ArcFace Loss Implementation for Tensorflow Similarity

Metric learning is a collection of machine learning approaches for learning to assess similarities and differences between data points from various modalities. It has a wide range of applications, including information retrieval, semantic search, and recommendation systems. One of the Metric Loss function, ArcFace loss determines the angle between the current feature and the target weight using the arc-cosine function since the dot product between the CNN feature and the last fully connected layer after feature and weight normalization matches the cosine distance.