ZTE achieves facial recognition breakthrough with deep learning

ZTE has achieved a breakthrough in deep learning and convolutional neural networks (CNN), working with Intel, which could enable new cloud applications performing picture search and matching, according to a Tuesday announcement. The companies say many others researching Internet search and artificial intelligence have been trying to reach the performance and accuracy benchmark.

ZTE engineers in Nanjing City, China, used Intel’s mid-range Arria 10 FPGA (or field-programmable gate array) for a cloud inferencing application using a CNN algorithm, and reached a record with more than a thousand images processed for facial recognition per second with “theoretical high accuracy” for its custom topology. The use of the Arria 10 FPGA maintained accuracy while accelerating the raw design performance more than 10 times, according to the announcement.

Intel says the Arria 10 PFGA can perform 1.5 teraflops (TFLOPs) single precision floating-point processing, provide 1.15 million logic elements, and high-speed connectivity over a terabit per second.

“Perception, such as recognizing a face in an image, is one of the essential goals of the ZTE 5G System,” said Duan Xiangyang, vice president of ZTE Wireless Institute. “Deep learning technology is very important as it can enable such perception in mobile edge computing systems, thus making ZTE’s 5G System smarter.”

The ZTE Wireless Institute team shortened its design time by using the OpenCL programming language, and Intel says the performance of such deep learning designs could further be improved by seamlessly migrating them to its high-end Stratix 10 FPGA family.

Article Topics

biometrics  |  biometrics research  |  deep learning  |  facial recognition  |  Intel  |  neural networks  |  ZTE