Characterizing Generalization under Out-Of-Distribution Shifts in Deep Metric Learning
This work addresses the need for more realistic evaluation in deep metric learning, which is incremental by extending existing protocols to better characterize generalization under distribution shifts.
The authors tackled the problem of evaluating deep metric learning methods under realistic out-of-distribution shifts by creating the ooDML benchmark, which systematically tests generalization across diverse and challenging splits, finding that some methods retain performance better as difficulty increases and proposing few-shot DML as an improvement.
Deep Metric Learning (DML) aims to find representations suitable for zero-shot transfer to a priori unknown test distributions. However, common evaluation protocols only test a single, fixed data split in which train and test classes are assigned randomly. More realistic evaluations should consider a broad spectrum of distribution shifts with potentially varying degree and difficulty. In this work, we systematically construct train-test splits of increasing difficulty and present the ooDML benchmark to characterize generalization under out-of-distribution shifts in DML. ooDML is designed to probe the generalization performance on much more challenging, diverse train-to-test distribution shifts. Based on our new benchmark, we conduct a thorough empirical analysis of state-of-the-art DML methods. We find that while generalization tends to consistently degrade with difficulty, some methods are better at retaining performance as the distribution shift increases. Finally, we propose few-shot DML as an efficient way to consistently improve generalization in response to unknown test shifts presented in ooDML. Code available here: https://github.com/CompVis/Characterizing_Generalization_in_DML.