Converting from TensorFlow or Keras

To convert our model from Keras or TensorFlow, another tool is needed—tf-coreml (https://github.com/tf-coreml/tf-coreml).

At the time of writing, tf-coreml is not compatible with TensorFlow 2. We have provided a modified version while the library's developers are updating it. Refer to the chapter's notebook for the latest installation instructions.

We can then convert our model to .mlmodel:

import tfcoreml as tf_converter

tf_converter.convert('frozen_model.pb',
                     'mobilenet.mlmodel',
                     class_labels=EMOTIONS,
                     image_input_names=['input_0:0'],
                     output_feature_names=[output_node_name + ':0'],
                     red_bias=-1,
                     green_bias=-1,
                     blue_bias=-1,
                     image_scale=1/127.5,
                     is_bgr=False)

A few arguments are important:

• class_labels: The list...