@ANDREAS

Hi! Welcome, and thanks for the links.
I have a question ANDREAS. I have done several tests with your matrix, and I get different results at different resolutions. It seems that the DCT algorithms in TMPEG give different result at different resolutions when using a constant matrix value. Like if I use your 99'er matrix, the quality is not constant at different resolutions.
Have you experienced this behaviour? I'm trying to find an optimal matrix, but it seems that I'm going to have to use different matrixes for different resolutions.

kwag

Sorry, but what is IVTC :?:

Also, I have TMPGEnc 2.53, should I update to latest TMPGEnc-2.55.38.142-Free.zip?

Although I always think of IVTC as "In-Vision Time Code" it is actually Inverse Telecine and is used (I think) to change the 30fps of NTSC televsion to 24fps film frame rates. This is only applicable to NTSC sources (ie. America / Japan) and not for PAL systems. Incidently, the 24fps film rate is speeded up slightly to 25fps for PAL transmission, that's why the same film will be about 3 minutes shorter in UK video release than it was in the UK cinema!

Anyway, I encoded my "Romancing The Stone" DVD at CQ 66 and it came out at 620 megs, so I think I can up the CQ a bit and still fit onto an 800mb disc. I will try again tomorrow.

MAX GOP....MAXED?!
 

Hi Kwag,
I did some testing using your GOP (1-18-3-1-48 )... I've been using, and liked the 12 GOP from your old templates, so I just plugged in (1-12-3-1-48 ) to check the differences in quality and filesizes. Surprizingly, they were exactly the same filesize. 8O Hmmmm...Since I used the log function in TMPGEnc, I checked it out...Here's what I found.

By using 18(P) and 3(B) frames per GOP (current template) but limiting the MAX GOP to 48...I find the maximum possible compression would only be 12. That's why the filesizes and logs came out exactly the same...Here's the arguement:

By using a GOP MAX limit of 48 it was getting to 12P+(3B*12)=48 than starting a new GOP(verified in test log).The 48 max was limiting the P frames from reaching the 18 requested... Should'nt the GOP MAX be changed to 72=18P+(3B*18 ) using the current templates 3B frames?...Here's a graph of what I'm thnking the different GOP structures should reflect in creating MAX GOP compression tweaks. Am I wrong with this? (By the way, using the MAX GOP of 72 in further testing did compress the filesize and reflect the 18P frames requested).

Also, would this long of a GOP negate the flashing effect caused before using (0-automatic)? Or would it be better to lower the B frames to (2)? Would appreciate your comments and wisdom on this one.... 8)


|P| |3B| |MAX GOP|
12 + 36 = 48
15 + 45 = 60
18 + 54 = 72

@ANDREAS
My player will only play 480x480 MPEG-2 files. This is close to standard SVCD. Well, same res anyway. You say that your QMatrix is optimized for SVCD. If I were to use your QMatrix would I get ebtter results with it or not?

optimal matrix
 

Well, the matrix. We discuss the matrix problem since ~1,5 jears. One matrix for all it´s impossible. Theoretical have every resolution with every motion area and every datarate this own matrix. Low datarates have other boundery conditions against SVCD standard CD. And 352x576 (480) are different to 704 x 576 (480) and so on. Maybe you found out that different resolution have the same matrix (e.g. KVCD 352x576 with motion area 320x224 and 320x160). But if you see exactly on "my" setting you can see I have turn on the block noise funktion (it is indirect a change of the matrix) with different values and sharpness.

If I search a matrix to a special problem I can only test, test and test. It takes a lot of time to find out in whitch way you can change the values in the matrix. For the examples in the artikel I need more than a week to find whitch matrix and on whitch way I have to change values in the matrix. Fist the intra matrix and later the non-intra matrix if is necessary.

I start with TMPEG default, then I tested MPEG standard and so on. Then you can compare the sequences and you can say the e.g. default matrix is the best picture. From this matrix I take attempts and change in slow steps values of the matrix.

