Apparent Age, Gender and Ethnicity Prediction

This work demonstrated ResNet-50 based Deep Learning model that can accurately classify between people of different age, gender, and racial origins. We employed Transfer Learning concept to account for relatively small dataset of over 23000 images. To be precise, the ResNet model pre-trained on the ImageNet dataset was applied to our classification task on the UTKFace dataset. Referring to Section IV, we achieve MAE of 7.7 years, 95%, and 80.4% for the first, second, and third classification problems, respectively. Further works can be done to improve all three numbers by training on a larger dataset as well as by transferring weights from a model specifically pre-trained to recognize human faces.