Fast Search of Face Recognition Model for a Mobile Device Based on Neural Architecture Comparator

This paper addresses the face recognition task for offline mobile applications. Using AutoML techniques, a novel technological framework is proposed to develop a fast neural network-based facial feature extractor for a concrete device. First, the Once-for-All SuperNet is trained on a large facial dataset. Each device is characterized by its lookup table, which contains the running times of inference in each layer of the SuperNet. An evolutionary search is then used to select the most accurate subnetwork within a limit on the maximum expected latency. It is proposed to train a neural architecture comparator using Gradient Boosted Trees to choose the better subnetwork in this search. Experimental face verification and recognition results demonstrate the robustness of the novel method to various facial region positions. The best model achieves an identification accuracy of 98.7% for the LFW dataset in less than 5 ms on the Qualcomm Snapdragon 865 GPU.