Hello, I don’t know anything about Edge Impulse but I’m interested in using it for targeting Apple Watch - is that possible? For example, can I export a model from Edge Impulse in a format that can be converted into an Apple Core ML model?
Interesting idea, Edge Impulse uses TensorFlow to train the models, looks like Apple Core ML is an entirely different framework (https://developer.apple.com/documentation/coreml) - at this time, unless you can integrate the Edge Impulse deployed WebAssembly library for example into your Apple Watch application, we cannot currently deploy models as is from Edge Impulse into an Apple Core ML format.
Apparently there is coremltools (https://github.com/apple/coremltools) which can transform a SavedModel to a CoreML format. You can get the SavedModel from the Dashboard under ‘Downloads’, but this will only include neural network part, not any of the signal processing code etc. But as Jenny said not something we support.
Thank you for your answers, that sounds hopeful - I’ll try it out!
@janjongboom would a transformed SavedModel to CoreML still do predictions the same? I would love to generate the model in Edge Impulse and then be able to use the model in Apple Watch and on hardware.
It should but the challenge is you would need to implement some of the DSP blocks to generate the input features of the NN. You can check our processing blocks implementation here if you want to move in that direction.
Thanks for the quick reply. I think I’ll look into web assembly called from swift. Might be an easier option.