[askytb]

Build an app for finding lost pets with facial recognition. Recognizing cats is the easiest thing for neural nets, yet I haven't seen it used for lost pets yet.

Basic idea, if my pet is lost, I open up your app, upload a few images of my cat and set a monetary reward. On the backend your neural net learns how that specific cat looks like, then other users of the app can snap pictures of random cats on the street and the app will tell them if that animal is lost or not. If someone snaps a picure of a lost cat, they'll get connected to the owner and the reward is transfered, the app can take a small cut of it.

[la_fayette]

Neutal nets can definitely identify cats and also tell if it is a black or white cat. But identifing a specific cat might be difficult...

[sebringj]

You'll also need the Boston Dynamic dog to go get her.

[askytb]

Not really, human face recognition is quite accurate and humans all look identical in comparison to the diversity of cats and dogs

[la_fayette]

... ok what approach would you use for that? Let's say you use yolo to detect th bounding box of the cat, but how do you want to identify the cat, based on your sample images?

[askytb]

Re-train https://github.com/davidsandberg/facenet on cats?

[la_fayette]

Ok. This could work, somebody tried it for dogs (https://github.com/GuillaumeMougeot/DogFaceNet)...

For the problem to solve in the first place, one might use a gps-tracker for pets or something.