Age & Gender Classifier for Edge Computing

Abstract

Deep learning models are known for being large and computationally expensive. It is a challenge to fit these models into edge devices which usually have frugal memory. A striking feature about neural networks is their enormous size. Embedded devices in edge computing scenario typically cannot handle large neural networks. We present two models of Convolutional Neural Networks (VGG16 type), with a modification in the prediction layer, potentially suitable for edge computing devices. Both networks have been designed for the classification of a human face by gender and age. The first one (VGG16/10) considers gender and age as two related characteristics and the final neurons have been conceived to hold these aspects together at the same time. The second one (VGG16/8+1) in the prediction layer has got a neuron for the prediction of gender and another eight for the prediction of age (according to the Adience benchmark). Such networks have been conceived to provide simultaneously information on both the gender and age of the person identified in the image, without the need to build two dedicated networks. The aim is to develop a solution that can be suitable in edge-computing scenario.