Converting Teachable Machine trained model to CoreML model

Teachable Machine

Teachable machine is a web based tool to train ML models using Tensorflow.

Training the model

• Create classes, and name the classes with classifying labels.
• Train the model, if required you can change parameter Learning rate and number of Epochs.
• We can upload the trained model to the cloud and then access it using the link provided or download the Tensorflow models locally and use it for further implementation.
• Tensorflow’s downloaded model can be saved/downloaded in two formats.

Tensorflow SavedModel (.pb extension)

Keras .h5 model

to know more about training on Teachable Machine you can read this tutorial.

Converting the model

• If you are working on mac then to support the script that uses Tensorflow packages, you can refer to the link below for setup instructions.

https://github.com/jeffheaton/t81_558_deep_learning/blob/master/install/tensorflow-install-mac-metal-jul-2021.ipynb

Use python scripts supported with corelmltools , keras.models and tensorflow modules for conversion.

labels.txt file that contains the class names, above has 3 classes as I had trained my model for 3 celebrities

from keras.models import load_model
import coremltools as ct
import tensorflow as tf
model = tf.keras.models.load_model('converted_keras/keras_model.h5')
image_input = ct.ImageType()
f=open("converted_keras/labels.txt", "r")
contents = f.read().splitlines()
for i, label in enumerate(contents):
 if isinstance(label, bytes):
   contents[i] = label.decode("utf8")
classifier_config = ct.ClassifierConfig(contents)
image_input = ct.ImageType(shape=(1, 224, 224, 3,),bias=[-1,-1,-1], scale=1/127)
mlmodel = ct.convert(
   model,
   source="tensorflow",
   inputs=[image_input],
   classifier_config=classifier_config,
)
mlmodel.author = 'Pranav'
mlmodel.short_description = 'Celebrity Recognition with TeachableMachine'
mlmodel.save('celebrityDetection.mlpackage')

• In the above script I have used the Keras (.h5) model and converted it to the CoreML model.
• Once you get the CoreML model from above script you can import it to desired applications and use it for predictions.
• The model takes images of size( 224 X 224 ) as input for predictions and gives labels as output, hence before giving input to model for predictions.

> Run a face detection function on the desired image.

> Crop faces to face bounds(You can use CIFeatureDetection or VisionFramework)

> Resize(224 x 224) the image(face) and convert to CVPixelBuffer

> Make the predictions