Re[2]: [XviD-devel] motion estimation for B/P/I decision

[Marc FD]

@gruel, sysKin

For I/PB frame decision, maybe my histogramm detection could be very useful.
but i need to work a lot on it to get it ready.

is SKIP mode possible for BVops ?
i think DIRECT mode could give almost the same results.
i really fear smearing in XviD. (and i think i'm not alone)
i've tested adaptive coef thresholding, who changed TOO_SMALL_LIMIT
 to 20 for non moving (0,0) MBs with low SAD16.
it gived me the (bad?) results you know (-15% size, -0.1% PSNR)
and the non-moving MB condition was needed to avoid smearing
very slow gradient moves are estimated by the ME, but you can regognise them
before.
early SKIP decision is impossible if you want to avoid smearing.

BTW, i'm ready to test as soon as the split is over.
and i've some french friends ready to test too ;)

(OT)
@MarcoAI
do you have any C implementation of Zhou Wang's structural distortion
measure ??
i could'nt find anything usefull on the site. you said it was better than
PSNR ?
how did you tested ? it would be usefull to have another objective value for
"quality" comparaisons.

Thx.