[XviD-devel] Early SKIP and early DC-only

Christoph Lampert chl at math.uni-bonn.de
Mon Apr 21 19:26:21 CEST 2003


On Mon, 21 Apr 2003, Marco Al wrote:

> From: "Christoph Lampert" <chl at math.uni-bonn.de>
> 
> > which finally gives some background about when early SKIP of a block might
> > be a good choice.
> 
> It only lets you guess when cbp will be 0, it does not adress the effects of
> the resultant discontinuous motion fields on perceptual quality and MV bits
> (of the subsequent MB). The way MPEG4 assigns MVs to skipped MBs causes a
> lot of problems which need more to solve them than simply knowing when cbp
> will be 0, it is not like a DCT+quant step takes meaningfull time on a PC
> anyway.

Yes, of course. I didn't mean "skipped" macroblocks in the way that
not_coded=1 etc. only no DCT coefficients. 

Of course DCT and quant are fast steps, but skipping them is still faster
;-) 
Actually _all_ routines are fast in XVID (ecxept for the non-ASMed GMC),
but I'm sure there is still some speedup possible. 

 
> Glad to hear H264 used a sane way of dealing with MVs of skipped blocks ...
> hard to comprehend how MPEG4 ended up with its braindead scheme.

Hm, I didn't find it that bad. What do you dislike about it? 

gruel
 



More information about the XviD-devel mailing list