Deploying Scalable Face Recognition Pipeline Using Distributed Microservices

Over the past few decades, deep learning has been a remarkable technique in solving numerous problems in application domains, such as facial detection and recognition. With the existence of facial datasets, neural network models, and deep learning frameworks, one can develop and train deep neural network models on a monolithic (single host) system with ease. However, at the deployment stage, this deployment method is no longer feasible due to the increasing volume of the given data. To address this problem, we propose a scalable architecture for deploying a deep learning-based facial recognition system using distributed microservices. In this work, we use Docker as the container platform, although practically one may use any platform with the same capabilities. By encapsulating the whole system to Docker images, we can deploy deep learning applications into containers and computational intensive containers are distributed throughout the cluster. With this horizontally scalable cluster, the system can process virtually any size of data. Experimental result suggests that the proposed method is a feasible solution, as there is no noticeable computational overhead when deploying deep learning-based facial recognition system when using container-based virtualization.