[ClassyJacket]

Fascinating. I love seeing practical applications of computer vision. I want to understand how image embeddings work.

I wonder how practical it would be to add specific multidimensional similarity metrics the user could adjust. For example, if they didn't care whether it was a photo or drawing, but still had to be the same type of object. Or if they didn't care about the art style. Or if it had to be an SUV driving thru a river, but they don't care what colour SUV.