In watching a few of your videos I thought of a project idea for your channel.
My use case is that I use photos on my phone but don’t really have a great way to find photos by the categories I want use and further tune.
So the project would pull all photos from all devices (cams, phones, iphotos) and use the google coral tpu to place labels on them so they are in the categories. This would be pushed to a maria sql database.
Meta data store in the sql database would be filename, md5hash of file, date, location, source (phone1, phone2, camera, directory)
The categories for my use case would be the following: people, documents, electronics, non-people, computer-screen-shots, login-screens, pets, non-pets.
A benefit of the TPU is that is a picture was not classified correctly a flag could be set so that that it is marked for retraining of the model.