Training with Core ML 3

This is the sample code for my blog post series On-device training with Core ML.

Included are:

• Dataset: a small dataset of 30 training images and 15 test images

• iOS App: the source code of the demo app described in the blog post

• Models: the empty and pre-trained models used by the app

  • TuriOriginal.mlmodel: the SqueezeNet classifier trained by Turi Create
  • HandsTuri.mlmodel: the TuriOriginal model but made updatable
  • HandsEmpty.mlmodel: like HandsTuri but with a classifier layer that has random weights
  • HandskNN.mlmodel: like TuriOriginal but with an untrained k-Nearest Neighbors classifier

• Scripts:

  • make_nn.py: converts TuriOriginal.mlmodel to HandsTuri and HandsEmpty.mlmodel
  • make_knn.py: creates the k-Nearest Neighbor model, HandskNN.mlmodel
  • TuriCreate.ipynb: the notebook used to train TuriOriginal.mlmodel

Credits:

• Camera icon made by Daniel Bruce from www.flaticon.com and is licensed by CC 3.0 BY.
• Picture icon made by Dave Gandy from www.flaticon.com and is licensed by CC 3.0 BY.

The source code is copyright 2019 M.I. Hollemans and licensed under the terms of the MIT license.