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

[Radoslaw 'sysKin' Czyz]

Hi again,

> I made a simple ME-similar routine for P/B/I-frame decision.

I definitely have good news for anime coders :) Encoder seems to put
a P-frame every time an anime scene is changed considerably (as it
happens in movement) but keeps virtually any number of B frames if
scene only pans or stays static.
As a result, all of my encodings at fixed quant are 70..75% smaller,
while I bet noone could ever say that B-frame encoding looks worse. In
fact, I'd say that b-frame clip looks better because B frames seem to
fight the "static noise on smooth moving wall" artifact.

I used max_bframes = 8 and b-quant ratio of 200%. It looks good with
250% too, because there is no danger that B frame will appear at a big
scene change (it will be coded as at least P frame then).

What do you think?

Radek