Using Firebase ML, TensorFlow, and Flutter I have put together a new image recognition app.
I began with teaching the ML Kit the difference between Chickens and Squirrels, it turned out to be very simple, just as my previous version using React and Clarifai had been, however at the moment AutoML is very new and doesn’t yet have Flutter extensions ( could change any day now ), so I changed tack and tried out one of their standard libraries instead.
Firebase ML will create an on-device API for you, it was a surprise to me, you just grab a package from your Firebase project and just import that into your App.
This is a big improvement on the clarifai app in that being local it happily works offline, there’s no usage cap to worry about, just install it on your phone and it will work anywhere anytime.
The initial Squirrel Chicken testing.
And a screenshot from the App. Pretty tricky for the AI this one, the generalised version can’t tell that it’s a Red Panda, I can forgive that, many people wouldn’t know. But I think it’s hedging its bets a bit by saying 82% dog and 64% cat – obviously can’t be both, probably it’s saying it has 82% dog characteristics and 64% cat characteristics. Think I would discard any results with less than 90% certainty but I need to test that.
