“…Recently, there has been a growing number of approaches that bridge existing modality gaps in order to perform heterogeneous face recognition. These approaches focused on various scenarios of heterogeneous face recognition such as infrared-to-visible [30,16,56], thermal-to-visible [51,48,76,21,27], and sketch-to-visible [10,39] [60]. Fundamentally, each approach seeks to either find a common latent subspace in which corresponding faces from each domain are "close" in terms of some distance and non-corresponding faces are "far."…”