Comparison of Age, Gender and Ethnicity Prediction Using Traditional CNN and Transfer Learning

Analysis of facial features has attracted a lot of attention since the development of deep learning. There are numerous applications of demographic features such as age, gender and ethnicity but this topic has not been thoroughly explored by researchers especially when it comes to prediction of ethnicity. This paper aims to show the comparison between a custom CNN model and a pre-trained model. This article uses two models, a pre-trained MobileNet model[1] and a customized CNN model make up the first and second, respectively. Prediction accuracy and MAE are used to compare the results of each model that was put into practice. While accuracy is utilized for gender and ethnicity, MAE is employed as a performance metric for age. The models are analyzed using the UTKFace dataset[2].