Log in

View Full Version : Do matrix-patch really improve SVCD quality ?


waldok
23rd January 2002, 18:41
Hi,

Sorry folks if this has been asked before.

I saw some threads discussing how to use custom matrices in CCE encoding, and I didn't know it was possible until recently. Still, I couldn't find any comment from SVCD makers out there on the efficiency or noticeable quality improvements on the final picture.
Can you tell me more about subjective benefits of this operation or point me to user comments or test bench? On what particular points can you notice improvements ? Less macroblocks ? Less mosquito effect ? More compression for same quality ? Something else ?

THanks a lot for the info !

Waldok:cool:

Deltis
24th January 2002, 02:30
Hi waldok!

I did some matrice-patching and tested some of the ultra-low-bitrate-ones. Unfortunately they did't seem to produce better results (e.g. "Braveheart " on two SVCDs).

You should test it with a short scene including normal and fast motion. AFAIK there were some weird errors in fast motion with some of the ultra-low-bitrate matrices. The standard CCE 2.6 matrice seems to handle most input quite well. IMHO there's no need for matrice-patching...default matrices of CCE perform well and resizing methods might have a greater and more predictable influence :)

Perhaps somebody else can tell you something about the results with special CG-matrices (for animes, etc.) ...

cu
Thomas

waldok
24th January 2002, 07:18
THanks a lot for your input here.:)

I'll still wait for some other opinions before I decide to give this matrix thing a try.

Any other experience ?

Waldok:cool:

kdiddy
24th January 2002, 07:31
Well I think it "may" depend on how you are viewing...if you are viewing just the video via PC, there is probably no difference....however if you are looking at the muxed mpeg version, you "may" notice slight differences...it was once posted on here on a thread I cant find right now...that CCE does not embed the matrix codes into its video streams, it only uses them for the encoding process...thus the your standalone players decoder will not decode the stream based upon that matrix...it went on to advise people to mux with the highy recommended BBmpeg, because it allows you to mux with the matrix settings you use to encode with, and then you player will decode correctly....now having said all that, to my poor eyes...I still couldnt see any difference.

Michael J
24th January 2002, 18:25
Hi,

I think a good matrix does dramatically improve image quality especially at low bitrates. Look into this posting:
http://forum.doom9.org/showthread.php?s=&threadid=12161&pagenumber=2

The matrix posted by mb1 is quite good, I think it was written by some German guy called Andreas (hence it's called Andreas matrix over here in Germany).

Plus: You don't need to patch CCE 2.62, only 2.50 needs to be patched. If you use the latter one, you need to add the intra and non-intra matrix for muxing the mpeg (e. g. in BBMPEG)! Don't forget this, or your result will be an MPEG with serious color shifting problems and maybe a grainy picture!

TactX
24th January 2002, 20:05
I've tested this matrix some time ago and it gave a bigger file (same CQ-value) without looking better (I did even compare frame by frame). So it can't be a very good solution for low bitrates. It might be better on some sources but for the average source the standard matrix (used as default by CCE, TMPGenc and most encoders) is propably the best. And the developers of these encoders are definetely more experienced with MPEG than any member of this forum.

Think about it...

Deltis
24th January 2002, 21:51
Hi!

I did the same frame to frame comparison (on my PC) and wasn't very satisfied with the results, too. The ultra-low-bitrate-matrices - what a word :) - didn't perform very well.

But perhaps it's just the mentioned problem with the matrice not being embedded in the encoded stream for proper decoding.

This might change the whole thing a little bit - further tests will be necessary using bbMPEG :(

There's only one problem left - the TV I own is too small for seeing the difference on my SVCDs - even if it really works ;)

cu
Thomas

Michael J
26th January 2002, 09:12
Originally posted by TactX
I've tested this matrix some time ago and it gave a bigger file (same CQ-value) without looking better (I did even compare frame by frame). So it can't be a very good solution for low bitrates. It might be better on some sources but for the average source the standard matrix (used as default by CCE, TMPGenc and most encoders) is propably the best. And the developers of these encoders are definetely more experienced with MPEG than any member of this forum.

Think about it...

I thought about it, but you DO know that there are settings made by FOX engineers (and Universal techies, too, I think) that differ largely from the standard CCE setting? Those guys are definitely experienced with MPEG, so there seem to be different opinions about appropriate CCE settings!

Btw: CQ-value? You were doing 1-pass-encoding? Or was that the first step of Robshot's 2-pass-encoding-way? Anyway: With Andreas Matrix I don't have to use TemporalSmoother, no need for AntiNoiseFilter, plus I get a very small, good quality MPEG2. And without the filters even four or five passes don't take too long. 144 minutes on 2 CDs in high quality ain't bad after all, right? And it looks great even on a large Panasonic tv-set.

Try it, and then trust you eyes. If it doesn't show the result you're looking for, then just stick to your encoding technique. We're all still looking for THE best way to encode MPEG2 with CCE, aren't we?

TactX
26th January 2002, 12:19
I need to be honest. I tested this matrix with TMPGenc, the encoder it was "developed" for.
But if this matrix is NOT performing well with TMPGenc it can't be too good for CCE.

I hate quoting myself but...
It might be better on some sources
I did NOT say that this almighty "Andreas-matrix" looks worse on any source. And the Fox-matrix you mentioned is definetely out-of-competition for SVCDs since it was designed for ultra-high-bitrate-DVDs.

And btw I think that developing an encoder needs much more "MPEG-experience" than DVD-encoding/mastering. And if I have a look at some (commercial) DVDs, I think that some of these "DVD-professionals" should visit Doom9's Forum a bit more often.