12. Core ML
• Announced at WWDC 17.
• Works only on iOS 11
• Inference engine
• Does not support training models.
• Heterogenous compute architecture - CPU & GPU.
• Open Source Python coremltool for converting
to coremlmodel file.
22. Core ML Model
• Single Document
• Public Format
• Reduced Size
• Improved Accuracy
• Decreased Prediction
Time
23. Core ML Architecture
• Unified Fine Tuned Inference Engine
• Xcode Integration
• Built on Accelerate and Metal
• Public Model Format
• Support for Popular Training Libraries
24. Core ML Model
• Function learned from data
• Observed inputs
• Predicts outputs
• Single Document
• Public Format
• Ready to Use
• Task Specific
27. Converting to Core ML
• Download a trained .caffemodel and
a .prototxt of the dataset, as well as a .txt list
of names related to the model.
• Install Python 2.7 and pip
• Install, create, then activate the virtualenv.
• Write and run a Python script to convert
the .caffemodel to a Core ML model,
using coremltools
• Add the generated .mlmodel to the Xcode project