_____ _ _
| _ |___ _ _| |_|___ ___
| | | | | | | | -_|
|__|__|_|_|_ |_|_|_|_|___|
|___|
I made a short sample project which demonstrates the capabilities of the new CoreML framework which Apple will release with iOS 11 later this month.
As an inspiration I used the Hotdog | Not Hotdog App idea from the latest Silicon Valley season.
The sample is pretty easy, most of the code is setting up the camera and converting the frames to the write size & format.
This Demo is written in Objective-C.
Keep in mind that you need the XCode 9 beta and iOS 11 beta to try the Demo yourself.
I used the model from the Apple CoreML Sample Model page. But you could easily convert the inception model by yourself with the python tool Apple provides. It's basically 2 lines of python code and the converting tool supports at the moment Caffe & Keras for neuronal networks. Keras can also be used inside of Tensorflow, the popular Deep Learning Library from Google.
I used the Inception model because it is with > 300 nodes not so small and can classify 1000 objects which one of them is luckily a hotdog. You can see a list of all the labels which inception can classify here. You can find a python script which generates the model and fills it with the correct weights here.
