Article
Vec2UAge: Enhancing Underage Age Estimation Performance through Facial Embeddings
Contribution Summary
Vec2UAge is a regression-based model for estimating the age of underage individuals from facial embeddings. The model is trained on the VisAGe and Selfie-FV datasets, which are two of the most popular datasets for facial age estimation. The authors evaluate the impact of random initializations, optimizers, and learning rates on the model's performance and find that Stochastic Weight Averaging (SWA) is the most effective optimizer. The model achieves a mean absolute error rate of 2.36 years, outperforming previous state-of-the-art models. The authors also evaluate the effect of data augmentation techniques on the model's performance and find that they improve the accuracy of the model. The Vec2UAge model is a significant contribution to the field of facial age estimation, particularly for underage individuals.
Keywords: Facial Age Estimation; Underage Age Estimation; Deep Learning; FaceNet Embeddings; Stochastic Weight Averaging; Data Augmentation; Machine Learning; Digital Forensics
Abstract
Automated facial age estimation has drawn increasing attention in recent years. Several applications relevant to digital forensic investigations include the identification of victims, suspects and missing children, and the decrease of investigators' exposure to psychologically impacting material. Nevertheless, due to the lack of accurately labelled age datasets, particularly for the underage age range, sufficient performance accuracy remains a major challenge in the field of age estimation. To address the problem, a novel regression-based model was created, Vec2UAge. FaceNet embeddings were extracted and used as feature vectors to train the model from the VisAGe and Selfie-FV datasets. A balanced, unbiased dataset was created for testing and validation. Data augmentation techniques were evaluated to further be used to expand the training dataset. The learning rate (lr) is one of the most important hyper-parameters for deep neural networks; a cyclic learning rate approach was used to find the optimal initial value for lr and the performance was evaluated. The distribution of model performance was presented per optimiser and one of the winning models with a Stochastic Weight Averaging (SWA) optimised training run reached a mean absolute error rate as low as 2.36 years. Additionally, the time of convergence using SWA was significantly faster than other optimisers evaluated, i.e., ADAGRAD, ADAM and Stochastic Gradient Descent. The evaluation model metric is presented in a form of a distribution rather than a single value, giving more insights into the effects of the random initialisations, optimisers and the learning rate on the outcome.
BibTeX
@article{anda2021Vec2UAge,
author={Anda, Felix and Dixon, Edward and Bou-Harb, Elias, Le-Khac, Nhien-An and Scanlon, Mark},
title="{Vec2UAge: Enhancing Underage Age Estimation Performance through Facial Embeddings}",
journal="{Forensic Science International: Digital Investigation}",
year=2021,
month=03,
publisher={Elsevier},
abstract={Automated facial age estimation has drawn increasing attention in recent years. Several applications relevant to digital forensic investigations include the identification of victims, suspects and missing children, and the decrease of investigators' exposure to psychologically impacting material. Nevertheless, due to the lack of accurately labelled age datasets, particularly for the underage age range, sufficient performance accuracy remains a major challenge in the field of age estimation. To address the problem, a novel regression-based model was created, Vec2UAge. FaceNet embeddings were extracted and used as feature vectors to train the model from the VisAGe and Selfie-FV datasets. A balanced, unbiased dataset was created for testing and validation. Data augmentation techniques were evaluated to further be used to expand the training dataset. The learning rate (lr) is one of the most important hyper-parameters for deep neural networks; a cyclic learning rate approach was used to find the optimal initial value for lr and the performance was evaluated. The distribution of model performance was presented per optimiser and one of the winning models with a Stochastic Weight Averaging (SWA) optimised training run reached a mean absolute error rate as low as 2.36 years. Additionally, the time of convergence using SWA was significantly faster than other optimisers evaluated, i.e., ADAGRAD, ADAM and Stochastic Gradient Descent. The evaluation model metric is presented in a form of a distribution rather than a single value, giving more insights into the effects of the random initialisations, optimisers and the learning rate on the outcome.},
doi={10.1016/j.fsidi.2021.301119},
}