Using 4,675 fully labeled bear faces in DSLR photographs, taken from bear research and viewing sites in Brooks River, Alabama, and Knight Inlet, they randomly divided the images into sets training and test data. Once formed from 3,740 bear faces, deep learning got to work “unattended,” Dr Clapham said, to see how well he could spot the differences between known bears. from 935 photographs.
First, the deep learning algorithm finds the bear’s face using distinctive landmarks such as the eyes, tip of the nose, ears, and forehead. Then, the application rotates the face to extract, encode and classify the facial features.
The system identified bears with an 84% accuracy rate, correctly distinguishing known bears such as Lucky, Toffee, Flora, and Steve.
But how does he actually differentiate these bears? Before the era of deep learning, “we tried to imagine how humans perceive faces and how we distinguish individuals,” said Alexander Loos, research engineer at the Fraunhofer Institute for Digital Media Technology , in Germany, who was not involved in the study but has worked with Dr Clapham in the past. Programmers would manually enter face descriptors into a computer.
But with deep learning, programmers capture images into a neural network that determines how best to identify individuals. “The network itself extracts the functionality,” Dr. Loos said, which is a huge plus.
He also cautioned: “It’s basically a black box. You don’t know what it’s doing ”and if the data set being examined is unintentionally biased, some errors may appear.