (or DivX) you actually loose horizontal resoultion on anamorphic sources right? So transparency between DVD and XViD isn't possible?

[PiXuS]

Originally posted by XiOUS
I made a DVD -> XViD copy of one of my DVD-9 R2 (PAL) movies yesterday following the guide located here (http://www.videohelp.com/forum/viewtopic.php?t=221896)
Using DGIndex (http://neuron2.net/fixd2v/decodefix.html) (instead of DVD2AVI) VitrualDub 1.5.10 and VirtualDubMod 1.5.10.1 (for muxing AVI with AC3), I used latest (to my knowledge) XviD-1.0.2-29082004.

Hmmm... I am not sure VDM 1.5.10.1/2439 is a good tool to mux AC3 with AVI. But, I am really not 100% sure. Do a search about that and AviMuxGUI.

Originally posted by XiOUS


As "Quantization type" I used MPEG (as it was recommended for higher bitrate), I guess I should have used your V3HR or VRUHR matrices now when I found out about them, right? I used bitrate 2400kbps (shows up as 2100 kb/s in GSpot v2.21). Which matrices would u recommend using with that kind of bitrate? And do you have any recommendations other then default settings when making XViD with that high bitrate?


There are MANY Custom Quant Matrices out there. Check that thread (http://forum.doom9.org/showthread.php?s=&threadid=73666) for example (in the first post there is a link to a collection of CQM). Also, do some research about Didée SixOfNine matrix. Some people (a few ;) ) still feel it is by far the best CQM for high-bitrate/high-quality/efficient encodes.

Originally posted by XiOUS

My other question is that my source was 720x576@25fps (PAL), in DGIndex when using the cropping feauture (to take away small black bars at top and bottom) I could crop 16x2 (16 at top and 16 at buttom) so it showed up as 720x544 (576-16*2=544). I guess the source is anamorphic DVD, so it uses 544 horizontal lines (and 720 vertical lines)? But when following the guide to keep original Aspect Ration of 1.85:1 (says on DVD-box) I would make 720/1.85 = 389 horizontal lines (after cropping), which is far less then original 544. To have a height that is a multiple of 8 I decided upon 392 horizontal lines (720/392 = 1.84:1). It all looks alright, but my question is that when encoding DVD to XViD (or DivX) you actually loose horizontal resoultion on anamorphic sources right? So transparency between DVD and XViD isn't possible?

I wouldn't crop with DGIndex. I would either use the wonderful AutoCrop plugin or Gordian Knot to do that job. Nevertheless, if after cropping a 16:9 movie the resolution goes from 720x576 to 720x544, then the AR becomes ((720/544)*1.823)/1.5 ~= 1.609. So, the resolution on playback would go from 1050x576 (AR=1.823) to 875x544 (AR=1.609). If you aim for an anamorphic encode, you shouldn't bother to keep an AR=1.823. As long as you know what the AR is after you crop, you can easily compute the required resolution on playback to keep the AR correct. Just test it yourself.

Make sure you crop using a mod16 x mod16 (or mod32 x mod16 if you want to be really safe) resolution.

Correct me if I am full of baloney. ;)

[Sharktooth]

Originally posted by XiOUS
As "Quantization type" I used MPEG (as it was recommended for higher bitrate), I guess I should have used your V3HR or VRUHR matrices now when I found out about them, right? I used bitrate 2400kbps (shows up as 2100 kb/s in GSpot v2.21). Which matrices would u recommend using with that kind of bitrate? And do you have any recommendations other then default settings when making XViD with that high bitrate?
I suggest you to use EQM V3HR.

