Cats are no respecters of personal property, as [Joe Mattioni] learned when one of his cats, [Layla] needed a special prescription diet. Kitty didn’t care for it, and since the other cat, [Foxy]’s bowl was right there– well, you see where this is going. To keep [Layla] out of [Foxy]’s food and on the vet-approved diet, [Joe] built an automatic feeding system with feline facial recognition. As you do.
The hardware consists of a heavily modified feed bowl with a motorized lid that was originally operated by motion-detection, an old Android phone running a customized TensorFlow Lite model, and hardware to bridge them together. Bowl hardware has yet to be documented on [Joe]’s project page, aside from the hint that an Arduino (what else?) was involved, but the write up on feline facial recognition is fascinating.
See, when [Joe] started the project, there were no cat-identifying models available– but there were lots of human facial recognition models. Since humans and cats both have faces, [Joe] decided to use the MobileFaceNet model as a starting point, and just add extra training data in the form of 5000 furry feline faces. That ran into the hurdle that you can’t train a TFLite model, which MobileFaceNet is, so [Joe] reconstructed it as a Keras model using Google CoLab. Only then could the training occur, after which the modified model was translated back to TFLite for deployment on the Android phone as part of a bowl-controller app he wrote.
No one, [Joe] included, would say that this is the easiest, fastest, or possibly even most reliable solution– a cat smart enough not to show their face might sneak in after the authorized feline has their fill, taking advantage of a safety that won’t close a bowl on a kitty’s head, for example–but that’s what undeniably makes this a hack. It sounds like [Joe] had a great learning adventure putting this together, and the fact that it kept kitty on the proper diet is really just bonus.
Want to go on a learning adventure of your own? Click this finely-crafted link for all the details about this ongoing contest.
…and the winner of the 2025 alliterative headline award goes to…
Semi seriously, can we have a vote on the best headlines?
Love the alliteration too hahaha. The project sounds fun too. Never thought about cat facial recognition. Gait detection or fur pattern recognition would probably be how I’d have tried this not knowing facial recognition was possible.
Sorry, but you have misread, it’s about fecondition my dear sir.
For my cats, the rear provides good contrast, one has a big hairy Maine Coon-cat tail, other has NO tail, and the third has a long, long tail.
Aim a bit lower and try an Iris~eye recognition program? Came to me as a joke, but the more ya think about it?
Just that any cats that I’ve lived with, tend to only raise their tail IF you are also around at that time (outside of “Spraying” moments).
I accidentally made an Ir cat identification system. I was building a simple touch less tachometer using an ATTiny85, OLED display, Ir LED, and a Ir Receiver Diode. I was setting it up and testing it on a bread board while at my desk. My Calico cat just had to investigate it (as cats always need to do). When she stood over the sensor, PING! it detected her. The Ir bounced off of her white belly fur and back to the sensor. Then when my grey Tabby came to investigate, his darker fur absorbed most of the Ir light, the sensor couldn’t see him. I was rather surprised as his belly fur is actually more of a creamy brown.
Right away my mind went to the possibility of separate feeders. If they ever need different diets, I’m going to look into this further.