In the first step it is important that the testsequence is a low bitrate sequence. You can see it in my movie examples. The problems on this scene is, that the encoder take normaly only 450kbps or less for the video with the original KVCD template. Normaly it is not enough to ?prevent? the movie about blockness in picture sections there are in the background. Second you can test relativ slow motions to see how sharp they are. Third you can see how sharp is the picture (e.g. hairs in the wind and the corn is moving in the field). Fourthly moskito-noises round corners (it is a very good thing to optimised the matrix). If you have found a good solution mostly you have a good solution for the high datarates. But needs the dataretes a higher value that the max. value is setting (e.g. action sequence) mostly you have to change the non-intra matix again. And then you have test the low sequence and look it is o.k. or not. To the end you found a compromise result about blockness and good picture in the low section.


@a_star62
I don´t know. Your resolution is a NTSC resolution I thing. I´m only working in PAL. Test it at this way that I describe it higher. Take a SMALL sequence and look it on the PC monitor. If you have good results the result is mostly one step better on the TV.

ANDREAS

Re: MAX GOP....MAXED?!
 

Hi slab:
You're right!. 100%. The 18 was left there, because I didn't pay attention, after I closed the GOP at 48. Before, when the last number was set to 0, there was a difference from 1-12-3-1-0 to 1-18-3-1-0, but not anymore.
So the GOP is maxed out even if you set the P's to 9999 or whatever. Which I have :lol: in the new template. To avoid ( or create ) confusion :lol: The new GOP I have set is now 1-9999-3-1-48 because it doesn't matter above 12. It's the same result because TMPEG controls the P's with the fixed size of 48 ( or 50 for PAL ) What I do have under my sleeve, is a surprise 8O , the KVCD Q.Matrix that I've been playing with for the last couple of days. It's a small modification to the "Default" TMPEG matrix. The quality is higher, as seen and analyzed with bit rate viewer, and there's a space savings of around 332Kb per minute :lol:
So I have uploaded the KVCD matrix only in the 352x480 NTSC template.
Please download it and make a measured test against the old template.
This will give us about 40MB savings per 2 hour movie, with higher quality as seen with "my eyes :P " and bit rate viewer. Please give it a run and let me know. I got the best size/quality with these values than with any other matrix. It's available in the download area. You'll see a small "new" indicator above the template

It's 6:30AM and I'm still awake 8O . Going to sleep now.

Enjoy!,
kwag

Hi Kwag !!! :lol:

The template is amazing, the quality is very,very good 8O

Just a question, since i have that "pioneer problem".... :evil:

Can i just use 352x240 instead of 352x480 ?

Is there much quality loss in that ?

Once again thanks !!!

Pacodoni 8)

Hi guys !
I'm happy that we have new MATRIX here . I didn't tryed yet but will very soon . Results will posted as soon as posible !!!!
KWAG !!!
As SLAB mentioned YOUR GOP 1-18-3-1-48 is not the end of way . I beleave that GOP can and HAVE to be longer than 72 even 96 . Somehow I can't encode with this GOP (I think WINme is the reason) system doesn't let me to go to the end of movie but as I wrote already we can achieve more compression in here .
Matrix - TMPGenc default matrix is worse than ANDREAS even it gives smaller file size . To get the same quality I have to go up with bitrate to 700-800/1750 .
I see this degradation in quality on BIG TV screen and beleave me its true . :(
I hope that u'r MATRIX is better than DEFAULT(TMPG). :lol:
I do not have all number coz didn't finished all tests but I'm sure that more compression of about 15% can be achieved with right GOP and MATRIX in wonderful KVCD templates . :D
Andreas is right about different matrixes at different resolutions. I've tryed and quality changes are very noticable . 8O 8O 8O
Make long to short :
KWAG u are THE MAN - U have knowledge and u have ability to put in order all this thing :wink:
I bealeave it's going to happen soon !!!!
bman

Re: MAX GOP....MAXED?!
 

Very cool Kwag, would this matrix also be usefull for the KDVD hald D1 template :?: :?: :?:

Thanks

I don't know MoovyGuy. I've only tested it at 352x480 MPEG-1 and pacodoni tested it at 352x240 with some amazing size result and also he reported an increase in encoding speed.
We need more tests with other resolutions. But I'm sure the results will be different.
I chose to start with the 352x480 because it's a center point between the lower 352x480 and the higher 704x480.
There's probably more room to tweak this matrix, but the current results are looking good :D

kwag

:D That´s right Kwag !

