Two tutorials will be organized at FG’24. These will be held on May 27, 2024, in the same venue as the FG 2024 main conference (the exact program will be announced closer to the conference).
Generation of Synthetic Data for Remote Verification System
Tutors: Juan Tapia (Hochschule Darmstadt, Germany), Naser Damer (Fraunhofer IGD, Germany), Juan M. Espín López (Facephi, Spain), Mario Nieto-Hidalgo (Facephi, Spain)
Abstract: Synthetic content creation has advanced in recent years with new deep-learning developments. Newer architectures like Generative Adversarial Networks (GAN) and diffusion models can now produce realistic face images with perceptually pleasing geometry and surface texture that challenge human perception. However, creating realistic images in other domains, such as remote verification systems, is still an open challenge.
Indeed, because of private concerns, access to bona fide ID cards and Selfies to train a robust fake-ID cards detection system is minimal. One solution is generating synthetic ID card images and Selfies in order to create faces, text and textures (colour and design) as a whole. This tutorial is an application complement for workshop such as “Synthetic Data for Face and Gesture Analysis”.
The tutorial will feature three presentations. The first will be focused on explaining the current challenge for biometrics companies in the remote verification system. The second talk will present a scientific point of view of how the researchers have explored this challenge, showing and summarizing the main work performed in the bona fide and attack ID-Cards scenarios related to databases, algorithms, and suggestions. The third talk also will present a scientific point, focusing on the Liveness Face Presentation Attack Detection to cover a whole remote verification system.
This tutorial is sponsored by Facephi company.
Bias Assessment, Explanation, and Mitigation in Deep Face Recognition
Tutors: Andrea Atzori (University of Cagliari, Italy), Lucia Cascone (University of Salerno, Italy), Mirko Marras (University of Cagliari, Italy), Fabio Narducci (University of Salerno, Italy)
Abstract: This tutorial provides an interdisciplinary overview about the topic of bias in the context of face recognition systems. We begin by delving into the foundational principles that characterize the adoption of these systems, drawing upon insights from academic literature and real-world examples that highlight the imperative need of addressing bias issues. Our exploration then extends to presenting a taxonomy that encompasses various dimensions of bias, including those pertaining to social, ethical, legal, and regulatory points of views. Subsequently, we introduce recent methodologies developed for assessing and explaining bias, and relevant mitigation techniques specific to face recognition systems. We alternate between lecture slides and hands-on sessions to allow participants to gain practical experience through implementations using open-source tools and public datasets. The final segments shift focus towards analyzing emerging challenges and future trajectories, emphasizing the need of a responsible approach in the development of face recognition systems.