Originally posted by XiOUS
My other question is that my source was 720x576@25fps (PAL), in DGIndex when using the cropping feauture (to take away small black bars at top and bottom) I could crop 16x2 (16 at top and 16 at buttom) so it showed up as 720x544 (576-16*2=544). I guess the source is anamorphic DVD, so it uses 544 horizontal lines (and 720 vertical lines)? But when following the guide to keep original Aspect Ration of 1.85:1 (says on DVD-box) I would make 720/1.85 = 389 horizontal lines (after cropping), which is far less then original 544. To have a height that is a multiple of 8 I decided upon 392 horizontal lines (720/392 = 1.84:1). It all looks alright, but my question is that when encoding DVD to XViD (or DivX) you actually loose horizontal resoultion on anamorphic sources right? So transparency between DVD and XViD isn't possible?
DVD image is resized while playing.
An exact copy means you don't resize and use the Aspect Ratio correction within the codec configuration dialog.
That procedure is called Anamorphic encoding.
Use the forum search function for more info

[XiOUS]

Originally posted by PiXuS
Hmmm... I am not sure VDM 1.5.10.1/2439 is a good tool to mux AC3 with AVI. But, I am really not 100% sure. Do a search about that and AviMuxGUI.
I've muxed several AVI's with AC3's using VDM and there is no synch problem or any other problems when playing them on the computer. Haven't tested to play on standalone DVD-player because I have none that support Xvid, but it should work there to I guess?

Originally posted by PiXuS
I wouldn't crop with DGIndex. I would either use the wonderful AutoCrop plugin or Gordian Knot to do that job. Nevertheless, if after cropping a 16:9 movie the resolution goes from 720x576 to 720x544, then the AR becomes ((720/544)*1.823)/1.5 ~= 1.609. So, the resolution on playback would go from 1050x576 (AR=1.823) to 875x544 (AR=1.609). If you aim for an anamorphic encode, you shouldn't bother to keep an AR=1.823. As long as you know what the AR is after you crop, you can easily compute the required resolution on playback to keep the AR correct. Just test it yourself.
...
I don't crop with DGIndex (only check how much can be cropped) I do it in AVISynth-script like this one:
LoadPlugin("C:\DGDecode.dll")
mpeg2source("H:\Tsam\tsam.d2v")
crop(0,16,0,-16)
lanczosresize(720,392)
isn't that the best way to go? And by 720x544 I meant that DGIndex showed that resolution after cropping.
Ok to get AR 1.85 in original then horizontal resolution should be: x/576 = 1.85 -> x = 1066
with cropping (16x2) -> 1066/544 ~= 1.96 is the new AR.
So to keep all horizontal lines (544, with black bars cropped) in original DVD then I should encode in resoultion 1066x544, like this:?
LoadPlugin("C:\DGDecode.dll")
mpeg2source("H:\Tsam\tsam.d2v")
crop(0,16,0,-16)
lanczosresize(1066,544)
Or do horizontal resolution have to be a multiple of 8 too? Then I should use 1064x544 as 1064 is multiple of 8?
I guess there is no way to encode XViD or DiVX in anamorphic so that it's decoded like anamorphic DVD on DVD-player?

[SeeMoreDigital]

If a PAL 720x576 source was converted to square pixels, it would not exceed 1024 pixels wide, so your 1050 and 1066 calculations are slightly off.

Also, if the source image is exactly 1.85:1, the vertical pixel count should be around 560 (to the nearest 16th pixel). More often than not however, I doubt that the source will be exactly 1.85:1.


Cheers

[XiOUS]

Originally posted by SeeMoreDigital
If a PAL 720x576 source was converted to square pixels, it would not exceed 1024 pixels wide, so your 1050 and 1066 calculations are slightly off.

Also, if the source image is exactly 1.85:1, the vertical pixel count should be around 560 (to the nearest 16th pixel). More often than not however, I doubt that the source will be exactly 1.85:1.


Cheers
It says 1,85:1 on the DVD-box for the movie. But when I encode it to XViD I have to have a horizontal wide of around 1064 to keep all 544 horizontal lines intact, right? As anamorphic DVD's are resized in horizontal resoultion when played on 16:9 TV's, right?

