@Kwag,

It's like the Mel Gibson Frankenstein movie where Egor ask GOD to make
his arm the same as the other. Lighting flashed and sure enough both of
Egor's arms were deformed and crippled. These pics now both look like equal
hammered Dog S***. This rabbit hole that we have fallen into with Alice
may be too deep with no bottom or end. I'm getting weird results from
file predictor, Q-Matrix, etc. I'm going back to CQ_VBR with GOP
1-12-2-1-24 for now. These changes are too hard to keep up with.

-black prince

:mrgreen: I agree. Things are chaotic, and that is good :D . That means the next generation of KVCD parameters should be of a "Higher Standard" with much better quality :D

You might want to use the matrix labeled "KVCD Notch BETA-1" which is identical to the original KVCD matrix, except for the "notch" adjustment on the low frequency end. That should help get a smoother picture by eliminating some DCT blocks with CQ or CQ_VBR :wink:

-kwag

@kwag and black prince:

Having viewed all 16 clips three times on my TV set, I've come to the conclusion that at least on an analog TV, there's very little difference between the old KVCD matrix and the new one. I believe that beta-1 version produces a slightly superior result when viewed on the monitor (and, perhaps, on a very large HDTV) than the old, but that beta-2 is unfortunately substantially worse in fast-action scenes.

I think the case is strong for using CQ mode at 528x480 and above, WITHOUT any added noise. It seems that the resolution is high enough that even though DCT blocks are very noticible on the monitor, they don't appear on the TV to any significant degree.

I also think that 352x240 and 352x480 benefit greatly from added noise (I'm still not decided whether I prefer noise or dither), in which case CQ_VBR mode should be used since it is much less aggressive with low-frequency components.

I think what I've settled on (at least for my own personal encodes, for now ;)) is:

* At 352x240 or 352x480, I'll use CQ_VBR with Blockbuster if necessary;
* At 528x480 or 704x480, I'll use CQ without Blockbuster; and
* I'll use either the old KVCD Q matrix or beta-1, depending on the results of my more specific tests tomorrow.

I think that sums up what I learned today. I must also mention that every time I see 528x480 on the TV I'm more and more impressed with it. Even with bilinear resizing it's very sharp, and yet doesn't exhibit many more artifacts than 352x480. I don't think I'll be using 704x480 except for multi-disc encodes :).

I think -- again, at least wrt my own encoding -- things should settle down a bit now. I'm looking forward to actually getting complete movies onto disc :mrgreen:.

Is there anything better than this hobby? 8)

Boy are we in sync :mrgreen:. 100% with you on every detail :D You bet we've learned a lot 8O :) And I think I'll stick to 528x480 even for 2 disks. There's not that much difference, even when viewed on HDTV. But I personally think that just about ANY movie can go in 2 CDs with the x3, and there will be hardly any difference from the original source.As for other hobbies :?: , no way :!: :mrgreen: 8)

-kwag

So I guess the last thing we need to thrash out is: old Q matrix, or beta-1? I'm going to run a few sample strips tomorrow with an eye on file size difference (I already know it's a little better in the low-freq areas). I'll probably choose beta-1 unless there's a significant difference in file size.

Glad we're in agreement on this one. It's nice to have independent confirmation of one's tests :D.

P.S.: Once we're "stable" again I think we should collaborate on a HOWTO and more complete FAQ, as well as update the templates download page. Some of those figures are way out now ;).

I'm going to try a couple of samples with the original KVCD matrix and the BETA-1. I think that BETA-1 is superior, and the actual "notch" idea did work. Probably not optimal, but it does show some improvement. And we need every quality improvement we can get, even if it's very small :D
I'll do some test at 352x240 and 704x480 (extremes) with both matrixes, using file prediction to get the same file size. The file size difference from the original to the BETA-1 is almost negligible.

