Maintaining Difficulty: A Margin Scheduler for Triplet Loss in Siamese Networks Training

TL;DR

This paper introduces a margin scheduler for triplet loss in Siamese networks that dynamically adjusts the margin to maintain training difficulty, leading to improved verification performance.

Contribution

The authors propose a novel margin scheduler that adapts the margin parameter during training based on triplet difficulty, enhancing learning effectiveness.

Findings

The margin often exceeds the fixed value during training, indicating the need for dynamic adjustment.

The proposed scheduler outperforms fixed and monotonically increasing margin schemes.

Experimental results show consistent performance gains across four datasets.

Abstract

The Triplet Margin Ranking Loss is one of the most widely used loss functions in Siamese Networks for solving Distance Metric Learning (DML) problems. This loss function depends on a margin parameter {\mu}, which defines the minimum distance that should separate positive and negative pairs during training. In this work, we show that, during training, the effective margin of many triplets often exceeds the predefined value of {\mu}, provided that a sufficient number of triplets violating this margin is observed. This behavior indicates that fixing the margin throughout training may limit the learning process. Based on this observation, we propose a margin scheduler that adjusts the value of {\mu} according to the proportion of easy triplets observed at each epoch, with the goal of maintaining training difficulty over time. We show that the proposed strategy leads to improved performance when compared to both a constant margin and a monotonically increasing margin scheme. Experimental results on four different datasets show consistent gains in verification performance.