[PiXuS]

Originally posted by SeeMoreDigital
If a PAL 720x576 source was converted to square pixels, it would not exceed 1024 pixels wide, so your 1050 and 1066 calculations are slightly off.

Also, if the source image is exactly 1.85:1, the vertical pixel count should be around 560 (to the nearest 16th pixel). More often than not however, I doubt that the source will be exactly 1.85:1.
Cheers

Can you point me to some thread or FAQ concerning this statement? I thought it was simply a matter of multiplying the horizontal lines by the PAR (1.823 if 16:9 and 1.367 if 4:3). For NTSC material, this means on playback a resolution of 875x480 for 16:9 stuff and 656x480 for 4:3 stuff. I find it weird, because I always computed the playback resolutions like so, and always got perfect results. 1024/586 is 1.77 which is 16/9, which isn't the TRUE ratio of a 16:9 presentation.

This guide on doom9 is saying what you are saying. (http://www.doom9.org/aspectratios.htm)

But then, read this thread (http://forum.doom9.org/showthread.php?s=&threadid=42708).


Originally posted by XiOUS

It says 1,85:1 on the DVD-box for the movie. But when I encode it to XViD I have to have a horizontal wide of around 1064 to keep all 544 horizontal lines intact, right? As anamorphic DVD's are resized in horizontal resoultion when played on 16:9 TV's, right?



Well, Goldfinger is a 1.85 presentation. That is close to 1.823 and you will notice in Gordian Knot preview mode that it fills the whole anamorphic frame (i.e.: if you are a PAL guy, it will basically use the whole 576 lines of resolution). A 2.35 presentation will really use less active lines than a 1.85 resolution (non-active lines are those which compose the black bars).

The way you computed 1066 is not correct. Let's use SeeMoreDigital numbers (PAR=1.77)

If you don't crop, the resolution on playback would be:

(1.77 * 576) x 576 ~= 1020x576

If you DO crop to 544 lines instead of 576, the resolution on playback becomes:

(((720/544)*1.77)/1.5) * 544 x 544 ~= 850x544.

And this is logic.. you have less columns because you have less lines (you got to keep the ratios right!).

Hope it helps!

[SeeMoreDigital]

PiXuS,

I have a link at the bottom of my "sig" that you may find helpful.

But in essence, digital images on DVD's contain pixels not lines. In the case of an PAL DVD, as you know it contains 576 vertical pixels.

576 multiplied by 1.7777 (aka: 16:9) equals 1024

In the case of an NTSC DVD, it contains 480 vertical pixels.

480 x 1.7777 = 853.3333

Don't forget the above calcs are based on converting your "anamorhpic" source images to "square" pixelled images.


Cheers

[PiXuS]

Originally posted by SeeMoreDigital
PiXuS,

I have a link at the bottom of my "sig" that you may find helpful.



I read the whole thing. But... to quote you: "In my opinion, the ITU standard is not really necessary when performing 'digital pixel to digital pixel' conversions.".

So, it is all about your humble opinion.

Which brings us back to square one.


But in essence, digital images on DVD's contain pixels not lines. In the case of an PAL DVD, as you know it contains 576 vertical pixels.


Do you know that pixels in a horizontal row form lines and pixels in a vertical row form columns?

FYI, each frame of a PAL DVD are made up of 720x576 pixels.


576 multiplied by 1.7777 (aka: 16:9) equals 1024

In the case of an NTSC DVD, it contains 480 vertical pixels.

480 x 1.7777 = 853.3333


We both know how to multiply!!

[SeeMoreDigital]

Try not to get yourself bogged down with pixel to line conversions!

Sooner or later the digital to analogue (PAL/NTSC CRT display) element of the image chain will be gone and all we'll be left with is pixels!

Do you have the new version of XviD-1.1.-127-13102004 (http://forum.doom9.org/showthread.php?s=&threadid=84039) and more importantly, its DSdec filter installed. If you have, then try playing these "pixel perfect" (http://forum.doom9.org/showthread.php?s=&postid=562776#post562776)720x576 and 720x480 anamorhpic encodes.

