Using a standard live view on With a generic library, the dogs are sometimes identified as cats. I suspect I can get accurate enough using a 2 stage recognition pipeline, but was thinking the RFID was going to be more reliable… Camera getting muddy for instance will break ML.