Just wanted to share it here to get some feedback. The app is intended to be a joke. We build the original version in the hackathon (in which we won a 2nd place) and never got around finishing it until now. It uses a dataset of 5k images, and an inception v3 model, retrained to recognise each category. Everything is done offline and optimised to run on a mobile.
Mainly because inception v3 alone takes around 80MBs. Add the project, few things to be able to read the model effectively (offline, on-the-go) and here we go - 100MBs :)
We have ideas of trying mobilenet as the model would be 5-15MBs then.