Revisiting Training Strategies and Generalization Performance in Deep Metric Learning