When you set up the AWS DeepLens, you need to set up a project:
When I first set this up, I tried the object detection project, but it detects objects your device probably won’t see (like a dining room table vs an airplane), so I think the face detection is a better demo.
Click through the next screen, and then select to deploy to the device:
This gives you a cryptic (contradictory) error, which you can feel free to ignore:
Face-detection contains only a Lambda function. Do you want to deploy it? This project does not contain a Lambda function. Lambda functions run an instances of the model.
Once the project deploys, you can “subscribe” to output from it, using an identifer AWS gives you that looks like this:
Once you subscribe to this stream, you’ll get a series of JSONs, that appear to represent the probability that there is a face in the image (I get 0 when I point it away, and .3-.8 when pointing at myself)