2 classes defined instead of just one

yeephycho / tensorflow-face-detection

A mobilenet SSD based face detector, powered by tensorflow object detection api, trained by WIDERFACE dataset.

Apache License 2.0

768 stars 264 forks source link

2 classes defined instead of just one #46

Closed eypros closed 5 years ago

eypros commented 5 years ago

I would like to ask for the 2 classes that you have defined. Is there a specific reason for that?

I have tried to train a one-class with persons and it seemed to me that sometimes multiple classes gives a precision boost for some reason. Is this the case here? Because as far as I can see you don't actually uses the background class at all. Are there any background class member that the model could detect for example?

yeephycho commented 5 years ago

I don't think so, if you found that a single class doesn't works out, it may because of other issue.