source: https://coremltools.readme.io/docs/image-inputs

classifier_conversion_tf_to_coreml.py

import os

import tensorflow as tf
import coremltools as ct

target_dir_path = # in my case, this path was output model path from tensorflow model maker
saved_model_path = os.path.join(target_dir_path, 'saved_model')
label_file_path = os.path.join(target_dir_path, 'labels.txt')

# convert tensorflow model into Core ML model
def get_bias_and_scale(preprocessing_type='scale_0_1'):
    if preprocessing_type == 'scale_0_1':
        return [0,0,0], 1/255.0
    elif preprocessing_type == 'scale_m1_p1':
        return [-1,-1,-1], 1/127.5
    elif preprocessing_type == 'pytorch':
        return [- 0.485/(0.229) , - 0.456/(0.224), - 0.406/(0.225)], 1/(0.226*255.0)
    elif preprocessing_type == 'none':
        return None, None
    return [0,0,0], 1/255.0

bias, scale = get_bias_and_scale('scale_0_1')
if bias is not None and scale is not None:
    inputs = [ct.ImageType(bias=bias, scale=scale)]
else:
    inputs = [ct.ImageType()]
class_labels = [label for label in open(label_file_path, 'r').read().split('\n') if len(label)>0]
classifier_config = ct.ClassifierConfig(class_labels)
mlmodel = ct.convert(saved_model_path, inputs=inputs, classifier_config=classifier_config)

mlmodel.save("MyModel.mlmodel")