Mahabali / TensorFlowForMobile

Create tensorflow model for mobile apps. TFLite for Android and CoreML for iOS.

TensorFlowForMobile

This is a simple tutorial for implementing machine learning with custom model in mobile apps. This tutorial consists of 3 parts

Repos

Simple TensorFlow Model for computing y= 2x+1. There are 3 repos

1. Tensorflow model
2. TensorflowLite Model for Android.
3. CoreML model for iOS (using CoreML tools)

I could have used tensorflow lite for iOS. I implemented CoreML because, its easier with documentation and debugging. CoreML also supports on device machine learning. TFLite in iOS does not support on device machine learning.

I have used 'Nadam' optimizer with Mean absolute error regression loss. Feel free to play with different optimizer.

How to use

1. Go to colab.google.com
2. Open the jupyter notebook(ipynb) in this repo
3. Click 'Connect' and wait till you get a free connection to google compute python backend
4. Run the code

I have added comments in the jupyter notebook which are self explanatory.

When you successfully run all the code, you get the following items in files

1. SampleModel.h5 - Tensor flow model which can be used later
2. SampleTFLiteModel.tflite - TFLite Model for Android
3. SampleMLModel.mlmodel - CoreML Model for iOS

iOS Screenshot

Android Screenshot

My Product

whichidiot.com

My consulting site

okchanges.com