Here what the author tell me in PM when i've asked him the purpose of these new 3&4 modes :
Quote:
Hi. The idea was to improve --aq-mode 2 in dark areas by allowing lower (for negatives numbers it mean higher absolute values) QP offsets there. Approximation of QP-offset/Variance curves of different AQ modes can be looked here: 1..100000 range but most of the difference between AQ 2/3/4 are in 1..5000 range.
P.S. As users didn't really take active part in my testing I decided to give up on mode 4 [(0.25f - 3.3125f / (qp_adj * qp_adj - 0.75f) one] as too weak (it changes curve only at extremely low variance).
|
So, basicaly, ignore mode 4.