Review: Google AIY Vision Kit

As part of our recent sprint on artificial intelligence, we decided to get our hands dirty with the latest version of Google’s AIY Vision Kit (v1.1).

Essentially, this is a cardboard cased intelligent camera based around a Raspberry Pi and is supplied with a few demos showing the potential of image based machine learning. You can also develop and run your own code on it.

The unboxing experience is good with all parts neatly packaged and labelled.

It has everything you need to run the demo that loads on start up— except for a USB power supply (and be warned you cannot run it from the USB of a computer).

If you want to run any of the other supplied demos, however, it is pretty much essential to have access to another computer and/or an Android phone, a HDMI monitor, USB keyboard, USB mouse, HDMI cable and a mini-HDMI adaptor.

The kit is coupled with a comprehensive step-by-step online guide which helpfully anticipates most of the problems you might encounter while building and running the in-built demos. We came slightly unstuck running one of the later demos as the camera preview window on the Pi completely obscured the terminal and didn’t seem to be resizable. We ended using SSH to get into the Pi from another computer to run the rest of the demos.

One of us did a out-of-the-cardboard-case pre-build a few days prior to running the workshop with the rest of the team and it went together easily enough in about 40 minutes. It was a similar build time on workshop day itself.

Playing around with the supplied demos kept us occupied for around another hour. There were demos covering object and face recognition as well as emotion detection.

What was interesting was how different team members responded to the finished camera. Those who’d had a hand in assembling it and actively running the demos felt it had demystified the technology and found it fun. Others, who’d only dipped in and out of the session, regarded the idea of a intelligent camera rather uncomfortable and alienating.

One of our team was sufficiently inspired by the Kit to consider how we’d build a bee recognition machine learning application to help with our Great British Bee Count.

Overall, this is very much a kit for techies to have a bit of a play. As Hollywood movies often demonstrate, typing obscure commands to run demos is not really a spectator sport and you’ll have to supply your own context if using it for educational purposes. Those with greater ambitions might find some of the hardware limitations frustrating. For example, we noticed the camera often struggled with focussing on smaller objects that were closer to it.

[embed]Connect with the Raven team on Telegram