https://github.com/ph1ps/MNIST-CoreML

Predict handwritten digits with CoreML
https://github.com/ph1ps/MNIST-CoreML

convolutional-neural-networks coreml coreml-framework coreml-models machine-learning mnist mnist-dataset

Last synced: 3 months ago
JSON representation

Predict handwritten digits with CoreML

• Host: GitHub
• URL: https://github.com/ph1ps/MNIST-CoreML
• Owner: ph1ps
• License: mit
• Created: 2017-06-19T06:18:55.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2018-02-15T17:37:35.000Z (almost 7 years ago)
• Last Synced: 2024-08-03T17:10:01.857Z (7 months ago)
• Topics: convolutional-neural-networks, coreml, coreml-framework, coreml-models, machine-learning, mnist, mnist-dataset
• Language: Python
• Homepage:
• Size: 4.22 MB
• Stars: 67
• Watchers: 4
• Forks: 11
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-CoreML-models - MNIST-CoreML

README

        
# MNIST for CoreML (CNN)

### Description

This is the MNIST dataset implemented in Apple's new framework CoreML. The MNIST dataset can predict handwritten (drawn) digits from an image and outputs a prediction from 0-9. The model was built with Keras 1.2.2.

To test this model you can open the `MNISTPrediction.xcodeproj` and run it on your device (iOS 11 and Xcode 9 is required). To test further images just add them to the project and replace my testing with yours.

An example of a handdrawn digit would look like this: 

![Digit 4](MNISTPrediction/Assets.xcassets/four.imageset/14978202165336.png)

**Be aware that your images have to have a black background and white line color!**

Furthermore your images resolution has to be 28x28px. If yours is bigger just use my `UIImage` rescaling extension I wrote. The line width has to be thick enough to be recognized as a digit.

### Information about the model

This CNN model achieves up to 99.5% of accuracy and the structure is as follows:

![CNN Model](model.png)