machine learning - tensorflow high false-postive rate -


i'm using object_detection tensorflow/models repo. i'm training faster r-cnn resnet101 model own images. in models, use 2-3 classes find particular objects on images.

after 60k training steps, quite results in finding searched objects. problem is, false-positive rate quite high. when searched object in image, result 99%, on 60% of results >90% searched object not there.

does has hints, how can reduce false-positive rate? there way in used model give training specific false-positive examples?


Comments

Popular posts from this blog

angular - Ionic slides - dynamically add slides before and after -

minify - Minimizing css files -

Add a dynamic header in angular 2 http provider -