coremltools_tensorflow_mobilenetv2_conversion_and_prediction.py

1import tensorflow as tf
2import coremltools as ct
3import numpy as np
4from PIL import Image
5
6# 1. Load a pre-trained Keras model
7keras_model = tf.keras.applications.MobileNetV2(weights="imagenet")
8
9# 2. Convert the model to Core ML
10# We use ClassifierConfig to map the ImageNet class labels
11model = ct.convert(
12    keras_model,
13    inputs=[ct.ImageType(shape=(1, 224, 224, 3), bias=[-1,-1,-1], scale=1/127.5)]
14)
15
16# 3. Save the converted model
17model.save("MobileNetV2.mlpackage")
18
19# 4. Load the model and make a prediction
20# Note: Making predictions with Core ML models via coremltools requires macOS
21loaded_model = ct.models.MLModel("MobileNetV2.mlpackage")
22
23# Prepare a dummy input image
24example_image = Image.fromarray(np.uint8(np.random.rand(224, 224, 3) * 255))
25out_dict = loaded_model.predict({"input_1": example_image})
26
27print(out_dict)