[XviD-devel] hq ac prediction mode

Christoph Lampert chl at math.uni-bonn.de
Sun Feb 9 20:25:51 CET 2003


On Sun, 9 Feb 2003, Edouard Gomez wrote:

> Very basic test with "Le fabuleux destin d'Amelie Poulain"
> 
> First 10000 frames sequence (frames resized to 640x272)
> 
> Quantizer |   Size   | Options
> ----------+----------+---------------------------------------------------------------
>     q3      81223900   base options
>     q3      80142222   base options + PMV_CHROMA16 + PMV_CHROMA8
>     q3      81110018   base options + HQ acdc prediction
>     q3      79989286   base options + PMV_CHROMA16 + PMV_CHROMA8 + HQ acdc prediction
> 
> It saves 0.15% for non chromaME and 0.19% for chromaME while
> just using chromaME can save 1.3%

Oh well, better than nothing, isn't it? 0.19% on a full CD is 1.4 MB,
sometimes that's just what's missing to fit on a CD-R ;-) 

gruel

P.S. Btw. Chroma_ME. Does anyone know _why_ chroma ME has such a positive
effect? Is it really the chromanance bits, or is it rather that search
doesn't terminate as early? Or is the result closer to "real motion" , so
motion field get smoother?





More information about the XviD-devel mailing list