Event Details

Face recognition is an active research field where many advancements in computer vision were developed and demonstrated. Especially the advancements in network architectures and the novelty of margin penalty-based softmax losses lead to achieving notable recognition accuracies. This talk will give an insight into recent state-of-the-art face recognition solutions, with a focus on training losses.

It will build on that to discuss innovations targeting deployment in embedded domains and further use-cases that are constrained by the computational capabilities, high throughput requirements, as well as environmental constraints.  The talk will also discuss different challenges related to face recognition including the effect of wearing a facial mask on verification performance. The talk will extend this discussion to present a recent effort to reduce such effect of mask-wearing on face recognition performance.

Speaker Bio

Fadi Boutros is a research scientist at the competence center Smart Living and Biometric Technologies, Fraunhofer Institute for Computer Graphics Research (IGD).

He received his Master of Science degree in Distributed Software System from the Technical University of Darmstadt (2019). During his studies, he was working as a research assistant at the Fraunhofer IGD, where he also worked on his master thesis dealing with the ethnicity bias problem in face recognition systems. He received the CAST-Forderpreis 2019 award and the best thesis award on the 21 Darmstadt Computer Graphic Night. He is a research scientist at Fraunhofer IGD since March 2019. His main research interests lie in the fields of biometrics and machine learning. Currently, he is working on the project Embedded Biometrics which is part of the mission Next Generation Biometric Systems at the National Research Center for Applied Cybersecurity ATHENE in Darmstadt. Also, he is pursuing his Ph.D. in enabling biometrics in embedded environments at the Technical University of Darmstadt.  Since he started his Ph.D. in September 2019, he published 9 scientific publications in this field as a first author and contributed to more than 18 scientific publications.