Biometric Authentication with a Timeless Learner


Many machine learning models have been developed under BATL and deployed in different Odin - Thor evaluation events. Several of these models achieved exceptional performance on presentation attack detection on different modalities. The details of these models can be found in the publications list. One universal model that was applied to all modalities, fingerprint, iris, and face, is shown above, which employs a fully convolutional neural networks architecture with fine- and coarse-grain supervision. Detailed evaluation results of this model in different attack categories are shown in the table below. Full details can be found in this paper.