[XviD-devel] [In progress] devapi4 -- mpeg matrices

[Edouard Gomez]

Michael Militzer (michael at xvid.org) wrote:
> I  don't  think it's  really  needed to  replace  the  whole code  and
> introduce   just  new   anomalies   (different  precision,   different
> dead-zone) that lead to a different overall behaviour of XviD.

h263 is identical. mpeg will be  identical too once i reintroduce the RD
optimized bias.

> I think  that the  speed-wise difference between  Skal's and  our code
> also shouldn't be  big - our code is quite fast  also (thanks Skal for
> tuning it up)...

It's not about  speed. It's about using a  well designed quant framework
(slightly modified)  that has  never suffered the  design errors  of our
current code.

-- 
Edouard Gomez