Bookmark and Share

Intel Demonstration of Scalable Deep Learning-based Face Detection and Recognition with FPGAs

Richard Chuang, Global Platform Solutions Architect at Intel, delivers a product demonstration at the May 2018 Embedded Vision Summit. Specifically, Chuang demonstrates an end-to-end face detection and recognition reference solution using the OpenVINO toolkit. Four primary algorithms are running in this demo system on top of OpenVINO: face detection, landmark detection, feature extraction, and face matching. The Core i7 system supports 500 fps/core running face detection and landmark detection, the Core i5 plus Arria 10 FPGA combination system is able to deliver 120 fps on a face feature extraction algorithm, and the Xeon SP Platinum server, Optane disk and IMDT are together able to find a specific face among 20 million candidate faces in 20 ms.