Figure 5
From: A new periocular dataset collected by mobile devices in unconstrained scenarios
Multi-task CNN architecture. In this model, each task has its own output and all tasks share the convolutional layers. The loss of all tasks is used to update the weights of the convolutional layers.
