H.263 QM: What are their coefficients?

[lcld]

I know, H.263 isn't a matrix-with-fixed-coefficients. At least, that's what I read several times. But how it is possible?

I can't understand how one can experiment with CQM without knowing what H.263 is.

And what about MPEG?

[Poutnik]

I do not know much about H263 but it uses different quantization approach,
comparing to MPEG ( no matter if standard MPEG or custom MPEG matrix ).

So CQM have nothing to do with matrices, so one need not to know it.
But fore sure it is useful to know what it is.

Try Google to find some H263 related resources....
I know there are..

[iradic]

sorry if OT but what reg value tells xvid which matrix to use?

[lcld]

Try Google to find some H263 related resources....reading http://www.uofaweb.ualberta.ca/mccl/pdfs/MEngReport-LinWang.pdf...
sorry if OT but what reg value tells xvid which matrix to use?HKEY_CURRENT_USER\SOFTWARE\GNU\XviD\quant_type
HKEY_CURRENT_USER\SOFTWARE\GNU\XviD\qmatrix_inter
HKEY_CURRENT_USER\SOFTWARE\GNU\XviD\qmatrix_intra
???

[Poutnik]

IMHO quant type says if h263,MPEG, or MPEG custom is used.
inter/intra keys say which koefs would be used, if Custom MPEG is chosed.
So there is no link to matrix.
When matrix is loaded, data are just copied to these reg items.

[lcld]

(The first time I tried to find the answer in xvid's source, I was completely lost.
Now, with that doc, I think I have understood.)

H.263 quantization woulld be a CQM full of 16 with a slightly different algorithm : results of divisions are "truncated", instead of being "rounded".

And MPEG quantization is just a predefined matrix.

Can anyone confirm this ?

[Sharktooth]

h263 is a quantization type, not a a quantization matrix.
DivX fusion uses an "optimized" h263 algo, while xvid uses standard h263.

[akupenguin]

DivX fusion uses an "optimized" h263 algo, while xvid uses standard h263.
I was under the impression that DivX's "Rate-distortion optimized H.263 quantization" was just another name for Trellis. After all, they can't change the behaviour of the decoder and still be MPEG4 compliant, so the only choices available are what dct coefficients to encode. And Trellis is precisely the rate-distortion optimal choice of coefficients.

[Manao]

akupenguin : It's indeed trellis that is used.

Lcld : I confirm you observation : roughly, h263 is equivalent to a matrix filled with 16s. Roughly only. MPEG is indeed equivalent to MPEG-Custom with a predefined matrix - this matrix is defined in the MPEG-4 norm.