digitalFAQ.com Forums [Archives]

digitalFAQ.com Forums [Archives] (http://www.digitalfaq.com/archives/)
-   Video Encoding and Conversion (http://www.digitalfaq.com/archives/encode/)
-   -   KDVD full-D1 with 2-pass VBR? (http://www.digitalfaq.com/archives/encode/1943-kdvd-full-d1.html)

syk2c11 12-27-2002 11:15 AM

KDVD full-D1 with 2-pass VBR?
 
Hi everyone,
For those of you who are interested in KDVD template, I just visited the following site:
http://pwp.netcabo.pt/0165394101/Compression.html

It talks about the setting in Tmpeg for DVD-compliant mpeg-2 encoding, I just gave it a shot and I found it worked. The Gibbs effect is substantially reduced (I did not say it is completely gone). However, the encoding time takes as twice as we normally do! It is for you judge if it is worth it.

The GOP suggested in the above-mentioned site is "1-4-1-1-18" (for NTSC) and I tried it with Kwag's matrix. I have also tried "1-4-1-1-18" with CQ_VBR mode, it give a better quality (as far as Gibbs effect is concerned) although it is still not the best (2-pass VBR is the best). So I am wondering if we can strike the balance between quality and encoding time?

MoovyGuy 12-27-2002 11:01 PM

Hey there syk2c11;

How does the quality compare to using Kwags KDVD template or my slightly more conservative version :?:

How about the compression ?

kwag 12-27-2002 11:12 PM

Using the standard MPEG matrix, the size will never be as small as with KVCDs Q. Matrix. That's why with the regular TMPEG templates you get about 2 hours full DVD quality on each DVD(+-)R, while if you encode with KDVD Full D-1 you'll get ~4 hours with the same quality on one DVD(+-)R. :wink:
Ask Bud ( Hey Bud, are you around :o ), as I believe he has used KDVDs more than anyone else :D

-kwag

syk2c11 12-28-2002 09:06 AM

I am still playing around the min, average and max values in 2-passVBR. One thing I would not change is Kwag's matrix. Currently, I found min=1000, average=2350, max=3750 is acceptable (as far as Gibbs effect is concerned). In that case, you can put 2 movies into ONE DVD-/+R.

MoovyGuy: are you interested in joining in? I am using "High Crimes" as the testing source, it has a lot of Gibbs effects when encoded by CQ or CQ_VBR mode.

Jellygoose 12-28-2002 09:21 AM

I never understood that... why would you put DVD-compliant files on a DVD-R? In my eyes there are too many restrictions (GOP length etc...) so couldn't you fit a hell of a lot more movies (let's say 3) in awesome quality on one DVD-R with KVCDx3 for example? why make it DVD-Compliant?

kwag 12-28-2002 11:47 AM

Quote:

Originally Posted by Jellygoose
I never understood that... why would you put DVD-compliant files on a DVD-R? In my eyes there are too many restrictions (GOP length etc...) so couldn't you fit a hell of a lot more movies (let's say 3) in awesome quality on one DVD-R with KVCDx3 for example? why make it DVD-Compliant?

The problem is the authoring programs don't let you import an mpeg file that has long GOPs. The DVD standard define a MAX GOP of 15 PAL and 18 NTSC. But then, if we can get a DVD authoring package that allows these kind of mpeg files, maybe most DVD players will pley them. Just like all the tricks we pull on MPEG-1 :wink:

-kwag

MoovyGuy 12-28-2002 11:14 PM

Quote:

Originally Posted by kwag
Quote:

Originally Posted by Jellygoose
I never understood that... why would you put DVD-compliant files on a DVD-R? In my eyes there are too many restrictions (GOP length etc...) so couldn't you fit a hell of a lot more movies (let's say 3) in awesome quality on one DVD-R with KVCDx3 for example? why make it DVD-Compliant?

The problem is the authoring programs don't let you import an mpeg file that has long GOPs. The DVD standard define a MAX GOP of 15 PAL and 18 NTSC. But then, if we can get a DVD authoring package that allows these kind of mpeg files, maybe most DVD players will pley them. Just like all the tricks we pull on MPEG-1 :wink: -kwag

Well one reason that I can see and the reason why I stay within bounds of the DVD spec is to allow for the highest level of compatibility possible when considering consumer DVD players ....

