A unifying mutual information view of metric learning: cross-entropy vs. pairwise losses

Benchmark Model Rank Results
metric-learning-on-cars196ResNet-50 + Cross-Entropy#10R@1: 89.3
metric-learning-on-cub-200-2011ResNet-50 + Cross-Entropy#11R@1: 69.2
metric-learning-on-in-shopResNet-50 + Cross-Entropy#9R@1: 90.6
metric-learning-on-stanford-online-productsResNet-50 + Cross-Entropy#15R@1: 81.1