Deep learning based face recognition system

Today, face recognition systems are able to achieve the state of the art performance using machine learning. Computer Division, BARC has designed and developed in-house face recognition system using deep learning. It recognizes more than 20,000 registered persons in an unconstrained environment. In this article, we discuss a life cycle of machine learning project with the help of the developed system. We discuss selection of performance metrics, datasets, and neural networks for achieving satisfactory performance. We talk about model deployment using edge computing. We emphasize the importance of data-centric AI development and model monitoring for maintaining an acceptable level of performance. (author)