Using the new template with 352X240, the results were very impressive 8O

I left TMPGENC encoding the part 2 of ALI with 74 mins and took 2 hours and 2 mins ( using a P-III 750 and 256Mb ).
THe filesize is about 5-6 meg per min, so it can pass the 120 min per 80min Media :roll:

Try it out everyone, and post the results in the forum.
I will do more tests and keep on touch.

See ya round

Pacodoni 8)

Is the new structure of 352x480 template application to KVCDx2, if so, can I just change from 352x480 to 704x480?

Doh! Just started my encoding runs for the night before I fall asleep, and just read this. I will try more encodings tomorrow with the new KVCD matrix... sound very intruiging.

Matrix
 

O.K. you write somthing about a new MATRIX. I hope we speak from the same thing. The matrix is the point in TMPEG under quantisize matrix, right? In the *.mcf file I found the "Andreas einfache 99er Matrix" and if I understand it right you use thisdmatrix to encode?

ANDREAS

Hi ANDREAS:

The matrix being talked here is a new test matrix I am testing. I modified the "default" TMPEG matrix with some new values, and that's the one being tested. I only modified the KVCD 352x480 template with this new matrix, for evaluation purposes. You can find it in the download page and you'll see a small "new" banner over it. It's the only template that currently has this matrix. The only real test so far I've done is file size comparison against all other matrixes. And I only tested it at 352x480. But I've received good results from other people using the matrix. I did compare the Q in bit rate viewer against the "default" template, and I do get a lower Q, which means higher quality, compared to the standard TMPEG template. I did notice some very small artifacts around some moving object. Just a little more than with the "default" template. So I have to analyze what is the frequency domain of these artifacts, and hope it's constant with every encode, to zero in on the 8x8 DCT value and clean that up. That will be the next step. A Steeeeep step :)

Regards,
kwag

Test Results
 

kwag,
just tried the new template and decided to make some comparisons with all the templates. the source was 1 min 31 sec section of alien4 ressurection 16x9 ntsc.

these are from the current templates:

template cq filesize encode time
352x240 80 13922 2:18 m
352x480 70 18508 2:43 m
704x480 50 20778 4:02 m

these are from the beta template:

template cq filesize encode time
352x480 70 17757 3:25 m
352x240 80 17201 2:20 m
352x240 70 13241 2:25 m
352x240 60 10965 2:17 m
352x240 50 9097 2:10 m

i viewed all these with windvd and all the tests except the 50cq on the
new 352x240 template look real good. i think 60 would be as low as i
would go. on the 352x240 template i just changed to 240 using the beta
352x480 template.

KVCD 352x480 template update
 

Hello everybody :lol:

I've made some minor changes to the template today.
Consider this still experimental.

The following changes were made:

B and P spoilage changed from 0/20 to 30/30
CQ level changed to from 70 to 72

This changes increase the file size about 60KB per minute, compared to the last update ( the GOP 1-9999-3-1-48 and KVCD Q. Matrix ) but still create a file size around 400KB per minute less than with the "default" TMPGEnc Q. Matrix.

I've tested these values in the 352x240, and the results are up to 700KB savings per minute, with no visual quality degradation. I just did a low action scene of one minute (video only) with the 352x240 template and the file size was 5,098KB. I also did a full action scene (also video only)with the 352x240, and the size was 7,953KB. I am, :lol: VERY :lol: pleased with the results.
Testers welcome 8) I need some feedback!.
I've updated the KVCD 352x240 NTSC template and the KVCD 352x480 template in the download page.

Let me know your results

BTW: The 9999 is a joke. There's no visual/quality changes after 12, because the 48 regulates the GOP size, etc.. So I set the ceiling value in TMPEG ( maybe as a KVCD signature :wink: )

kwag

Another update: 8O

Hopefully, now there's not that much more to tweak ;)

I've dropped the KVCD Q. Matrix and reverted to the "default" matrix :cry: The KVCD matrix is excelent as far as file size, but it has more "mosquito" effect around edges. This is monitoring a movie on a HDTV. It's not noticeable on a regular TV, but on high definition TV's, it is.
So until I solve that, we're back to the default. ( which is not bad at all ).
Both templates 352x240 and 352x480 NTSC have been tested.
Both templates have been changed from "Normal" motion search precision to "fast". The B spoilage has been tightened from 20 to 10.

