Martin Cerman Martin Cerman - 1 month ago 22
Android Question

TensorFlow on Mobile Devices (Android, iOS, Windows Phone)

I am currently looking on different deep learning frameworks out there, specifically to train and deploy convolutional neural networks. The requirements are, that it can be trained on a normal PC with a GPU, but then the trained model has to be deployed on the three main mobile operating systems, namely Android, iOS, and Windows Phone.

TensorFlow caught my eye, because of its simplicity and great python interface. There is an example application for Android (https://jalammar.github.io/Supercharging-android-apps-using-tensorflow/), but I am not sure if it can be also deployed on iOS and Windows Phone? If not, can you recommend an alternative framework, which would meet these requirements? Ideally with a simple scripting interface for fast prototyping?

Thank you very much for your answers!

EDIT: Currently I'm testing Microsoft's CNTK. Building on Windows and Linux from source works perfectly, it can be extended in a "Lego blocks" fashion, and the proprietary NDL (Network Description Language) is really easy to read and learn, and provides enough freedom to build a lot of different Neural Network architectures. The execution engine is only a small part of the framework, and it can read in the NN model defined by the NDL, as well as the trained parameters. I will keep the post updated, on how the porting process to ARM processors goes.

Answer

TensorFlow currently doesn't support iOS or Windows. Here are the open github issues tracking them :