Deep neural networks on mobile

This repo is work in progress and currently migrated here from my several private repos

This repository is meant to be a playground for mobile deep learning. It compares and analyzes state-of-the-art mobile deep learning frameworks with focus on Android, iOS and mobile browsers.

The project is structured as a mono-repo and consists of the following sections:

• dnnmobile-app-android: contains the android app
• dnnmobile-app-ios: contains the ios app
• dnnmobile-app-web: contains the web app
• common: contains the preprocessing of datasets and training of neural network models in form of docker containers and jupyter notebooks.

Exploring mobile deep learing

Mainly, there are there populare mobile deep learning libraries Tensorflow Lite, PyTorch Mobile as well as Core ML, all having differences in their capabilities.

Library	Company	Platform	on-device inference	on-device training
Tensorflow Lite	Google	Android, iOS, embedded	✓	✓ (transfer)
PyTorch Mobile	Facebook	Android, iOS	✓	✗
Core ML	Apple	iOS	✓	✓ (transfer)

Setup

tba.

Roadmap

• add pytorch preprocessing
• add pytorch to android
• add iOS project
• add web project
• add datasets
• extend documentation
• add popular models (e.g. MobileNet)
• add benchmarking
• tba.