I. Introduction
The realm of face verification and recognition systems is undergoing rapid evolution, marked by notable advancements. A pivotal requirement for successful differentiation among diverse face samples is the clear distinction of intra-class and inter-class similarities within the latent space. The efficacy of Deep Learning-driven face recognition systems is fundamentally shaped by the diversity and adequacy of the training and evaluation datasets. In our empirical endeavors, we harnessed a siamese network in conjunction with limited data, employing two distinct loss functions: contrastive loss and triplet loss, for the purpose of gauging the similarity among different facial images.