Core ML Conversion Script for the Keras Facenet Model

convert.py

import coremltools
from coremltools.proto import NeuralNetwork_pb2
from coremltools.models.neural_network.quantization_utils import *
import numpy
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
from keras.utils import np_utils
from keras.models import load_model
import os.path
import sys

sys.path.append('./code/') # Import "code" from https://github.com/nyoki-mtl/keras-facenet 

from inception_resnet_v1 import *
model = InceptionResNetV1(weights_path='facenet_keras_weights.h5')

def convert_lambda(layer):
    if layer.function == scaling:
        params = NeuralNetwork_pb2.CustomLayerParams()
        params.className = "scaling"
        params.parameters["scale"].doubleValue = layer.arguments['scale']
        return params
    else:
        return None

coreml_model = coremltools.converters.keras.convert(
    model,
    input_names="image",
    image_input_names="image",
    output_names="output",
    model_precision='float16',
    image_scale=2/255.0,
    red_bias=-1,
    green_bias=-1,
    blue_bias=-1,
    add_custom_layers=True,
    custom_conversion_functions={ "Lambda": convert_lambda })
coreml_model.save('facenet_keras_weights_coreml.mlmodel')

where did you do the L2 Normalization? on the Swift side or ?