An Empirical Study and Analysis of Generalized Zero-Shot Learning for Object Recognition in the Wild

TL;DR

This paper investigates generalized zero-shot learning for object recognition, highlighting challenges in balancing recognition of seen and unseen classes, and proposes a calibration method to improve performance in realistic settings.

Contribution

It introduces a calibration technique for GZSL, develops a new performance metric, and analyzes the gap between current methods and idealized benchmarks.

Findings

Naive ZSL classifiers perform poorly in GZSL settings.

Calibration improves the balance between seen and unseen class recognition.

Significant performance gap remains between current approaches and idealized benchmarks.

Abstract

Zero-shot learning (ZSL) methods have been studied in the unrealistic setting where test data are assumed to come from unseen classes only. In this paper, we advocate studying the problem of generalized zero-shot learning (GZSL) where the test data's class memberships are unconstrained. We show empirically that naively using the classifiers constructed by ZSL approaches does not perform well in the generalized setting. Motivated by this, we propose a simple but effective calibration method that can be used to balance two conflicting forces: recognizing data from seen classes versus those from unseen ones. We develop a performance metric to characterize such a trade-off and examine the utility of this metric in evaluating various ZSL approaches. Our analysis further shows that there is a large gap between the performance of existing approaches and an upper bound established via idealized semantic embeddings, suggesting that improving class semantic embeddings is vital to GZSL.