Learning a Metric Embedding for Face Recognition using the Multibatch Method

TL;DR

This paper introduces the Multibatch training method for face recognition, significantly improving training efficiency and accuracy on embedded systems by reducing gradient variance.

Contribution

The novel Multibatch method for similarity learning reduces gradient variance and accelerates convergence, enabling high-accuracy face recognition on minimal hardware.

Findings

Achieved 98.2% accuracy on LFW benchmark

Training time of 12 hours on a Titan X GPU

Prediction runtime of 30ms on ARM Cortex A9

Abstract

This work is motivated by the engineering task of achieving a near state-of-the-art face recognition on a minimal computing budget running on an embedded system. Our main technical contribution centers around a novel training method, called Multibatch, for similarity learning, i.e., for the task of generating an invariant "face signature" through training pairs of "same" and "not-same" face images. The Multibatch method first generates signatures for a mini-batch of $k$ face images and then constructs an unbiased estimate of the full gradient by relying on all $k^2-k$ pairs from the mini-batch. We prove that the variance of the Multibatch estimator is bounded by $O(1/k^2)$, under some mild conditions. In contrast, the standard gradient estimator that relies on random $k/2$ pairs has a variance of order $1/k$. The smaller variance of the Multibatch estimator significantly speeds up the convergence rate of stochastic gradient descent. Using the Multibatch method we train a deep convolutional neural network that achieves an accuracy of $98.2\%$ on the LFW benchmark, while its prediction runtime takes only $30$msec on a single ARM Cortex A9 core. Furthermore, the entire training process took only 12 hours on a single Titan X GPU.