Gender Recognizer Based on Human Face using CNN and Bottleneck Transformer Encoder

Abstract

Several applications, such as Human-Robot interactions and offline advertising platforms, perform gender recognition based on a human face to profile their audience. These applications demand gender recognition that can operate in real-time on a low-cost or CPU device. This work proposes a gender recognizer based on a human face using a Convolutional Neural Network (CNN) and Bottleneck Transformer Encoder (BTE) that renders low parameters and operation. BTE is offered to support the primary CNN feature extractor in learning the global representation of the feature maps efficiently. This work uses three face gender datasets benchmark, namely UTKFace, Labeled Faces in the Wild (LFW), and Adience, to train and validate the proposed network. The CNN network consisted of the BTE achieves competitive accuracy compared to the state-of-the-art network. The recognizer can operate in real-time on a CPU with 150 frames per second.