Optimizing Rank-based Metrics with Blackbox Differentiation
Abstract
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. The code is available at https://github.com/martius-lab/blackbox-backprop
Cite
@article{arxiv.1912.03500,
title = {Optimizing Rank-based Metrics with Blackbox Differentiation},
author = {Michal Rolínek and Vít Musil and Anselm Paulus and Marin Vlastelica and Claudio Michaelis and Georg Martius},
journal= {arXiv preprint arXiv:1912.03500},
year = {2024}
}
Comments
CVPR 2020 conference paper (oral). The first two authors contributed equally