transforms.Normalize(mean(a,a,a,),std(b,b,b))
the torch(transforms.Normalize) formulas : x = (x-mean(x))/std(x) ;
haisi_AIPP formulas :pixel_out_chx(i) = [pixel_in_chx(i) - mean_chn_i - min_chin_i]*var_reci_chn;
calculate the final answer is :
min_chn_i = mean_i * 255; formula(1)
var_reci_chn = 1/(255*std_i) formula(2)
when u use the transforms.Normalize() you must give 2 parms which named mean and std;
so , if u want deploy them on haisi .om you must trans them use formula1 and formula2.