Optimizing Rank-Based Metrics With Blackbox Differentiation

Michal Rolinek, Vit Musil, Anselm Paulus, Marin Vlastelica, Claudio Michaelis, Georg Martius; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 7620-7630


Rank-based metrics are some of the most widely used criteria for performance evaluation of computer vision models. Despite years of effort, direct optimization for these metrics remains a challenge due to their non-differentiable and non-decomposable nature. We present an efficient, theoretically sound, and general method for differentiating rank-based metrics with mini-batch gradient descent. In addition, we address optimization instability and sparsity of the supervision signal that both arise from using rank-based metrics as optimization targets. Resulting losses based on recall and Average Precision are applied to image retrieval and object detection tasks. We obtain performance that is competitive with state-of-the-art on standard image retrieval datasets and consistently improve performance of near state-of-the-art object detectors.

Related Material

[pdf] [supp] [arXiv]
author = {Rolinek, Michal and Musil, Vit and Paulus, Anselm and Vlastelica, Marin and Michaelis, Claudio and Martius, Georg},
title = {Optimizing Rank-Based Metrics With Blackbox Differentiation},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}