Sure any PC can play the KVCDs as well as many consumer DVD players, including two of my own, but not all....

I want to be sure ... that when I put together a DVD, that it will work for me, my grandmother & my aunt in California and in this case the only thing I need to know is if their player takes DVD-R or DVD+R so that I can burn on the appropriate media .....

Anyway, that's my two cents on that .... :wink:

MoovyGuy 12-28-2002 11:18 PM

Quote:

Originally Posted by syk2c11
MoovyGuy: are you interested in joining in? I am using "High Crimes" as the testing source, it has a lot of Gibbs effects when encoded by CQ or CQ_VBR mode.

Sure, I'll jump in ... as it happens, the VOBs for High Crimes is currently sitting on my main processing machine. I just got finished making a 1 CD KVCD of the movie for my wife ......

What GOP are you currently working with ? And yeah, I'll also be sticking to Kwag's Q.Matrix ... nothing beats it ..

As I have the same VOBs on board you should send me your D2V/AVS scripts so that we are working from the same samples ... I assume of course that you are experimenting with a subset of the movie ?

Anyway, I'll PM you with an address to send the scripts to .....

kwag 12-28-2002 11:22 PM

Quote:

Originally Posted by MoovyGuy
I want to be sure ... that when I put together a DVD, that it will work for me, my grandmother & my aunt in California and in this case the only thing I need to know is if their player takes DVD-R or DVD+R so that I can burn on the appropriate media .....

Anyway, that's my two cents on that .... :wink:

And I absolutely agree with you :D
Right now, every KDVD you make, should play now and in the future in any DVD player. That's because it is 100% compliant. The only thing that is not standard is the Q. Matrix, and that is not defined in the DVD specifications, so it's up to manufacturers and developers to put in whatever Q. Matrix they want. You could call that a "competitor's advantage" facility :wink: So the KVCD matrix won't affect compatibility, but you'll gain better quality in the same space. And btw, the matrix is getting better and better :wink:

-kwag

MoovyGuy 12-28-2002 11:36 PM

Quote:

Originally Posted by kwag
And btw, the matrix is getting better and better :wink:
-kwag

Hey Kwag,

What is the latest version of your Q.Matrix ? Cause the one I'm using is from way back when the first KDVD template came out .....

If there is a newer one in the latest KVCD templates, which template should I clip it out from ...?

Thanks

kwag 12-28-2002 11:42 PM

Quote:

Originally Posted by MoovyGuy
What is the latest version of your Q.Matrix ? Cause the one I'm using is from way back when the first KDVD template came out .....
Thanks

Check this thread: http://www.kvcd.net/forum/viewtopic....=asc&start=120

Warning :!: , still beta. But you might want to try it out. Actually, yes please, try it out :D . And let us know your results :wink:

-kwag

MoovyGuy 12-29-2002 12:06 AM

Quote:

Originally Posted by kwag
Check this thread: http://www.kvcd.net/forum/viewtopic....=asc&start=120

Warning :!: , still beta. But you might want to try it out. Actually, yes please, try it out :D . And let us know your results :wink: -kwag

Alrighty then !!

I'll be starting up a test run in a few and going comatose ... back in the morn ... :wink:

MoovyGuy 12-29-2002 10:35 PM

Well guys,

here's what I'm seeing so far .... using a test clip from High Crimes, that is 1 minute and 50 seconds long, video only, no resizing, encoded at full 720x480 ...

All encodes are using Kwag's matrix, of course 8)


I've run the following tests and as usual all results are subjective ....

Quote:

1 GOP 1-5-2-1-18 CQ 70, 4000 Max, Min 750 with padding = 53943 Bytes
2 GOP 1-4-1-1-18 CQ 70, 4000 Max, Min 750 with padding = 56266

3 GOP 1-5-2-1-18 2xVBR, 4000/2375/1000 with padding = 86922
4 GOP 1-4-1-1-18 2xVBR, 4000/2375/1000 with padding = 86936

5 GOP 1-5-2-1-18 2xVBR, 3000/2500/1000 with padding = 91215
6 GOP 1-4-1-1-18 2xVBR, 3000/2500/1000 with padding = 91410
Visually the difference between 5 & 6 is negligible, and these appear to be the best quality of the bunch, but suffer from large file size ...

