Realigned Softmax Warping for Deep Metric Learning

Michael G. DeMoor, John J. Prevost

Deep Metric Learning (DML) loss functions traditionally aim to control the forces of separability and compactness within an embedding space so that the same class data points are p…