Despite significant progress made over the past twenty five years,
unconstrained face verification remains a challenging problem. This paper
proposes an approach that couples a deep CNN-based approach with a
low-dimensional discriminative embedding learned using triplet probability
con