Summary
The invention relates to a method of using a deep neural network (DNN) accelerator for implementing traditional computer vision algorithms. This includes receiving a sequence of traditional computer vision algorithm operations and mapping them to a set of mathematically equivalent neural network primitives. A neural network accelerator processes input data, including images, according to the generated neural network that represents the traditional computer vision algorithm.
The Hearing Officer applied the Aerotel/Macrossan approach and found the invention made a technical contribution over and above its implementation as a computer program, so it was not excluded as a computer program as such. The claimed combination of generating a neural network representing the traditional computer vision algorithm and processing images via that network, using a hardware neural network accelerator, was found to produce a technical effect of processing image data more efficiently and to be a technical solution to a technical problem.
Full decisionO/296/21 110Kb