Department of Computer Engineering
Representation Learning for Improved Facial Analysis
Mehmet Oğuz Sakaoğlu
(Supervisor: Asst. Prof. Dr. Hamdi Dibeklioğlu)
Computer Engineering Department
Facial analysis is becoming more important than ever in the era of deep learning. Understanding human faces is a necessary asset for many real world applications such as face recognition and verification, behavioral and emotional analysis, age estimation and many more. However, the power of deep learning models becomes more imminent with large volumes of data, and obtaining large amount of facial images/videos under controlled settings is hard and unrealistic for many approaches. Current domain trend of facial analysis shifts swiftly towards in-the-wild images. In this research, we aim to approach facial analysis on in-the-wild images employing generative network based representation learning. In this particular study, we propose to reconstruct dense 3D human faces from a single in-the-wild color image estimating/extracting meaningful parameters such as shape(identity), pose or expression. Our research goal is to use such a rich latent representation for robust and reliable facial analysis tasks such as emotional expression recognition.
DATE: 25 November 2020, Wednesday @ 13:15