They have been generated using my calculations and the same theory apply to cropped encodes!

They've also been tested on Mpeg4/DVD stand-alone players that support the playback of anamorphic .AVI files.


Cheers

[PiXuS]

Originally posted by SeeMoreDigital
Try not to get yourself bogged down with pixel to line conversions!

Sooner or later the digital to analogue (PAL/NTSC CRT display) element of the image chain will be gone and all we'll be left with is pixels!


Is that Newspeak you are trying to impose on us? ;)


Do you have the new version of XviD-1.1.-127-13102004 (http://forum.doom9.org/showthread.php?s=&threadid=84039) and more importantly, its DSdec filter installed. If you have, then try playing these "pixel perfect" (http://forum.doom9.org/showthread.php?s=&postid=562776#post562776)720x576 and 720x480 anamorhpic encodes.

They have been generated using my calculations and the same theory apply to cropped encodes!

They've also been tested on Mpeg4/DVD stand-alone players that support the playback of anamorphic .AVI files.


Cheers

I don't have any beta versions of XviD. I am using the latest stable build from Koepi (1.0.2). I downloaded your 16:9 NTSC "TestCard" and put the AVI in a Matroska container. I created two files.. one with a resolution of 875x480 (1.823) and one with a resolution of 853x480 (1.777).

The 1.777 version appeared to be the correct one. Though, I will refer you to this thread (http://forum.doom9.org/showthread.php?s=&threadid=75998). From all I read on this wonderful subject, if the DVD was well mastered, you should follow the ITU standard and use a PAR of 1.823. If the DVD wasn't well done, use a PAR of 1.777. Geez... if you read the thread, you will notice even len0x seems to change is mind about following the ITU standard as often as he changes underwear.

So, that being said, I guess it all boils down to a case by case affair. Do a preview @ 1.777, then @ 1.823, find something which should be round and decide what seems correct.

[SeeMoreDigital]

Okay play the files in a software player that supports AR signalling detection by default, such as VLC media player or Nero's ShowTime player.

Why did you feel the need no mux my Mpeg4 streams into the MKV container and alter their AR... this is just asking for trouble!

And by the way, if you want your anamorphic Mpeg4 streams to play back at the correct AR in either AVI or MKV container, replace your XviD.ax file with this one (http://82.2.167.24/Uploaded_Files/Doom9_Forum_files/XviD_DSdec_filter_with_AR_detection_2.zip).


Cheers

[PiXuS]

Originally posted by SeeMoreDigital
Okay play the files in a software player that supports AR signalling detection by default, such as VLC media player or Nero's ShowTime player.

Why did you feel the need no mux my Mpeg4 streams into the MKV container and alter their AR... this is just asking for trouble!


Asking for trouble? Why would you say that? I don't want to take another tangent in our conversation (we already hijacked Sharktooth thread pretty badly :eek: ), but I don't understand your point regarding MKV. Try it.. it is very good.

I use MPC and MPC doesn't care about the PAR signalling in the stream so I needed to use MKV so that the PAR signal in the container could be detected and used.


And by the way, if you want your anamorphic Mpeg4 streams to play back at the correct AR in either AVI or MKV container, replace your XviD.ax file with this one (http://82.2.167.24/Uploaded_Files/Doom9_Forum_files/XviD_DSdec_filter_with_AR_detection_2.zip).


Well, okay.. thanks.. but.. my system is already pretty clutered as is. ;)


All that being said, I think XiOUS has enough pointers now to forge is own opinion on the subject. And.. btw... if you are afraid you make a mistake in your encode vis-à-vis a given PAR (say 1.777 or 1.823), just make an anamorphic encode. You then will always be able to change the PAR as you wish (even to some weird values). So you can't f*** up your encode for ever..