Because the tightening of the spoilage, the CQ could be lowered a little. On the 352x240, the CQ is now 78, and on the 352x480, it's 68.
The quality is better than the original templates, and the "mosquito" effect is almost gone. The file size is still lower that with the original template and the "default" Q. Matrix.

I'm going to leave the templates this way for a while, until I get more feedback, and more test runs are made.
I wish I could get the quality that the current templates generate with the KVCD Q. Matrix. That would be the perfect combination. Maybe some day I ( or someone :wink: ) will find the optimal coefficient values for the matrix.
I just updated the download page. Both templates are available for download now.

kwag

So what effect does motion search precision mean really? I've always read it should be at 'high', and almost never at 'fast' unless quality wasn't a concern. Of course, not sure where I read this... Very cool if 'fast' is acceptable, because.. well, it's fast ;)

Hi Daagar:

The "Highest" quality setting just takes too long to encode. There's almost no difference from it to the "Fast motion estimate". I tested the "fast motion estimate" and the result is that I get less "mosquito effect" around edges in the picture.

Here's a 2,492KB 30 second sample ( video only to save space ):
http://ns1.shidima.com/kwag/paysamp.mpg

The sample is 352x240 with the template that is currently posted. The complete movie video size is 416,197KB. The audio file, which I created with headac3he at 224Khz, is 166,033KB. So the complete multiplexed movie is 589,502KB. The movie's playing time is 1 hour 41 minutes and 12 seconds.

kwag

Good enough reason for me! Anything to cut back in encoding time is fine. My machine only has 256megs of ram, and I found out that with the 704x480 template, I was pegging the amount of physical ram I had, and spent most of the time swapping. VERY painful encoding times. Dropped back to 352x480 and things are cooking along much better. Dropping to 'fast' motion search will help even more!

Hi KWAG !
Nice work , NICE
Just a FEW questions :
352x240 template bitrate 1150/300 ???? 1150 is the right value in here
( vs 352x240 - 1750/300 )
Kwag !
As I see that u'r KVCD matrix is better than Default - on 34" screen I really don't notice "mosquitoes" (u r talking about HDTV but how meny of us are using it ????)
I defenetly preffer KVCD matrix vs default !!!
With Default matrix there is some color problem too : Colors become somehow FADED . Black becomes dominant and colors lose strength (can't change even with tv tuning).
U'r previous template was much better ( I think )
What about ANDREAS matrix : It was just fine too .
So many changes and tweacks ....
Can u check 1-9999-1-2-48 GOP for me ?
It saves few Kb too !!!
All in All good job , MAN !
So many efforts and on such short period !!!
Best regards
bman

Hi bman:

Well, surprise probably tomorrow 8)
I've got the KVCD matrix working with NO "mosquito" effect in my HDTV monitor 8O
I'm still running some tests, as I am syncing the 352x240, 352x480 and 704x480. They are all using the KVCD matrix. BUT using CQ_VBR insead of plain CQ. I did this, because some people's DVD players can only play CQ_VBR, but not CQ mode. But nobody has complained that they can't play CQ_VBR. The resulting bit rate is very different viewed with bit rate viewer. I hope that some time tomorrow I'll have something ready to upload.

In the mean time, here's a peek at two samples so that you can feel what to expect :wink: Each is 15 seconds.

http://ns1.shidima.com/kwag/samp-352x240.mpg
http://ns1.shidima.com/kwag/samp-352x480.mpg

No audio on both samples. Just a dummy 64k blind filler mono chanel.
Same sample clip at two different resolutions. Both WITH the new KVCD Q. Matrix. Feedback always welcome :lol:

kwag

bman,
About your questions. 1750/300 is for the 352x480. 1150/300 has always been the value for the 352x240.
About the 9999 :lol: I explained it before. If you increase the P's above 12, there's no file size/quality difference, because the GOP size is being controlled by the max of 48. So I just set it to the max that the field permits. Which is 9999. Just for the hell of it :o

kwag

@ KWAG !
Sorry about my bad english !!! :cry: :oops:
What I meant was - set in u'r temlplates 1-9999-1-2-48 instead of 1-9999-3-1-48 .
Just change numbers and give me your opinion about this change . :?: :!: I found that this can make smaller file size !!!
bman

Hi bman.

