`ValueError: Keras layer '<class 'keras_resnet.layers._batch_normalization.BatchNormalization'>' not supported.` when trying to convert to coreml

I learned a model with keras-retinanet which seems to work pretty well, now I would like to convert it to a coreml model to use it in my iOS app:

import coremltools
coreml_model = coremltools.converters.keras.convert(model)
coreml_model.save('my_model.mlmodel')

Unfortunately I get the following error:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-7-ba230c07a72c> in <module>()
      1 import coremltools
----> 2 coreml_model = coremltools.converters.keras.convert(model)
      3 # Saving the Core ML model to a file.
      4 coreml_model.save('my_model.mlmodel')

/home/jonas/projects/keras/keras-env/local/lib/python2.7/site-packages/coremltools/converters/keras/_keras_converter.pyc in convert(model, input_names, output_names, image_input_names, is_bgr, red_bias, green_bias, blue_bias, gray_bias, image_scale, class_labels, predicted_feature_name, model_precision, predicted_probabilities_output, add_custom_layers, custom_conversion_functions)
    743                       predicted_probabilities_output,
    744                       add_custom_layers,
--> 745                       custom_conversion_functions=custom_conversion_functions)
    746 
    747     return _MLModel(spec)

/home/jonas/projects/keras/keras-env/local/lib/python2.7/site-packages/coremltools/converters/keras/_keras_converter.pyc in convertToSpec(model, input_names, output_names, image_input_names, is_bgr, red_bias, green_bias, blue_bias, gray_bias, image_scale, class_labels, predicted_feature_name, model_precision, predicted_probabilities_output, add_custom_layers, custom_conversion_functions, custom_objects)
    541                                            add_custom_layers=add_custom_layers,
    542                                            custom_conversion_functions=custom_conversion_functions,
--> 543                                            custom_objects=custom_objects)
    544     else:
    545         raise RuntimeError(

/home/jonas/projects/keras/keras-env/local/lib/python2.7/site-packages/coremltools/converters/keras/_keras2_converter.pyc in _convert(model, input_names, output_names, image_input_names, is_bgr, red_bias, green_bias, blue_bias, gray_bias, image_scale, class_labels, predicted_feature_name, predicted_probabilities_output, add_custom_layers, custom_conversion_functions, custom_objects)
    185 
    186     # Check valid versions
--> 187     _check_unsupported_layers(model, add_custom_layers)
    188 
    189     # Build network graph to represent Keras model

/home/jonas/projects/keras/keras-env/local/lib/python2.7/site-packages/coremltools/converters/keras/_keras2_converter.pyc in _check_unsupported_layers(model, add_custom_layers)
     98         else:
     99             if type(layer) not in _KERAS_LAYER_REGISTRY:
--> 100                 raise ValueError("Keras layer '%s' not supported. " % str(type(layer)))
    101             if isinstance(layer, _keras.layers.wrappers.TimeDistributed):
    102                 if type(layer.layer) not in _KERAS_LAYER_REGISTRY:

ValueError: Keras layer '<class 'keras_resnet.layers._batch_normalization.BatchNormalization'>' not supported. 

I’m confused by that batchnormalization as i didn’t find that string anywhere in this repo…

Is there a way to convert it to coreml and/or use it in an Android app? I searched for tutorials but didn’t find any unfortunately.

[mxvs]

Just want to add the following: I’ve looked into this before, you need to re-implement all custom layers used by keras_resnet and keras-retinenet in Swift for this to work, this is possible since the latest version of coreml but it will be a lot of work to port everything over. Also I have my doubts if the Retinanet architecture will fit on a mobile device.

[jegarcian]

I would also love to see RetinaNet on iOS. If someone manages to convert it a .mlmodel (or find it somewhere in the web), I would love to test the performance on an iPhone 11 Pro.

I was not able to convert it or adapt it. But in case it helps, I ended up using MobileNet and tflite instead. This tutorial was quite helpful.

https://towardsdatascience.com/deeppicar-part-6-963334b2abe0