The largest bank of Russia, Sber, has developed an animal recognition system that can be used to search for missing pets, TASS reports with reference to the bank’s press service.
It is announced that Sber has taught the neural network to distinguish animals that are very similar to each other by certain features of their muzzles, which allows people to find differences between separate animals of the same species as accurately as possible. It is specified that at the moment the quality of comparing two faces of pets is 94 per cent for dogs and 84 per cent for cats.
The model is based on the upgraded CosFace algorithm used for recognizing human faces. According to the Director of the “Protective Products and Services” Division of Sberbank, Denis Kuzmin, now the algorithm is integrated into the pet insurance programme, and in the future, this development can also be used to search for missing animals.
It is noted that earlier Sber studied other algorithms for recognizing animals. For example, the identification of pets by a nose print, the uniqueness of which was equated to a human fingerprint. Unfortunately, such a system turned out to be unsuitable for wide use due to the lack of a distinct nose pattern in the representatives of the cat family.