Multi Proxy Anchor Loss and Effectiveness of Deep Metric Learning Performance Metrics

Deep metric learning (DML) learns the mapping, which maps into embedding space in which similar data is near and dissimilar data is far. In this paper, we propose the new proxy-based loss and the new DML performance metric. This study contributes two following: (1) we propose multi-proxies anchor (MPA) loss, and we show the effectiveness of the multi-proxies approach on proxy-based loss. (2) we establish the good stability and flexible normalized discounted cumulative gain (nDCG@k) metric as the effective DML performance metric. Finally, we demonstrate MPA loss's effectiveness, and MPA loss achieves new state-of-the-art performance on two datasets for fine-grained images.