My First AI-Powered Android App — Car Classification

What did I do?

I built an Android application to use AI for classifying cars, very similar to the Hot Dog/Not Hot Dog App, but better...

Why did I build this?

I really like cars. Not as much as Top Gear’s Jeremy Clarkson, but a decent amount. I walk down the street and often see a car I don’t recognize. Is it a Jaguar F-type or an Aston Martin Vantage? Particularly the less well known cars like recognizable cars like the Fisker Karma, the first time I saw that I thought it looked like some awesome futuristic sports car, and had to ask my friend what it was later that day. So, generally I’m curious about cars and like to know what I’m looking at, and maybe how ridiculously unaffordable it is so I can cry.

How did I build it?

I used Java for the app and Python/Flask for the server. It uses a ML API in the cloud, and is queried with a POST request. The app consists of two main activities — the welcome screen (with options to take a picture or upload an image), and the results screen where the detections are shown. The server is hosted on AWS Elastic Beanstalk and costs about $14 a month to maintain. The server uses a Torch model that runs detection and classification on the uploaded images and returns a json object with the bounding box and predictions.