[ChronoCross]

really the mkv container shouldn't be used unless your dealing with multiple audio/subs. using it purely for the aspect ratio configuration serves no point. getting the aspect ratio to work fully in avi is probably in the long run going to be better. I support mkv for it's capabilities but it should be used for this purpose as a last resort.

[SeeMoreDigital]

PiXuS,

Unfortunately, it seems pretty obvious that you don't require any help in this matter!

This is strange because you asked why your anamorphic encodes don't come out properly... I've given you the answer and provided some anamorphic encodes for you to try!

I also find it strange how you expect to view anamorphic encodes correctly without installing the "all important" and "necessary" Mpeg4 direct-show decoder filter - which I also provided a link to!

Personally, I don't have any problems generating or playing back anamorphic encodes (via software or hardware players). Neither do many other forum members... Can you say the same?


Cheers

[PiXuS]

Originally posted by ChronoCross
really the mkv container shouldn't be used unless your dealing with multiple audio/subs. using it purely for the aspect ratio configuration serves no point. getting the aspect ratio to work fully in avi is probably in the long run going to be better. I support mkv for it's capabilities but it should be used for this purpose as a last resort.

Nah. I use Matroska for 1 avi + 1 ac3 + subs + chapters. I use mkvmerge to do the job of multiplexing everything. The tool is easy to use (especially through the mmg GUI!). I agree, if all I wanted to do was fix the AR, I would actually use the MPEG-4 container. I don't like the idea of using AVI.

Originally posted by SeeMoreDigital
Unfortunately, it seems pretty obvious that you don't require any help in this matter!

This is strange because you asked why your anamorphic encodes don't come out properly... I've given you the answer and provided some anamorphic encodes for you to try!

I also find it strange how you expect to view anamorphic encodes correctly without installing the "all important" and "necessary" Mpeg4 direct-show decoder filter - which I also provided a link to!

Personally, I don't have any problems generating or playing back anamorphic encodes (via software or hardware players). Neither do many other forum members... Can you say the same?


Huh... your tone gives a weird turn to our conversation. I am certainly not going to fight over aspect ratio stuff. (!!) I thought I had made it clear I feel the ITU standard should be followed when a DVD is properly mastered and not if not. You think 16/9 should be used all the time. Fine, I respect that (never thought I would need to explicitly tell a humain being I respect is choice of divisions....arg!)

You say: This is strange because you asked why your anamorphic encodes don't come out properly.... Huh... are you confusing XiOUS with PiXuS? I never really asked for your help...

Anyway... this is too weird. Me afraid, shudering and running away !!

:scared:

[SeeMoreDigital]

Can one of you re-name yourself to say... Stan :)

[LoKi128]

Just wanted to post some filesizes from my encodes of The Princess Bride (the softest, noisiest DVD I've ever seen):

@ 720x416

HR - 974,926,523
LR - 815,469,405
ULR - 773,913,937
ULR - 762,522,443 (using Deen then LimitedSharpen)

The first three have no processing except IVTC, Lanczos4 and UnDot. The last one I added Deen and LimitedSharpen before the resize.

The problem with this movie is that since it is so noisy, scenes that show the sky, or some other constant-color background, it seems that the brightness of that area fluctuates, and there is some blocking. The blocking is not so obvious in the HR matrix and very evident in the ULR matrix, as it averages more. To correct the problem i just used UnFilter, Resize to 640 and Undot. Deen was making the already soft image just look like a VHS transfer anyway.

Anyway, that ULR matrix is pretty good! I even made a 1CD encode of The Truman Show at 640 with the full AC3 that looked great! Thanks!

[SeeMoreDigital]

Originally posted by LoKi128
....The problem with this movie is that since it is so noisy, scenes that show the sky, or some other constant-color background... You've just reminded me that my DVD of "The Big Blue" also displays these kinds nasty effects...


