A Vision Transformer Architecture for Open Set Recognition