[XviD-devel] BFRAMES

[elcabesa]

> It's being done after each motion search step - in the last lines of
> SearchBF() and SearchInterpolate(). The code compares the best SAD
> found in current search against "best_sad" variable. If "bast_sad" is
> higher (== worse), it updates MACROBLOCK information with new data
> (new mode, motion vectors etc), and also updates best_sad with new
> value.
> If best_sad was better, nothing happens.
> SearchDirect() is first, so we don't do this, it just updates the
> macroblock info with "direct" mode.
>
> Regards,
> Radek

thank you
i found it in code  and look them
while i was waiting for a reply i hacked a little xvid code and found 
something i can't undertand

before explain what i don't understan i need some help
MODE_FORWARD  or MODE_BACKWARD are one direction  prediction
MODE INPTERPOLATE is 2 direction prediction  frame i = (frame i-1 +  frame 
i+1)/2
am i right??
what is MODE_DIRECT?

now this is the result i found in a 140 s encode
70855 mblock encoded as b    (20MacroBlock /frame is tipe B) (image size 
640*272 -> 40*17 macroblock = 640)   (only 3% of mb are B block)

32 MB  of type MODE_INTERPOLATE (only 0.045%)
12960 of type MODE FORWARD (18%)
6409 of type MODE_BACKWARD (9%)
and
51451 of type MODE_DIRECT (72.5%)

isn't mode direct to used?? and mode_interpolate not too used..


another question thast came to my mind now

somedays ago i was talking with gomgom and he told me that using 
DIVX_compatiblity  you have not bframe problem couse it add p frames before 
each I frame
but a good working  b frame should notice (NOTICE ==found a high SAD)  that 
using the I frame (scene change) for B frame prediction is not good and use 
only n-1 frame for his prediciton changing it to MODE_FORWARD OR BACKWARD .. 
i don't know the right one

this is what i mena

PBBI
1234

frame 4 should be tipe b but refer only to frame 1  , or almost all his MB 
should refer only to frame 1 and not to frame 4

am i totally wrong??