I built this app over a week as a way to learn how to use TensorFlow with Mobilenets and to get some experience with Google Play (and partly as a dare). It's written in Java, as I wasn't able to find a Kotlin API for TFLite. It was built with Bazel.
I'm pretty satisfied with the actual detector's performance, although I expect I could improve the UI a little bit.
It's a weird UX, as you want it to be as simple and fast as possible (loss/notloss) but you also want it to have some sort of recognizability. I would love to hear your thoughts.