[x264-devel] Question about MV initial cost calculation

[Shen Sha]

Dear experts,

I'm reading the motion estimation realed x264 code now.  I know that when performing mode decision, cost for all possible partition/mv should be calculated.  I see the following code in file "Analyse.c", which is obviously doing some initialization for the MV cost calculation.

int x264_analyse_init_costs( x264_t *h, int qp )
{
    int lambda = x264_lambda_tab[qp];
    if( h->cost_mv[lambda] )
        return 0;
    /* factor of 4 from qpel, 2 from sign, and 2 because mv can be opposite from mvp */
    CHECKED_MALLOC( h->cost_mv[lambda], (4*4*2048 + 1) * sizeof(uint16_t) );
    h->cost_mv[lambda] += 2*4*2048;
    for( int i = 0; i <= 2*4*2048; i++ )
    {
        h->cost_mv[lambda][-i] =
        h->cost_mv[lambda][i]  = lambda * (log2f(i+1)*2 + 0.718f + !!i) + .5f;
    }
........
 
}


I have the following question about the equation "h->cost_mv[lambda][i]  = lambda * (log2f(i+1)*2 + 0.718f + !!i) + .5f;
" used above:

(1) why the value 0.718 used in the above calculation? Is there any papers/documents which introduce why this number is chosen?
(2) why the log2f(i+1)*2  is used instead of just log2f(i+1)?  In my opinion, the bit lenght of mv is log2f(i+1), why here this bit length is doubled?


Thanks a lot for your kindly help.

Sha