On every measured test I've made so far, changing the number of B pictures from 3 to 2, increases the file size. Not much, but it does.
Only on dark scenes and long low action scenes, the B=2 creates a smaller file. But on average, on a complete movie, the B=3 creates the smallest file.

kwag

Kwag !
We are missing something here ?!!! :oops: :cry:
In 1-9999-1-2-48 GOP =>>> I=1 P=9999 B=1 Output interval = 2 Mux=48
Am I right ??? :(
bman

I=1
P=9999 ( for the time being :lol: )
B=3
Output interval in sequence header = 1
MAX number of frames in GOP = 48

So the order is: 1-9999-3-1-48

kwag

Surprise ready?
 

Kwag,
Are you ready to give us surprise (new template with KVCD matrix)? I am really looking forward to tasting it.

Hi teacher!

I did some tests again and found that... you're right! the Gop 1-18-3-1 give the smaller file size if you use the standard matrix but 1-18-1-1 give smaller file size for an entire movie if you use the andreas matrix, so there is no sense in using andreas matrix wich increase the file size and after that change the gop to decrease the size unless someone can say that the quality with this procedure is better, right now i'm testing your new templates so i leave the original gop as it is, by the way i have a question, did you say you changed from CQ to CQ_VBR? do i need to make this manually cause the templates doesn't changed it... it changed CQ68 motion search to estimated and changed the Q.matrix to yours... or you will post that template today?

Well, so sad, i thought i found something useful but i didn't, better luck next time... i hope.

Saludos.

DaDe.

@syk2c11
Almost! There is one thing which I have been trying for the last two days, and that is to reduce the size of the files, with the new templates. Right now the best I've been able to do is that the 352x240 template can still fit an average 120 minute movie. But the 352x480 will fit "maybe" 120 minutes, but probably more like 100-110 minutes max per CD :cry: . And the 704x480 about 95-100 minutes per CD :cry: .
BUT the quality of the three templates just blows away every other template!. The 704x480 just looks incredible, even better than the 2 CD 704x480!

So I think that even that the file sizes are larger than with the older templates, I'm going to go ahead and post them, in a separate beta section. If someone wants to fit more time with these templates, it will be a matter of just lowering the CQ_VBR values ( which are completely different than the values of CQ mode! ). After seing the quality produced by these templates, I'm not going back to the old ones 8) . Even if I have to use 2 CD's. Of course, I'm looking at the VCD's in a HDTV, and there you can see the humongeous difference 8O

@DaDe

The new templates are based on CQ-VBR instead of plain CQ. I read that many DVD players don't play CQ, but play CQ-VBR. Also the quality I'm getting with the new KVCD Q. matrix works better with the CQ_VBR.

@ALL

The templates are available at the download page now :lol:
Here are three samples, one from each template, for your evaluation.
They're coded with silent 128Kbps audio track.

http://ns1.shidima.com/kwag/film-352x240.mpg
http://ns1.shidima.com/kwag/film-352x480.mpg
http://ns1.shidima.com/kwag/film-704x480.mpg

Hack'em all you want! :lol: And let me know your results.
Enjoy,
kwag

cool
 

Ill be sure to try this new template out on my Apex 1500.. hopefully the cq-vbr will work with my DVD player...


