For Nvidia Systems Software Engineer Robert Bond, however, there was only ever one possible solution: training a deep learning neural network to turn on his garden sprinklers whenever it recognized a cat in his front yard.
“I mounted a camera on the garage that looks out over the garden,” Bond tells Digital Trends. “The camera is set to send images to an [Nvidia Jetson TX1] whenever something changes. The [TX1] then uses deep learning to decide whether the thing that moved is a cat or something else. If it’s a cat then it activates the sprinkler system.”
Bond says that he started training his neural network with a data set of around 1.2 million pictures gathered from the internet. These included plenty of pictures of cats, but most of them turned out to be cutesy close-ups portraits, rather than the kind of pictures he was looking for. “It was almost worthless for what I needed to do, because I was hunting for pictures of cats slinking around in my bushes,” he explains.
Bond notes that at this point his neural network was only successfully classifying around 30 percent of the cats that came into his garden. He decided to modify the neural network’s training examples by adding pictures he had taken himself. Recognition scores immediately jumped to around 90 percent.
The system hasn’t been perfect, however. For one thing, during the training process shadows were regularly classified as cats because they represented dark shapes moving across Bond’s lawn. Even Bond wasn’t immune from being classified as a feline. “I went to mow my lawn and the system somehow decided that the combination of me and my lawnmower counted as a cat,” he says. “I wound up getting pretty thoroughly sprayed.”
But for the most part the deep learning network has been a resounding success — and Robert Bond’s garden is now virtually cat free.
It’s simple when you know how, right?
