using the new mobileNet weights.

[RadEdje]

Hello, I'm really new to machine learning, Keras, and tensorflow. I'm currently looking for an efficient way to retrain a mobileNetV2 with imagenetWeights when I saw your github repo. Sorry If I have a totally newb question here.

I'm currently reading the keras github repo for moblieNetV2 here: https://github.com/keras-team/keras/blob/master/keras/applications/mobilenetv2.py

so to use your keras model weights I just put that in a directory AND assigned that at line 106.

BASE_WEIGHT_PATH = 'https://github.com/JonathanCMitchell/mobilenet_v2_keras/releases/download/v1.1/' etc?

what is the difference between the larger file at around 6mb and the one with no the top 5? (smaller mb).

Any advice would be much appreciated. Thanks.

[RadEdje]

oh sorry... reading other docs... so the no_top.h5 file is the one withhout the top layer or final fully connected layer classifying the 1000 categories of imagenet. I will use this so I can re-add a new layer with my own number of classes during retraining correct?

[pedromlsreis]

@RadEdje might be handy for you: https://stackoverflow.com/questions/46742491/why-do-we-call-the-fully-connected-layers-in-cnn-the-top-layers