[MPlayer-dev-eng] comprehensive lavcopts test - in progress.

[Romain Dolbeau]

Michael Niedermayer wrote:

> btw, it would be interresting to know how many local minima are there in your 
> results
> i mean cases where no single changed option leaded to an improvement

that would be local maxima wrt PSNR ?

That would be 3 if I'm not wrong:

[cmp=6, dia=2, mbd=2, mv0, precmp=6, predia=2, qpel, subcmp=6, trell, v4mv, vmax_b_frames=0, vme=4]:44.63
[cmp=3, dia=-2, mbd=2, mv0, precmp=3, predia=-2, qpel, subcmp=3, trell, v4mv, vmax_b_frames=1, vme=4]:44.65
[cmp=3, dia=2, mbd=2, mv0, precmp=3, predia=2, qpel, subcmp=3, trell, v4mv, vmax_b_frames=1, vme=4]:44.65

You can't go from the first to one of the other 2 with
only 1 change, and the last two are the same, PSNR-wise.

The opposite (local minima for PSNR, i.e. you can't worsen
the result with 1 change):

[cmp=0, dia=2, mbd=2, precmp=0, predia=2, qprd, subcmp=0, vmax_b_frames=1, vme=4]:42.75
[cmp=0, dia=-1, mbd=2, precmp=0, predia=-1, qprd, subcmp=0, vmax_b_frames=1, vme=4]:42.75
[cbp, cmp=0, dia=-1, mbd=2, precmp=0, predia=-1, qprd, subcmp=0, trell, vmax_b_frames=1, vme=4]:42.96
[cbp, cmp=0, dia=2, mbd=2, precmp=0, predia=2, qprd, subcmp=0, trell, vmax_b_frames=1, vme=4]:42.96

But those are not very useful in practice :-)

(this is from the 2352 results vme=4, mbd=2)

-- 
Romain Dolbeau