Tests 3 & 4 were interesting in that #4 was noticably better, although not by much, than #3. Overall very good, Mosquitos were almost completely unseen.

Tests 1 and 2:

#2 was considerably better than #1, however does suffer slightly from a larger file size. Since I'm only working with a 2 minute clip here, we can see that the file size difference will be considerable.

Having said this, #2 was visually much better than #1, so much so, that with a little CQ/bitrate tweaking would be as good as 5 and 6, but smaller overall. Mosquito effect was quite minimal, but visible on PC monitor running @ 1280 x 1024. Not visible at all on 32" JVC TV (4x3)


What I am seeing so far is that, although 2xVBR provides excellent quality, the file size difference is to great to make it effective enough to use ....

But, I'm not yet finished messing around with this .....

I will now incorporate Kwag's new experimental Q.Matrix tweaks and runn all of these tests again.

I will also be varying the Min/Max/Ave settings on some 2xVBR tests to see if I can bring them into the same realm, file size wise, as the CQ tests and still keep the same apparent visual quality ......

Cheers 8)

SansGrip 12-29-2002 10:40 PM

Some good testing there -- nice to see there's still room to tweak even within the (currently) more restrictive DVD specs :). I'm unable to test any of this yet, but am looking forward to getting a DVD burner probably sometime in the summer... 8)

MoovyGuy 12-29-2002 11:05 PM

Yeah,

I tell ya it's a real pain in the ass having to stay within the confines of the DVD spec, but the compatibility is important for some of what I produce ...

The stuff that never leaves the house is all KVCD .......

I have all of the same tests as I detailed above running in a batch now ... the difference being that the "Notch 2" test matrix has been incorporated....

Gonna go comatose soon & check them in the morning .....

syk2c11 12-30-2002 05:42 AM

Hi MoovyGuy and others,
I always have problems with CQ mode. What is your value for "VBV buffer size"? My problem is the same as mentioned by Kwag somewhere in the forum that video is unstable and jerks occasionally (just a very short interval, 1-2 seconds). I tested with VBV buffer size = 0(automatic)

syk2c11 12-30-2002 07:19 AM

I tested again the GOP 1-4-1-1-18 with CQ_VBR (I have problem with CQ mode), I also notice that it is better than GOP 1-5-2-1-18. Gibbs effect is not that worse. One thing to note that the smaller the range between max and min (say max=4000, min=1000) the larger the file size. So, I now using min=800 (with padding) and max=5000.

MoovyGuy 12-30-2002 07:19 AM

My VBV buffer size is always 224 for DVD encodes, this setting is the same as the DVD standard template included with TMPGenc ...

And now some new results, using the same test sequence, only the CQ ones are of interest here .....

First of all, all of the test look very good, but the new Q.Matrix does indeed increas the file size on CQ encodes ....

Will have to report the number later, don't got no time ....

But the CQ encodes are larger than before, but still smaller than the 2xVBR ones .... Mosiquitos are almost completely gone !!!!!

Interestingly, the 2xVBR encodes are prectically the same size, and the quality is of course great.

Gotta go to work, will report numbers later ...

kwag 12-30-2002 09:23 AM

Quote:

Originally Posted by MoovyGuy
But the CQ encodes are larger than before, but still smaller than the 2xVBR ones .... Mosiquitos are almost completely gone !!!!!

You should use file prediction from now on, and set CQ value accordingly.
Quote:


Interestingly, the 2xVBR encodes are prectically the same size, and the quality is of course great.
And they will look better as long as the file size is larger than the CQ counterpart. Now, if you use file size prediction and target for the same file size your 2-pass VBR encode created, look again and compare the quality :wink:

-kwag

MoovyGuy 12-30-2002 02:32 PM

Quote:

Originally Posted by kwag
[Now, if you use file size prediction and target for the same file size your 2-pass VBR encode created, look again and compare the quality :wink: -kwag

Yeah, but I'm not trying to make the CQ encode as large as the 2xVBR, I actually want to bring the 2xVBR down to the same file size as the CQ encode and then compare quality .......

