Angular Margin Loss for Deep Metric Learning: From turning point to state of the art

Tóm tắt báo cáo

Metric Learning is an approach based directly on a distance metric that aims to establish similarity or dissimilarity between images, and Deep Metric Learning uses neural networks to learn discriminative features from the images effectively and then compute the metric. Various loss functions have been developed for Metric Learning. For example, the contrastive loss guides the objects from the same class to be mapped to the same point and those from different classes to be mapped to different points whose distances are larger than a margin. However, the traditional softmax loss of deep CNNs usually lacks the power of discrimination; there was a rise of angular margin methods in 2017. These improved losses share the same idea: maximizing inter-class variance and minimizing intra-class variance. We will describe the transition from contrastive approaches to current angular margin SOTA models as the advances in Metric Learning in recent years.