how to reduce the false detection in person detect?

[wzf19947]

I'm training datasets with one class "Person", I found that many objects related to humans like clothes on the chair or hang on the hanger offen got detected, what's the good solution of this problem? Add some clothes pics as negative or sth?

[bulatnv]

Hello @wzf19947. Add more negative examples, more hard ones. Read about Hard Negative Mining: Vehicle Detection in Aerial Images Based on Region Convolutional Neural Networks and Hard Negative Example Mining Efficient Sample Mining for Object Detection