SAFR was tested against the Large-scale CelebFaces Attributes (CelebA) Dataset for gender detection accuracy.  Below table provides results of this testing.

 

CelebA
Accuracy
CelebA
False Male Rate
(41.69% Male) 
CelebA
False Female Rate
(58.31% Female) 
98.3%2.5% of actual males
(1.25% of entire set)
1.2% of actual females
(0.6% of entire set)


False male rate is number of females who were falsely reported as male.  Likewise, False female rate is number of males falsely reported as females.