Handwritten digit recognition with Keras/MNIST on iOS

This deep learning sample application shows how to use a Keras model with the iOS CoreML framework in Objective-C. The used model implements a convolutional deep- neural network that is written in Keras in order to recognize handwritten digits based on the MNIST data model. The .h5 model that is created by Keras is then converted into a .mlmodel using coremltools that can be used in iOS new CoreML framework. Although the results of this model are not perfect, it is a good start into Machine learning in iOS in general. You can clone or download this repo and try to play around with the application and test the model.

A complete Tutorial for this App can be found here

Implementation Prerequisites

In order to implement this completely by yourself you will need to have your working-station setup as follows:

Deep Learning

• Tensorflow 1.2.1
• Python 2.7
• Keras 2.0.6
• coremltools 0.5.1

iOS Development

• MACOS Sierra 10.12.6 (this is mandatory for Xcode Version 9.0 beta)
• Xcode 9.0 (accessible with a developer account, currently beta)
• optional iOS 11 installed on your iPhone (if you want to test the app on your iPhone, Simulator is totally sufficient.)