Computer vision is a tough thing. Sometimes even Schwarzenegger is identified as female. But overall we reach quite good accuracy on determining gender from face.
Yeah I totally understand that- even humans have trouble sometimes. I'm surprised you even tackled that since everything else seemed to be pretty good.
Just out of curiosity what are some things you expect people to use the gender selection for? I could imagine a toy app that puts hats on people and, if the gender confidence is high, assigns them a "female" or "male" style hat (and a more unisex style if the confidence is lower).
Gender estimation is a must-have feature in modern face detection / recognition software. However, I'm not sure where it is used in real-life besides targeted advertising or some toy applications as you mentioned. For now, we will leave it for our users to decide how to use this feature.
Just out of curiosity what are some things you expect people to use the gender selection for? I could imagine a toy app that puts hats on people and, if the gender confidence is high, assigns them a "female" or "male" style hat (and a more unisex style if the confidence is lower).