Asymmetric metric learning for knowledge transfer

TL;DR

This paper introduces asymmetric metric learning for instance-level image retrieval, enabling effective knowledge transfer from teacher to student models, and demonstrates its advantages in both asymmetric and symmetric testing scenarios.

Contribution

The paper proposes a novel asymmetric metric learning paradigm that combines knowledge transfer with metric learning, improving retrieval performance and enabling students to outperform teachers.

Findings

Plain regression is highly effective for asymmetric testing.

Asymmetric metric learning performs well in both asymmetric and symmetric testing.

Students can outperform teachers using the proposed method.

Abstract

Knowledge transfer from large teacher models to smaller student models has recently been studied for metric learning, focusing on fine-grained classification. In this work, focusing on instance-level image retrieval, we study an asymmetric testing task, where the database is represented by the teacher and queries by the student. Inspired by this task, we introduce asymmetric metric learning, a novel paradigm of using asymmetric representations at training. This acts as a simple combination of knowledge transfer with the original metric learning task. We systematically evaluate different teacher and student models, metric learning and knowledge transfer loss functions on the new asymmetric testing as well as the standard symmetric testing task, where database and queries are represented by the same model. We find that plain regression is surprisingly effective compared to more complex knowledge transfer mechanisms, working best in asymmetric testing. Interestingly, our asymmetric metric learning approach works best in symmetric testing, allowing the student to even outperform the teacher.