acc and val_acc are good but there is no prediction

pierluigiferrari / ssd_keras

A Keras port of Single Shot MultiBox Detector

Apache License 2.0

1.86k stars 938 forks source link

acc and val_acc are good but there is no prediction #352

Closed mouad47 closed 4 years ago

mouad47 commented 4 years ago

hello @pierluigiferrari first thank for this work, it's amazing i'm having problem with the prediction, i don't know where is the bug, loss and val_loss are zeros, but i when i ran the prediction it's show nothing even with test images in the data it's self, i'm ussing a defrent dataset, and it's labelled properly, i had some error before that but i fixed them, do you have any idea? please

stale[bot] commented 4 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

Ezzysci commented 4 years ago

I have the same problem and it has to do with the confidence threshold. When I put it 0.5, I get no prediction, when I put it as 0.09, I get predictions with too many bounding boxes. What could be the general reason for this? Just as a guiding principle.