Hard Negative Mining for Metric Learning Based Zero-Shot Classification

TL;DR

This paper enhances zero-shot classification by introducing hard negative mining techniques into a metric learning framework, significantly improving performance on challenging datasets.

Contribution

It extends previous metric learning approaches for ZSC by proposing schemes for better negative pair selection, achieving state-of-the-art results.

Findings

Improved accuracy on three ZSC datasets

Effective negative pair generation schemes

Significant performance gains over prior methods

Abstract

Zero-Shot learning has been shown to be an efficient strategy for domain adaptation. In this context, this paper builds on the recent work of Bucher et al. [1], which proposed an approach to solve Zero-Shot classification problems (ZSC) by introducing a novel metric learning based objective function. This objective function allows to learn an optimal embedding of the attributes jointly with a measure of similarity between images and attributes. This paper extends their approach by proposing several schemes to control the generation of the negative pairs, resulting in a significant improvement of the performance and giving above state-of-the-art results on three challenging ZSC datasets.