[XviD-devel] Comparison "old" vs "new" adaptive quant code

[Dirk Knop]

...and hi again ...

Dirk Knop wrote:

> Attached the modified vfw/codec.c which does exactly this 
> (PMV_ADVANCEDDIAMOND16 for search precision 5, PMV_USESQUARES16 for 
> search precision 6).

I'm redoing matrix again now with these modifications all implemented 
(heh, they work flawlessly, someone _may_ want to add that to CVS, 
finally no need to have three different copies of xvid.h, and a bug is 
smashed in VFW, too), and I'm somewhat amazed what I can see. I have to 
admit, I changed from "dirty" simple resize to the higher quality 
neutral bicubic resize (impact: nearly 5h for 1st pass instead of 4h), 
and what I can see is at least different. Some frames which before were 
around 2-3k show up as 1.0-2.5kb, others like the "telephone number 
search" at the beginning are slightly bigger. We'll finally see when I 
finished the encoding, but it looks promising.

Just wanted to report that those changes have an effect :)

Regards,
Koepi

...who's somewhat amazed that things seem to work better now...