Cheers

[Sharktooth]

@LoKi128: EQM V3ULR produces blocks when compressibility is really too low. Helping compressibility and removing noise with filters like UnFilter(x,x), UnDot() or RemoveGrain(mode=2) will produce a great looking 1CD backups.
Temporal denoisers like Convolution3d or TemporalCleaner will help a lot with extremely noisy sources.
Keep in mind most DVDs have added noise too and and low-bitrate matrices tends to quantize noise (high-frequencies) with higher coefficients (less details).
That may result in fluctuating blocks in flat or gradient coloured areas. If you want to remove blocks just use a high bitrate matrix (and 2 CDs or more CDs...) or remove as much noise as you can (with filters) but trying to keep the details.
Anyways dont confuse the noise with the film grain...

[Didée]

Originally posted by Sharktooth
Anyways dont confuse the noise with the film grain...
But the encoder will confuse them for sure. And it's the encoder which you want to help in producing a harmonious result.

Lately, I've come across the aiming "kill the noise, keep the grain" several times. But I still wait for an answer, from any of the proposers, to this question:

1. What's the difference, from a filter's or an encoder's point of view, between "noise" and "grain", so that either one shall be able to distinct between them two?

1a. What's the actual benefit of the distinction, which most likely will be very unsure anyways?


From my limited experience, impressing the audience with details on terminology helps the filter very little, and the encoder even less. Try encoding Alien² with "kill the noise, keep the grain". I give an ensurance on having BIG fun :)

[Teegedeck]

From the encoder's side there isn't really a difference, granted. It is more the viewer's impression that determines which is which. Still, the subjective difference can be quite staggering. A certain amount of 'grain' will usually create the (false) impression of a higher level of detail, while just a bit more 'grain' will seem annoying and hence be called 'noise'. I strongly feel that everything below the personal, subjective 'grain'-threshold should be kept, even if it means using a higher bitrate. My personal threshold of what I perceive as 'pleasant' grain is very low, of course. When I denoise I would not denoise to a perfectly clean picture (I guess you could summ that up as 'kill the noise, keep the grain').

Sure, I could add the noise on playback via ffdshow but it seems nonsensical to me to first remove information from the video by filtering out the noise in order to 'recreate' or better 'invent' a clean picture that wasn't really there in the first place, and afterwards try to attain a 'natural' (as I perceive it) look to it again. Even if this would safe me quite a lot of diskspace/bitrate.

Still, I've got to say I understand fully that others perceive this completely different - people I really respect, like -h for example, preferred to denoise video as much as possible.

Excuse this late-night rant.

Good night.

[Sharktooth]

IMHO noise and film grain are 2 different things.
They're distingushable coz the "dot" diameter is different: grain is "larger" while noise is much "smaller" (but sometimes not too much) and keeps randomly moving every frame.
My ideal filtering is exactly "kill the noise, keep the grain".

[Sharktooth]

IMPORTANT NOTE:
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 were 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.

[pwh04]

Just want to say thanks Sharktooth for these cqms. I'm a noob with custom matrices. I'm using bicubic sharp and tomsmocomp with no other filtering and I am very happy with the results.

[Sharktooth]

Thanks for your feedback pwh04

[Sharktooth]

I have updated EQM V3ULR. Any feedback will be appreciated.

[len0x]

Originally posted by Sharktooth
I have updated EQM V3ULR. Any feedback will be appreciated.

What are the changes and expected quality outcome?

[Sharktooth]

Better I-Frame quantization. Ark provided me a clip where I-Frames were blocky, and those blocks were compromising also the next P and B frames.
I think i have fixed it, but some feedback will be appreciated.

