TMpgEnc Default Matrix

[ookzDVD]

I have been using the TMpgEnc default matrix,
I've found that it perform between the H.263 and MPEG.

Mind if you try and tell me about your comments ?

Thank you.

PS. Download the matrix (http://www.freewebs.com/ekoboy/tmpgenc-default.zip).

[WyldeF1r3]

the link just cycles back to this post

[Mel Maconoo]

right click save target is how i did it.. ^_^

[WyldeF1r3]

heh, silly me, works now. thanks

[BoNz1]

Hi, I have been using the tmpgenc matrix a couple times myself as well. I have found it to be about the same as mpeg in some of my tests maybe even a bit worse. The matrix that I really like is CCE's low bitrate matrix. In my tests, 1st pass size with it was about equal to h.263 size but I would say that the quality of it was a bit better. I did my testing on very noisy sources. :) I won't post the CCE low bitrate matrix because I think someone else already has a while ago, so use the search if you want it.

[sam_b]

I am still having brilliant results with andreas 78ers matrix that wotef linked to a while back in the Divx3.11/XviD thread. Will try the default Tmpgeng matrix you posted in a bit.

[ookzDVD]

@sam_b,

Any chances to test the matrix ?

I have tested the andreas 78er matrix, it seems it will produce the smaller 1-pass size compared with the default tmpgenc matrix but
it's still larger than the h.263.

Thank you.

[sam_b]

Only have "So I married an axe murderer" vobs on drive ATM which are pretty bad source but will give it a go now. Mixing 22Gigs of 10-track audio at the moment which is eating quite a bit of space on my drive.

The 78ers matrix I find to have a huge q2 size but I find it can to down to quant 4-5 and still look prett good. At this point I find it rings less than MPEG, and q2 with this matrix makes a good alternative to q1 with the default matricies. Keeps a good bit of detail to my eyes.

Will give yours a go and get back to you.

[sam_b]

Okay, tried it out (1/150/100/0 BF VHQ1 CM MP6)

I'm afraid I can't really spot the difference compared to default MPEG. Both look damn similar. Also, the test filesizes (approx 10Mb) differ by only 21 bytes at same quality, 97. 78ers matrix gave 14.4mb at same quality. Quality 94 gave 10mb.

Are you sure that this is not just the standard MPEG matrix?

[Didée]

Originally posted by sam_b
Are you sure that this is not just the standard MPEG matrix?
It is the other way round: you used Andreas78er in both cases ...

After fiddling around with custom matrices, it often may happen that the custom matrix gets "locked", so that when you're switching back to "standard" mpeg quantization, in fact you're still using the last used custom matrix.
And that seems to be the case here ... Andreas' 87er matrix produces much higher bitrate at a given quant than standard mpeg. It is quite impossible that those two should only differ by some bytes in the end, using the same quality setting!

BTW, hitting "load defaults" won't get you out :P You must re-install XviD to unlock the matrix again. But hey, that's one double-click and some seconds waiting, it's okay ... I got very used to that procedure in the meantime!

[sam_b]

Didee, I deliberately swapped between the three matricies twice (6 tests) to check this result. Both 78er's were 14.4MBs and the rest were all 10 MBs. These were all at quality 97. The 14.4MB one looked like andreas and the other two looked like MPEG. I have never experienced your problem, which would be easily spotted due to the size differences. All my switches between H263, MPEG and custom have worked fine.

I will reinstall xvid a couple of times to check, to I'm pretty sure my results are accurate.

----EDIT----

Procedure:
1: Reinstall Nics latest after loading defaults, then reload defaults
2: encode clip (just ChromaM this time) using MPEG : 9.12Mb
3: encode clip TMPGEng matrix: 9.24MB
4: re-encode with MPEG: 9.12MB
5: encode with 78ers: 13.5MB
6: re-encode with MPEG again: 9.12MB

Although this time there was a slight size difference, I still couldn't see any difference. I also noticed that TMPGENC inter-matrix is the same as Xvid's default custom inter-matrix. Intra matrix is different.

I'm afraid I cannot find any locking of matricies, as this test shows. It seems to behave itself perfectly.

[Didée]

Try also to create some matrices, and load and save a handful matrices one after another, making a short test with each.
It does not happen immediately, and it does not happen everytime.
But it does happen sometimes, and there have been other reports, too.
However, it is a "feature", not a bug. Once the matrix has locked, you are able to use a custom matrix along with "modulated" quantization ;)
Furthermore, I didn't test that behaviour with 08-03 and 16-03 builds from Koepi and Nic. Perhaps its gone in the meantime?

[sam_b]

Havn't really come across this problem at all before that I'm aware of, maybe I'm just lucky. Havn't really used Modulated for a long time, as I like B-frames. What matrix do you mix H263 with via this 'feature'?

P.S. What is your opinion on the 78ers matrix, as I'm now using it A LOT? At quant 2 I find it's simply stunning if you've got the bitrate.

[Didée]

I don't use modulated at all, but (because) bframes a lot ...

Well, IF you have the bitrate, Andreas78 looks amazing. Confirmed.

Now, if you have EVEN MORE bitrate, you should try "DoubleBRmatrix". It looks unbelievable.

:) Didée

[kilg0r3]

Originally posted by Didée
I don't use modulated at all, but (because) bframes a lot ...

Well, IF you have the bitrate, Andreas78 looks amazing. Confirmed.

Whoops, I am using the 78er matrix for encodes with a target size of 42% of the first-pass size. What do you use for this range?

Now, if you have EVEN MORE bitrate, you should try "DoubleBRmatrix". It looks unbelievable.

Where did you get that one? Search and Google didn't yield anything. And for which compression ratio would you use it?

[sam_b]

It's ok kilg0r3, I often use it at quality <90. I find it scales pretty well, but it won't be putting RealVideo's streaming out of business anytime soon :)

But it doesn't ring much, so I'm happy.

[Didée]

Originally posted by kilg0r3
Whoops, I am using the 78er matrix for encodes with a target size of 42% of the first-pass size. What do you use for this range?

Where did you get that one? Search and Google didn't yield anything. And for which compression ratio would you use it?
Well, I found that matrix in N:\CusomMatrices\ on my PC - maybe the Google robots have missed that one ;)

For what compressability - well, depends on your Bframe-settings ... above 65%, I'd say.
Speaking of quantizers, that matrix is my choice when average quant will be at least < 3, or better < 2.8.

...

Argh, still no attaching possible.
I'll try e-mail, kilg0r.

<edit>
... what is difficult without an adress, and I have no place to upload that file. Perhaps drop me an adress by PM?

[sam_b]

I will put it on our server if you'd like. email stuff@briarwood.f9.co.uk. Possibly a repository of matricies?