Yep, though if the quality increase is significant I'd say it would be worth it -- at least for those with faster machines.

Personally I'd let mine (Athlon XP 2100+) encode all night if it meant significantly better quality, at least for important jobs.

Well, so far I have my sample clips of Resident Evil (2424 frames) encoded at 704x480 CQ_VBR with dither, 704x480 CQ without dither, 528x480 CQ_VBR with dither, 528x480 CQ without dither, and 352x480 CQ_VBR with dither. All are at 11.4mb. I'm very close to having 352x480 CQ without dither finished (it's at 11.3mb right now), and then tomorrow I'll do 352x240 and watch them all on my TV.

After that I have to redo the whole thing with kwag's new Q matrix .

Now will that be too long, because now I'm sitting on the edge of the chair waiting for your results ( Ouch )

You may as well get comfortable since it's taken me hours to get this far . I find getting precisely sized samples with CQ takes a lot longer than CQ_VBR. There's no way of knowing if you need to increase by 0.5 or by 2. It's very frustrating .

...mainly because, 5 sample encodes later, I discover, for example, that I've been doing CQ 352x480 WITH dither instead of without.

Like now.

GRRRRR .

Oh you're 100% right. It's a pain . Every time I calculate "((wanted/current) * current CQ)" the curve is completely off on the next encode . Try it with the BETA-2 matrix, it's a little more linear

-kwag

That's next on my list .

By the way, how did the method="dither" sample strip compare to method="noise"?

On 704x480 and 528x480, I can see a clear advantage. At 352x240, the macroblocks are just so big, that anything I try they still show . I'm currently encoding with CQ=53.39 and the new matrix the movie K-19 at 528x480. Here's a short cut from the prediction sample: http://www.kvcd.net/test-kvcd-new-matrix.m1v

Made is with this script:

LoadPlugin("C:\encoding\MPEG2DEC.dll")
LoadPlugin("C:\encoding\fluxsmooth.dll")
LoadPlugin("C:\encoding\sampler.dll")
LoadPlugin("C:\encoding\blockbuster.dll")
LoadPlugin("C:\encoding\legalclip.dll")
mpeg2source("K:\K19\VIDEO_TS\k19.d2v")
LegalClip()
LancZosResize(496,336)
FluxSmooth()
Blockbuster(method="dither", detail_min=1, detail_max=10, variance=1)
AddBorders(16,72,16,72)
LegalClip()

-kwag

Kwag please:
how can we "found" this values?

AddBorders(16,72,16,72)
why is 16,72,.....?

Really? Cool .

Yes, there's probably not much we can do about that. If you set the intra-frame Q matrix to all 8s (the lowest you can go) you'll still see tons of DCT blocks. MPEG-1 I-frames are blocky, period .

FitCD

thanks (fitcd)!

i see the "test-kvcd-new-matrix.m1v
and seems strange (a little) aspect radio.(faces like eggs)

Rename the file to .mpg and play it in WinDVD or PowerDVD
DON'T USE WMP
It's a 528x480 file.

-kwag

@Kwag,

The first noticable benifit of the new KVCD Notch (Beta-2) Q-Matrix is
test file size has decreased from 22,865,194 to 21,896,514. I'm using
704x480, lanczosresize, Fluxsmooth, CQ 50. Picture quality is difficult
to judge, but it looks as good as the old Q-Matrix. Going to try Notch
Q-Matrix on 352x240 CQ_VBR. Let you know results later today.

-black prince

You should notice less artifacts (Gibbs) on high frequency components and less macroblocks on fast scenes. Also slightly lower dancing DCT blocks on low lit scenes. It's just a little, but it helps

-kwag

Hey Kwag,

Kwag wrote:
I did notice improvements using "avscompare" under magnification mode.
They were small but did appear improved. I have a question about
528x480 Plus (KVCDx3). What needs to be done to the Template so that
I can make it work as CQ instead of CQ_VBR I haven't tried the
new Q-Matrix with this resolution yet. Are you still using Blockbuster
dither with CQ

Thanks

-black prince

Just go to Settings/Video/RateControl and change to CQ.
I'm still experimenting with dither

-kwag

Ok, I've got 352x240, 352x480, 528x480 and 704x480 all done using both CQ_VBR with dither and CQ without. I'm now encoding them all again, this time using the new matrices...

Can't wait for the results , either good or bad. The truth, and nothing but the truth

-kwag

Hey SansGrip Tick, Tick, Tick, Tick

Are the results or are they :P

-kwag