[Ark]

I'll try it on the same Matrix encode, same settings, to better viewing of changes

[dr.Prozac]

Hi !
About EQM V3ULR
I made many tests today. Really good matrix. This new one is about 1,5% less compressable than the previous version. It is always about 1,5% less compressable than Jawor's matrix. I made a comparison including eqm ulr, hvs good, hvs best and jawor's matrices. All XviD settings default. Just VHQ set to 2. We made many tests at our polish forum and got best results with vhq=2 so I always use it. Bitrate for all samples was 750-850 kbps. Beyond a doubt, eqm ulr brings the best quality. :)
Btw
Was this new one used in Soulhunter's Custom matrix comparison - V3 - Round 2 ? I voted yesterday.

[Ark]

I've done some tests with the new ULR, and it seems that the I-frames-blocking issue is almost resolved. I say almost because in comparing ULR with other few matrices (h263, mpeg, 6of9), still some "ringing" remain.

It's not blocking now, but it's like I-frames are sharper than other frames, but also more noisy, and the jump from I to the next frame is quite apparent (not in normal playback btw), something that doesn't happen with other matrices.

Even LR "suffers" from this, but you have to look really hard to see what i mean.

It seems to me that this is somehow costant about your matrices, but other than that i can't say if it's good or bad, as i said, in normal playback it's almost impossible to see it.

[acrespo]

I am searching for a custom matrix to Anime with 900-1100 kbps bitrate. I usualy do encodes with H.263 because any custom matrix I tried give me many block artefacts. My videos are analog capture episodies from VHS and resolution is 640x480, IVTCed filtered with Removedust (script from removegrain page), AAA script from Soulhunter, LimitedSharpen and FastLineDarken.

I do 2-passes because I need near 175Mbytes each episode:
Xvid settings:
VHQ 1
Chroma Optmizer
No B-Frame
Trellis
No Qpel
No Adaptative Quantization
No GMC
MSP 6
No Cartoon Mode

[Soulhunter]

My advice...

- Enable B-VOP's

- Stay with H.263 for anime @ low bitrates

- Go for 233MB and give Y.A.C.Q.M. a try... :D


Bye

[Sharktooth]

Originally posted by dr.Prozac
Hi !
About EQM V3ULR
I made many tests today. Really good matrix. This new one is about 1,5% less compressable than the previous version. It is always about 1,5% less compressable than Jawor's matrix. I made a comparison including eqm ulr, hvs good, hvs best and jawor's matrices. All XviD settings default. Just VHQ set to 2. We made many tests at our polish forum and got best results with vhq=2 so I always use it. Bitrate for all samples was 750-850 kbps. Beyond a doubt, eqm ulr brings the best quality. :)
Btw
Was this new one used in Soulhunter's Custom matrix comparison - V3 - Round 2 ? I voted yesterday.
1.5% is a small price to pay for the better quality but it's also a small margin and it should not interfere too much.
However in the SH comparison there was the old ULR.

[Sharktooth]

Originally posted by Ark
I've done some tests with the new ULR, and it seems that the I-frames-blocking issue is almost resolved. I say almost because in comparing ULR with other few matrices (h263, mpeg, 6of9), still some "ringing" remain.

It's not blocking now, but it's like I-frames are sharper than other frames, but also more noisy, and the jump from I to the next frame is quite apparent (not in normal playback btw), something that doesn't happen with other matrices.

Even LR "suffers" from this, but you have to look really hard to see what i mean.

It seems to me that this is somehow costant about your matrices, but other than that i can't say if it's good or bad, as i said, in normal playback it's almost impossible to see it.
Yes, all my matrices (except UHR and EHR) have better intra-frame than inter-frame quantization.
It's an "I-Frame pumping"-like effect to ensure a better quality for reference frames (the codec does a similar thing also). If it hurts you, you can always lower the I-Frame Boost (%) parameter in xvid 2nd pass options.
BTW those "differences" are not visible during playback but IMHO they help to keep a good amount of details in both P and B frames.

[Sharktooth]

Updated EQM V3UHR to raise a bit (but only a bit) the compressibility.
I've also renamed some files to *_rev2.xcm so you will be sure to have the latest version of the matrices.
ULR will be updated soon (at least as soon as i finish testing).

