Deep learning inference on android phone

Mobilenet architecture

The demo that you’ll always see at any AI/IoT expo/exhibition!
The difference is that I’m using my Android instead of a PC or a development board like Rasberry Pi. 

The application could be downloaded from Google Play (not mine!). Its name is Object Detector and Classifier (Try it! Believe me, it’s fun!). It uses Mobilenet v2 architecture model that has been already pre-trained with thousands of samples. Mobilenet v2 has a minimal architecture that fits well in mobile devices like Android phone. Precision is not really high, but very good in identifying basic daily objects.

The bottom line is that we can develop, train, and optimize any deep learning model in the cluster/cloud based on our need/requirement, then export the trained model to any edge computing devices (e.g., Android phone, Raspberry pi, NVIDIA TX1, camera processing board, etc.). Combining with other datasets in the edge device might deliver the business values that we are looking for! (e.g. plate number detection + vehicle registration database can recognize an illegal vehicle on the spot)

Leave a Reply

Your email address will not be published. Required fields are marked *