A Deep Adversarial Framework for Visually Explainable Periocular Recognition

Joao Brito, Hugo Proenca; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021, pp. 1453-1461

The ability to portray the reasoning behind a decision has been at the core of major research efforts. It serves not only to increase trust amongst the stakeholders of the automated agent, but also to potentially improve the entire system as a whole. In this work, we present our efforts towards a visually explainable periocular recognition framework, with a simple, yet effective solution that automatically provides a visual representation of the features in each region that sustained an impostor pairwise comparison. Based in our quantitative and qualitative experiments, the results validate the proposed goals and reiterate the notion that exploitability should be strongly considered when designing ML algorithms.