I want make a KSVCD with 480x576 (PAL) and MPEG-2.

I doubt these settings:

GOP size is 25?
I have to check KVCD notch matrix?
Maximum bitrate is 2500?
I must check detection change scene?

Thanks.

-Maurus

Hi Maurus,

Your settings are correct, but make sure you uncheck "Trellis".
There's a quantization bug in there, which is visible mostly when using KVCD notch matrix.
I've seen it happening even without using KVCD's matrix (very sporadically).
So just disable trellis, and you'll be fine.

-kwag

I never used Trellis, .... as this came up in the early time of the mencoder thread. After comparings, I immediately dropped it.

Yep. I agree, and I couldn't see any substantial difference selecting or deselecting trellis.
Notch, however, does makes quite a visual difference
Seems libavcodec's quantization algorithm is our friend

-kwag

Thanks to both.

I also read GOP size = 24 (not 25), because 24 is mod 3.

How about detection change scene?

-Maurus

What does MOD have to do here
It's 24 for NTSC and 25 for PAL.
If you choose 24 for a PAL movie, and you select closed GOP, you'll get a MAX GOP size of 22.
If you select 25, you'll get a MAX GOP size of 25, and those 3 extra frames DO make a difference Select it
You'll get better quality..

-kwag

I see in PAL KVCD templates for TMPGEnc GOP = 24.

But I will set GOP = 25

Thanks.

-Maurus

Those are old settings
I never did update the templates, because we just don't use templates anymore. We use "parameters", useable for any capable encoder

-kwag

It seems that you think than AviSynth+TMPGEnc method is old...

It seems also than best quality, very fast and best file prediction is Mencoder.

I will migrate to KSVCD in MPEG-2 with MEncoderMe 0.23.

What is the best quality method in MEncoderMe 0.23?

1) Script Avisynth + MakeAVI option
2) 1 Big VOB with DVD option.

Thanks.

@kwag, inc., ...
... oh yeah, depends on the eye ... but why does it mean trellis is wrong ?
however, all my tests having completed so far says; with trellis i got higher compressibility and(!) higher psnr (ok, psnr's not the best metric but good for comparison) i found the same for mpeg4. so i wouldn't go wout.
why do u think, trellis is so wrong ? imho, trellis is our friend. give it one more chance before draining it
the bests
y

I don't mean trellis is wrong. I mean the algorithm it uses, just doesn't work correctly with the Notch matrix. Well, maybe with the standard matrix, it does work correctly, but so far all the tests I've done, (+Notch -Trellis) is better than (-Notch +Trellis) or (-Notch -Trellis).
I think I'm not the ony one who did those tests. Ask dialhot, Inc. and others here

-kwag