Always trying to find the best balance between size & quality ..

kwag 12-30-2002 02:44 PM

Quote:

Originally Posted by MoovyGuy

Yeah, but I'm not trying to make the CQ encode as large as the 2xVBR, I actually want to bring the 2xVBR down to the same file size as the CQ encode and then compare quality .......

Ok, that would be the same result :) . As long as they are both the same size, then a comparison can be made.

-kwag

MoovyGuy 12-30-2002 07:14 PM

OK Boys and girls,

As a follow up, here are my new numbers with the Notch 2 beta Q.Matrix.

Old Numbers:
Quote:

1 GOP 1-5-2-1-18 CQ 70, 4000 Max, Min 750 with padding = 53943 Bytes
2 GOP 1-4-1-1-18 CQ 70, 4000 Max, Min 750 with padding = 56266

3 GOP 1-5-2-1-18 2xVBR, 4000/2375/1000 with padding = 86922
4 GOP 1-4-1-1-18 2xVBR, 4000/2375/1000 with padding = 86936

5 GOP 1-5-2-1-18 2xVBR, 3000/2500/1000 with padding = 91215
6 GOP 1-4-1-1-18 2xVBR, 3000/2500/1000 with padding = 91410
New numbers - Same encoding methods, but new Q.Matrix
Quote:

1 GOP 1-5-2-1-18 CQ 70, 4000 Max, Min 750 with padding = 67732 Bytes
2 GOP 1-4-1-1-18 CQ 70, 4000 Max, Min 750 with padding = 71255

3 GOP 1-5-2-1-18 2xVBR, 4000/2375/1000 with padding = 87009
4 GOP 1-4-1-1-18 2xVBR, 4000/2375/1000 with padding = 86992

5 GOP 1-5-2-1-18 2xVBR, 3000/2500/1000 with padding = 91439
6 GOP 1-4-1-1-18 2xVBR, 3000/2500/1000 with padding = 91526
So as you can see, the 2xVBR encodes didn't change much file size wise, but the CQ encodes did.

Quality wise, as I said, the 2xVBRs are great. The CQ encodes are very much improved with the 1-4-1-1-18 GOP being slightly better overall. However, it is larger than the CQ encode with a GOP of 1-5-2-1-18.

Why do I use 1-5-2-1-18 as my standard GOP ? I found that it was capable of a good balance between stability and compression. I found that Kwags GOP for the KDVD template of 1-18-3-1-18 was causing some jerky image problems as well as occasionally missing some scene changes .........

Anyway, as I said, I will now be working on bringing down the 2xVBR encodes to that same file size range as the CQ encodes .........

Later y'all ..

MoovyGuy 01-01-2003 11:04 AM

OK kiddies,

Have some new results here and what I'm finding, much to my own chagrin :grr1: that the 2xVBR encodes that are roughly at the same file size as the CQ encodes are not as good quality wise.

Now you have to under stand, I have always been a huge supporter of xPass VBR encoding as the principle behind this approach makes perfect sense to me.

However, I'm not closed minded either & when I first came across Kwags KVCD & KDVD templates and saw that they used CQ_VBR or CQ I said OK, tried some tests comparing the various methods and settled with Kwag's way of doing things. But in the back of my mind, still thought xPass was better.

So now here we are with another challenge to KDVD and once again when your parameters are best quality for least file size, on KDVD with Kwags first Q.Matrix or the new experimental one, the CQ encoding methods do indeed appear to turn in better results.

So once again, I'm proven wrong .... But I'm happy :ole: 'cause I still get really fantastic encodes regardless of the methodology.

Anyway here's the latest numbers ....

Two sets of tests were run, using a GOP of 1-5-2-1-18 and the other using 1-4-1-1-18.

The goal wast to bring the 2xVBR encodes down to the same filesizes as the last set of CQ encode tests that I performed.

