When I use '--no-weights' parameter for training ap is 0

fizyr / keras-retinanet

Keras implementation of RetinaNet object detection.

Apache License 2.0

4.38k stars 1.96k forks source link

When I use '--no-weights' parameter for training ap is 0 #1475

Closed omeryasar closed 3 years ago

omeryasar commented 4 years ago

When I want to initialize weights randomly and training the model from scratch I get 0 ap even if I test my training set ? Any reccomendations or predictions why might it happens. Thanks

hgaiser commented 4 years ago

Why do you want to train from scratch? This is usually a very bad idea. When possible I would always suggest to start from som pretrained set of weights.

omeryasar commented 4 years ago

Experimental reasons. The image resolutions and relative object sizes are too small according to the task that pre-trained models trained on. I wonder what would I get if I train whole model from the very beginning. But as I mentioned something goes wrong and I can't figure out what is it

stale[bot] commented 3 years ago

This issue has been automatically marked as stale due to the lack of recent activity. It will be closed if no further activity occurs. Thank you for your contributions.