-kwag

quote parts:

1
At 352x240 or 352x480, I'll use CQ_VBR with Blockbuster if necessary...
yes,in my tests with 352x240 shows it. :wink:

2
I also think that 352x240 and 352x480 benefit greatly from added noise (I'm still not decided whether I prefer noise or dither), in which case CQ_VBR mode should be used since it is much less aggressive with low-frequency components.

with dither : more clean,less size
with noise : more saturation(color),more size
...of course,same chapter in test. :? :?:
don't know what1s better...seems...dither? :wink:

3
You bet we've learned a lot ...
oh yes,i'm croos-eyed. 8O

:)

You bet :D. I think that we should be set in a couple of days. The only thing left is just some basic tests with the matrixes. Then "Freeze" and document, and have fun for a while :lol:

-kwag

There, that's three people. How much more proof do we need? ;)

Interesting. If you're noticing a decrease in saturation, try lowering the variance a little to see if that helps.

I need to add more emoticons to the forum, so that we can really express ourselves :mrgreen:

-kwag

Since I've settled on 528x480 as my preferred resolution I'm going to do a little bit of testing with the different resizers (mostly Lanczos with subsequent light smoothing), but apart from that I think we're good to go.

Then I will work on a new version of KVCDP (I'm thinking this time of rewriting in regular Win32 so that .NET won't be necessary) optimized for our new discoveries.

But my computer will be idle at night, so I think I might take this opportunity to back up some movies :mrgreen:.

please, i need one "BIG stupid" :lol:

Like this: HELLO Jorel :!: :BangHead:

:lol: I'm working on it now :wink:

Edit: :alien: :angel:

-kwag

:bouncy: :bigooh: :angryfire:

Hope you like them, many more by the minute :D . Check back in about half hour :beard:

:lol:

Night all ;).

yes,:bigooh: IT'S ME.....my photo! :lol:

Good night SanGrip, have good :zzzsleep: and dream with :angel:

-kwag

@All,

SansGrip wrote:
I agree. I'm comparing KVCDx3 to it's CQ version to see which one has
the most advantages (e.g. file size, picture quality, etc.)

Can't wait for this :) I'm going to give Q-Matrix Beta-1 a try with
528x480 resolution, because I'm curious.

Finally, file prediction for me is not working for CQ tests. :( Followed
the sticky for manual process to the letter. After determining the target
file for video and creating the test file, the formula that gives me a
new CQ is constantly too high and computes to a lower CQ than the one I
started with only to bounce the new CQ too high again. I can't
understand how no one is having problems. If anyone is any problem
with file prediction please speak-up. If you don't speak-up Kwag will
assume nothing is wrong and move on.

-black prince

emoticon orgy
:smileysex::smileysex: :smileysex::smileysex: :smileysex: :smileysex:
:smileysex::smileysex: :smileysex::smileysex: :smileysex: :smileysex:
:smileysex::smileysex: :smileysex::smileysex: :smileysex: :smileysex:
:smileysex::smileysex: :smileysex::smileysex: :smileysex: :smileysex:
:zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep::zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep:
:zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep::zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep:
:zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep::zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep: :zzzsleep:

LMAO :flip:

:bouncy: :hopline: :bigooh:
:mrgreen: :mrgreen: :mrgreen: :mrgreen:

this is the real life!

after much reading,write,test,adjust,teach,learn,discuss,ask, think.......

the people here (like me),

is gonna crazy! :lol:

I still haven't found your requested icon jorel :teeth:

-kwag

how can i put a little bitmap here?(3,02k) :?

Do you have an icon? You can send it to my e-mail. Send me a PM.

how i put the image?(bitmap)
tick in what?

Check your PM.

-kwag

Here are my findings for pal,

I`ve found the following to work best for me,
For darkish movies like Blade2 Men in black, I found using CQ without blockbuster 528x576 to be the best,
and all other movies using blockbuster Cq-Vbr looks best to me, I`ll try the new dither blockbuster 7 tommorow, after new years coz i`m working on a new years bash tonight. I did try the new matrix not extensively though so i`m still experimenting with that.

Thanks for all the hard work guys, and have a great new year :D :D :D ,

God bless.

thanks,.........i see you all in 2003.
Happy New Year,progress,fortune,happiness!
GOD bless you all,good morning!

I agree with SansGrip when he said after things slow down you guys could make a how to guide. I don't know about every one else but I'm going crazy trying to keep up with Kwag and SansGrip and all the new software. Don't get me wrong because all the hard work is paying off. Everyone on this forum loves your templates,and look how far we all have come. But with all the new software and you guys fitting longer movies onto one cd some of us aren't sure how you did it or were to start. But a guild could show everyone how you get very great results,and a new faq would be nice also. And the new icons look
great! :ole: :litesbr: :gun: :punch: :tongue2:
Ps: Thanks everyone for your hard work testing and testing again,and thanks kwag for you great templates and everyone helping kwag with his great templates!

