Try this:

It's an object detection model that also gives you a representation of the thing it detected that's supposed to be generally useful for new downstream tasks. 10 images isn't a lot to work with, but you could definitely try fine tuning with that data and see what happens. Start with the pre-trained weights available in the git repo.

About

Geeks Mental is a community that publishes articles and tutorials about Web, Android, Data Science, new techniques and Linux security.