Revisiting Training Strategies and Generalization Performance in Deep Metric Learning

By Karsten Roth, Timo Milbich, Samarth Sinha, Prateek Gupta, Björn Ommer, Joseph Paul Cohen. 2020

191 citations recorded by the source corpus.

Deep Metric Learning (DML) is arguably one of the most influential lines of research for learning visual similarities with many proposed approaches every year. Although the field benefits from the rapid progress, the divergence in training protocols, architectures, and parameter choices make an unbiased comparison difficult. To provide a consistent reference point, we revisit the most widely used DML objective functions and conduct a study of the crucial parameter choices as well as the commonly neglected mini-batch sampling process. Under consistent comparison, DML objectives show much higher saturation than indicated by literature. Further based on our analysis, we uncover a correlation between the embedding space density and compression to the generalization performance of DML models. Exploiting these insights, we propose a simple, yet effective, training regularization to reliably boost the performance of ranking-based DML models on various standard benchmark datasets. Code and a publicly accessible WandB-repo are available at this https URL.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning (2021, methodology): “[27] uses eigenvalue analysis on representations of Deep Metric Learning (DML) models.”
Class-Incremental Learning via Dual Augmentation (2021, background): “A recent paper [54] has shown that strong compressed representations can actually hurt the generalization ability in the deep metric learning setting.”
Are We Learning Yet? A Meta Review of Evaluation Failures Across Machine Learning (2021, background): “For instance, a series of recent papers have attempted to benchmark a variety of deep metric learning algorithms, controlling for aspects such as network architecture, optimizer, image augmentations, hyperparameter compute budget, etc. [41, 101, 129].”
Characterizing Generalization under Out-Of-Distribution Shifts in Deep Metric Learning (2021, methodology): “Additionally, we also evaluate results over mean average precision (mAP@1000) [54, 42], but provide respective tables and visualizations in the supplementary when the interpretation of results is not impacted.”
Towards Total Recall in Industrial Anomaly Detection (2021, background): “coverage of mini-batches for improved GAN training [49] or representation learning [41].”
Prototype Augmentation and Self-Supervision for Incremental Learning (2021, background): “4(d) shows that SSL leads to a higher feature space density πratio, and the improvement on generalization is consistent with the observation in [45].”
A Metric Learning Reality Check (2020, background): “[44], which addresses many of the same flaws that we find, and does an extensive analysis of various loss functions.”
COVID-19 Image Data Collection: Prospective Predictions Are the Future (2020, background): “, 2017], deep metric learning [Roth et al., 2020], and domain adaptation [Motiian et al.”
Why Do Better Loss Functions Lead to Less Transferable Features? (2020, background): “[56] show that objectives that lead to greater class separation generally produce higher recall, and find a similar relationship for a measure of the uniformity of the singular value spectrum of the representation space.”