In this video I show how a humble Raspberry Pi can be turned into an artificial intelligence that recognises objects you put in front of it and tells you what it sees, using Google’s open source TensorFlow machine learning library and the Inception data model trained on ImageNet data. No Internet connection required as everything lives on the Raspberry Pi itself!
This uses the rather wonderful Raspberry Pi version of TensorFlow developed by Sam Abrahams, which is available from GitHub at:
By default TensorFlow’s image classifier script doesn’t speak out loud the name of the object it has recognised, and the whole thing is quite fiddly to get up and running. So a thousand thanks to Libby Miller for documenting all the necessary steps on her blog, and to Dan Brickley for the tweaked image classifier:
If you tried this out and found it interesting, I’d love to hear what you use TensorFlow for – leave me a comment below!