Make sure you get the correct meaning of the 90%. This article isn't talking about a 10% false positive rate, which would mean that 10% of OK pictures are false

Make sure you get the correct meaning of the 90%. This article isn't talking about a 10% false positive rate, which would mean that 10% of OK pictures are falsely flagged as bad. That would be awful. The article is actually talking about 90% precision, AKA 10% false discovery rate, which means that 10% of the pictures flagged as bad are actually OK. That's a fairly good precision - however it is only half the picture, as the false positive rate definitely needs to be specified (and be very low).