Re-loads the weights from the tensorflow hub module.
These weight files can be used with keras's mobilenetv2 model.

I designed the MobileNetV2 model to mimic that of MobileNetV1. The usage is the same, and it can be used in very similar applications in a plug and play manner. I attempted to mimic the style writtein in the original mobilenet: https://github.com/keras-team/keras/blob/master/keras/applications/mobilenet.py

To use MobileNetV2 a user simply has to import it the same way they would import MobileNet

from keras.applications.mobilenetv2 import MobileNetV2
from keras.layers import Input
input_tensor = Input(shape=(224,224, 3)) # or you could put (None, None, 3) for shape
model = MobileNetV2(input_tensor = input_tensor, alpha = 1.0, include_top = True, weights=’imagenet’)
# Now you have a fully loaded model.

This is a keras implementation of MobilenetV2 with imagenet weights for a width_multiplier = 1.0 and input image resolution (224, 224, 3) RGB that is pre-trained on the imagenet challenge.

Individual weights are extracted in the jupyter notebook called "mobilenet_example" and saved to a weights directory, and a layer_guide is provided inside layer_guide.p that has descriptions for all the layers received from reading the Tensorflow checkpoints.

A testing script has been provided and can be found in test_mobilenet.py.