One other quick question Kwag.. You were mentioning before with the Motion Search Precision in Tmpgenc on Motion Estimate Search (fast) that it wouldnt change the quality at all. Im wondering if this is still true with yer new 704x480 beta template.. An increase in speed would be nice :)

tks

Re: cool
 

The BETAS have the Motion search presicion set to "Normal".
After some tests, with the current paramers and the Q.Matrix, I got lower file size with the "Normal" mode instead of "Fast".

kwag

Okay, a few questions about getting the most out of the new templates. Let's assume we have a 150min movie. A bit large for the new templates (actually, too large for the old as well most times), so we know we need two CDs. No problem. Since you've gone to CQ_VBR, does this give us any better ability to 'max out' the space on each CD? IE., I know the movie is 150min, so let's just put 75min on each CD, and fill each CD to capacity. Or does it still follow the CQ model of 'you don't know filesize until you try'?

This also begs the question 'why not 2pass VBR'... (time, obviously, would be one reason).

Hi Daagar:

Based on the estimated file sizes generated by the BETA templates, I would put the 150 minute movie in 2 CD's, using the 352x480 template. Now this is what I recommend, if what you want is to max the size in the CD.
The quality generated by the new templates, it's pretty much maxed out. You won't get more quality by increasing the CQ value. Maybe a little, but it's barely noticeable.
What I am doing, and I strongly recommend everyone to do, is to encode ONLY the video stream. Then, depending on the video stream size, you encode the audio with HeadAC3he. When you run headAC3he and open your AC3 or WAV file, the program tells you what the size of the MP2 file will be. So just add your video size to your audio size and see if it will fit in one CD. If it doesn't fit in one CD, with your audio say at 128Kbps, then encode at full 224Khz audio quality and mux your video/audio in two pieces with BBmpeg. It never fails!. The audio and video are always in perfect sync using this method. So instead of waisting time re-encoding the video, DON"T. It's a waste of time.
If you still want to fill your disk with CQ mode, I suggest you try DVD2SVCD. It will calculate an average CQ to fill a CD. But keep this in mind, if you are thinking about x-pass VBR. The longer your movie, the worse the quality. The longer the movie, the lower will be the average bit rate, for a calculated size. With CQ modes, the longer the movie, the larger the file size. But your quality will be maintained constant in every part of the movie.

kwag

Kwag,
I hope you can bear with me (us) since a brand new template is out, I (we) may have a series of questions.

(1)---If I am not anxious to "fill up" each CD-R, what would be the suggested set up to have a movie (say 120 minutes) safely put into 2 CD-Rs by using the NEW 704x480 template. All I care about is quality, I don't mind to have 2 CD-Rs for one movie but not 3 CD-Rs. I don't want to have a separate procedure for audio, I just want Tmpeg to take care of all Video and Audio for me.

(2)---Will there be a noticeable difference between 128 and 224 kbps in audio setting? Thanks a lot.

Hi Kwag,

With switching modes in your new templates you've opened the door to more questions for those of us trying to keep up with you 8O .

I've started a new thread here http://www.kvcd.net/forum/viewtopic.php?p=3352#3352

Thanks

1) Just like me 8) I don't mind 2 CD's. But 3 is a no no :lol:
For 120 minutes at 704x480 in two CD with the new template, maybe!, but you'll probably have to lower the CQ value a little. It all depends on the type of movie. I'm encoding "The Matrix" later tonight at 704x480, to find out the size, with the template without modifications. I'll let you know tomorrow. That's a 136 minute movie, so we'll know what to expect from the new template as far as file size.

2) There is a HUGE difference from 128Kbps to 224Kbps. As a matter of fact, if you use headaAC3he which uses mp2enc to create the audio, it will sound better at 128Kbps than with TMPEG's built in audio encoder at 192Kbps 8O
I haven't tried audio lower than 128Kbps in the regular templates. But I can tell you this: I tried audio at 64Kbps for the Pocket PC templates, done with headac3he in joint stereo, and it sounds good. DAMN good! Don't try that with TMPEG's audio encoder. It will probably sound like metal cans rolling down a street :lol:

kwag

Ahhh, my mistake Kwag. I was thinking CQ_VBR was one of the modes where you specified your own 'average' VBR rate, not just min and max. If you could specify an average, you could calculate a 'max out the CD size' as you can with dvd2svcd. However, it appears with CQ_VBR, you are still simply defining a min/max and CQ level, so what you suggest is probably the best bet. Thanks for pointing out headache... I've used besweet in the past but will check this alternative out.