Adapting gender and age recognition system for mobile platforms

Abstract

Human gender and age recognition is an emerging application for intelligent video analysis. However, offline pretrained recognition models often show degraded performance in a specific application scenario. To alleviate this issue, this paper presents a client-server system design adapting gender and age recognition models for mobile platforms. Specifically, the client program on Android smart phones streams face images to a cloud computing service where the recognition models based on convolutional neural networks are adapted leveraging the face correspondences in successive frames as weak supervision. The prototype system demonstrates the proposed design effectively reduces estimation variances and enhances user experiences.