Deep Learning for Age Estimation Using EfficientNet

The human face constitutes various biometric features that could be used to estimate important details from humans, such as age. The automation of age estimation has been further limited by variations in facial landmarks and appearances, together with the lack of enormous databases. These have also limited the efficiencies of conventional approaches such as the handcrafted method for adequate age estimation. More recently, Convolutional Neural Network (CNN) methods have been applied to age estimation and image classification with recorded improvements. In this work, we utilise the CNN-based EfficientNet architecture for age estimation, which, so far, has not been employed in any current study to the best of our knowledge. This research focused on applying the EfficientNet architecture to classify an individual's age in the appropriate age group using the UTKface and Adience datasets. Seven EfficientNet variants (B0-B6) were presented herein, which were fine-tuned and used to evaluate age classification efficiency. Experimentation showed that the EfficientNet-B4 variant had the best performance on both datasets with accuracy of 73.5% and 81.1% on UTKFace and Adience, respectively. The models showed a promising pathway in solving problems related to learning global features, reducing training time and computational resources.