I think CQ mode has less Gibbs and responds better to fast motion. I'll be interested to hear what you think.

CQ prediction does work for me, but it's a very slow process. Generally I do it manually, and go up or down maybe 2 or 3 at a time until I end up on the "other side" of my target, then I'll go up (or down) by 1.

It is possible to get very accurate sample sizes, but it generally takes many more iterations than CQ_VBR. That's the one downside of CQ...

8) I feel kinda spacy-punk-cool that I started a 4000-views threat at my favorite website :buzz: :wavey:

yeah! and as for the results: I'm totally with SansGrip and kwag.
But since I only encode in higher resolutions, I'll stick with CQ! :lol:

CU next year!!

kwag -

Now I look at beta-1 more closely I see you're using values lower than 8. This concerns me somewhat, since I'm under the impression (correct or otherwise ;)) that 8 should be the minimum value in the matrix.

This is because the result of a discrete cosine transform is a number between 0 and 2047, and if you were to divide by less than 8 you could get a number greater than 255. This would result in an overflow and subsequent artifacts.

I could be wrong, though.

@SansGrip,

SansGrip wrote:
That's similiar to the way I'm doing file prediction for CQ. I pay
attention to the test file size result and less to the predicted new CQ.
When I'm < 8MB near the target file size, I increase CQ by 1 until
I'm within 2% to 3%. CQ creates a final file size a little larger
than the predicted file. I also use 128kb audio minus CD size, so
if the final video file size is too big for target, I can reduce the audio
size to fit. It's a little tricky, but it's working :D

-black prince

I thought so too :eeks:, but only if I use 0 or 1, then video starts to jump. Maybe what's happening is we're getting a real "clamping" effect ( again a notch ) where I'm using the 6 and 7 values. Or maybe changing the 6 and 7 for 8 will have the same effect. Then we'll know if there's no sense in going below 8. If this is the case, then I will sit in a corner for a while because :imstupid: :cyclops:

-kwag

@SansGrip,

For Cropping I used Tmpgenc's Clip Frame to get the Crop settings.
e.g. Crop(left,top,-right,-bottom). I mask in multiples of 4 to avoid
macro block errors. Next I use FitCD to get correct aspect ratio and
save this in my script:

mpeg2source("D:\Temp\movie.d2v")
Crop(4,64,0,-68 )
LegalClip()
LanczosResize(496,352)
FluxSmooth()
AddBorders(16,64,16,64)
LegalClip()
Sampler()

This seems to work and is simpler than using FitCD to get crop settings.
Can you tell me if I missed something :)

-black prince

That's pretty much what I do, except I usually use VDub to figure out how big the borders are :).

If you use zero you should get divide-by-zero errors -- I'm surprised it works at all :). If you use 1 then you're producing values from 0-2047 instead of 0-255. This is probably why it's jumping.

I'm not totally sure this is the case, but I know that TMPGEnc won't let you enter a value lower than 8 in the very top-left box. I'm thinking that the reason you can use a lower value elsewhere is simply a bug.

I think I'll go ask on Doom9 -- this one's outside my area of knowledge ;).

It's a HUGE difference. Look at this, and note specially around the eyes:

http://www.digitalfaq.com/archives/i.../2002/12/6.png

The top image is using the BETA-1 matrix. The top left of the matrix is this:


Where the bottom image is using this:

I think the notch is dead on target :wink:

Here are both samples:

http://www.kvcd.net/test-cq-beta-1.m1v
http://www.kvcd.net/test-cq-8s.m1v

-kwag

You're right, the lower values are better. However, that still doesn't rule out the case where it overflows and causes weird artifacts. Before going with it I'd like to get a definitive answer on whether one should use values less than 8 in the matrix :).

My tests with beta-1 showed quite a serious impact on CQ level. With the old matrix I got 69.9, and with beta-1 I got 62.8. This will probably equate to more macroblocks. I also saw more Gibbs :(. Are my results really that out of whack with yours?