[Velocity 7]

I have to say I'm mighty impressed by these matrices. However, I've been testing these on actual video captured movie-like content rather than anime, and by doing two-pass I've noticed that ULR doesn't have that much of a difference compared to EHR in quality.

So, is that new ULR ready?

[Soulhunter]

Wow, improved EQM3 versions and Jawors 2CD matrix...

Should I post the CQMC round 3, or directly start CQMC v4 ???


Bye

[IgorC]

actually im trying some matrices, so it will be usefull for me some results of test
and what about update EQM_ulr_rev2?

[Sharktooth]

Since im a bit busy with the usual xmas fuzz i have no time to update ULR. But it's going to be updated in the first days of the new year:)

[Palmnut]

I think I'm a little hard-of-thinking after Christmas. Any chance you can recommend, in one place, suitable ranges of bitrates for each matrix?

[Sharktooth]

I can't. There are other factors (motion, resolution, fps, interlaced/progressive material, compressibility etc.) that can influence the answare.
So i can only say "high/med/low" bitrate without specifying the "exact" numbers coz they highly depend on those factors too.

[gurabli]

Which matrix would you recommend for an avarage encoding with 850 - 1200 kBit/s avarage bitrate and resolution around 576*320 - 640*352?





B-VOPs 2, 1.50, 1.00
bframes vhq on
vhq 4
adaptive quant. ON
Qpel ON
GMC ON
closed GOV
trellis ON
quant.restriction 2-31-2-31-2-31
chroma motion ON

2 pass enconding.
Xvid 1.1 Celtic Druid CVS 2004.12.20


Quality is the most important factor! Please suggest a 1 and 2 CD matrix. Thanx in forward.
(I think I should use eqm_v3lr , or?)

[Sharktooth]

Well having no other data i suggest you to try EQM V3LR, EQM V3ULR or Jawor's 1CD matrices.
For the lowest bitrate choice (850) it depends on the res. you choose.
The higher the res. the most "conservative" matrix should be chosen.
For example: 640x352 @ 850kbits, Jawor's 1CD or EQM V3ULR will give the best results. if you bring down the resolution to 576x320 maybe you could think to use a higher bitrate matrix (V3LR maybe).

[gurabli]

Thank you Sharktooth!

With these custom settings should I almoust always get better results than with the built-in matrices (H263 or MPEG)?

Which matrix is the best for the 2CD encoding at 576x320 @ ~1100kbits?

Thanx in forward! You help me a lot! Can you give me a link where can I read more about the custom matrices? I would like to learn what does these numbers mean:)

[Sharktooth]

H.263 is a good quant. type for 1 CD backups, but usually a tuned matrix should do a better job (that means h.263 is a generic, and generic "things" are not the best for a particular usage...).

I dont like MPEG matrix at all. I would never use it, but it's there and it was the "standard" in old MPEG codecs...

I will suggest EQM V3LR for 576x320 @ 1100kbps, but it's a safe and conservative choice coz i dont know the compressibility of your material.
Also if you look around on this board you will notice there's a Jawors's 2 CD matrix, give it a try.
Just use the forum search function to find all you need to know about custom matrices (search for "custom matrices", "custom matrix", "custom quantization", "CQM", "HVS").

[Didée]

gurabli: Most of these threads (http://forum.doom9.org/search.php?s=&action=showresults&searchid=1141260&sortby=lastpost&sortorder=descending) should make a good read about the topic.

[gurabli]

Thanx for your help!

I am going to try out these suggested matrices!

[gurabli]

If I use standard H263 matrices I always get undersized files. The movie is good compressible, it would fit one one CD but I would like to make a 2 CD encoding.

Is the problem with the matrice choice?

Bitrate around 1600. 7xx X 3XX resolution.
Quant 1-31.
Should I use for eg: v3HR for this high bitrate to get the proper desired size? 1400MB