Quote:

1 GOP 1-5-2-1-18 2xVBR, 3250/2000/750 with padding = 66320
2 GOP 1-4-1-1-18 2xVBR, 3250/2000/750 with padding = 66783

3 GOP 1-5-2-1-18 2xVBR, 3250/2000/750 with padding = 71410 NewQ
4 GOP 1-4-1-1-18 2xVBR, 3250/2000/750 with padding = 71734 NewQ
Now don't get me wrong, the encodes look very good, but even #4 from this test is not quite as clear of mosquitos as #2 from that last round of testing.

I do find that even at the cost of a larger file size, the tighter GOP produces better results, I honestly don't know yet whether it's enough of an improvement to justify the size increase .....

More testing will go on .... :insight:








P.S. Dig the new emoticons..

syk2c11 01-01-2003 08:09 PM

Hi MoovyGuy and others,
I haven't tried CQ mode, what I have doing is to use CQ_VBR mode (with GOP 1-4-1-1-18 and Kwag's matrix) and use TMpeg's built-in size predictor (Wizard) to reach the targeted size. I am pretty happy with the result so far (by using CQ_VBR mode). The wizard is pretty accurate which allows you to adjust the average bitrate.

Side issue: I had a terrible error recently, I did not know what have I done wrong (I might have checked "Output bitstream for edit") and encoded the whole movie spending 3.5 hours. when I open the m2v, PowerDVD just show me a black screen, it appeared that the movie was running (the time was moving), but the length showed in PowerDVD was only 1 minute or so. It showed the correct size in file manager.

MoovyGuy 01-01-2003 10:03 PM

Quote:

Originally Posted by syk2c11
what I have doing is to use CQ_VBR mode

Hey syk2c11, howz it going ??

I remember something a while back about using CQ_VBR with KDVD. Something about it being a no no.

Kwag, do you remember any issues with it ? I seem to remember that when when you switched the KVCD templates over to CQ_VBR, that some folks had trouble with the KDVD templates set that way. Something about compatibility that didn't seem to be a problem for CQ.

Can you confirm ?

Anyway, about the closed GOP. I've made many encodes where the GOP is closed and never had any issues other than the fact that the file size of the encode tends to be larger than when it is open ....

Every once in a while though, I'll do an encode & when I open it in PowerDVD it'll freeze on the first fame & not play. I can run the jog bar through the clip, but it won't play. WinDVD or Hollywood+ has no problem though ..... not a big deal, but a little annoying .....

syk2c11 01-01-2003 10:22 PM

Sorry for causing confusion. I meant GOP (1-4-1-1-18) with 2xVBR mode, I mis-typed it! As mentioned earlier, I just play around with the average bitrate and it will lead me to the targeted file size.

So, have you decided which one to use yet (CQ or 2xVBR)?

MoovyGuy 01-01-2003 11:02 PM

Quote:

Originally Posted by syk2c11
So, have you decided which one to use yet (CQ or 2xVBR)?

Well so far I've produced about 12 DVDr discs that have two full length movies on them, 1 with 3 movies (half-D1) and a Lord of the Rings extended edition on a single DVDr.

For every one of these encodes I have used CQ encoding.

Each time I deal with a new source, I repeat the same type of testing that I have here, comparing CQ to 2xVBR with my goal always being best quality for resonable file size. That size being around 1.8 gig. I always end up going for CQ.

The reason for my 1.8 gig target size ? This size allows for mistakes in judgement and for the average size of 350 to 400 meg for the AC3 5.1 track.

So using these parameters I usually can make a 2 in 1 DVD first time out. And of course encoding CQ is considerably faster with only the one pass.

Does this mean that I would never use xPass encoding? Not at all, I'm always comparing, testing, judging ....

Who knows, maybe I or someone else will come up with a Q.Matrix that is better suited to xPass VBR. But Kwag's most definitivly favors CQ or CQ_VBR.

Come to think of it, I do have another test that I want to run ... It's based on an SVCD template that I picked up a while ago. This template, I've found, does favor 2xPass VBR. I used it exclusivly when making pure SVCD's as I find the quality of the encodes produced to be second to none...

I'll letcha know ...


All times are GMT -5. The time now is 03:04 AM  —  vBulletin © Jelsoft Enterprises Ltd

Site design, images and content © 2002-2024 The Digital FAQ, www.digitalFAQ.com
Forum Software by vBulletin · Copyright © 2024 Jelsoft Enterprises Ltd.