Vec2UAge: Enhancing Underage Age Estimation Performance through Facial Embeddings

Congratulations to Felix Anda and co-authors on the paper Vec2UAge at Forensic Science International: Digital Investigation.

AI-generated summary of the contribution: The Vec2UAge model addresses underage age estimation challenges through facial embeddings, leveraging FaceNet embeddings and achieving a mean absolute error rate as low as 2.36 years. This research has significant implications for digital forensic investigations, particularly in identifying victims, suspects, and missing children, and reducing investigators’ exposure to harmful material. The study’s use of data augmentation techniques and evaluation of various optimizers provides valuable insights into the effects of random initializations and learning rates on model performance.

Read the publication.