Bridging the Projection Gap: Overcoming Projection Bias Through Parameterized Distance Learning

TL;DR

This paper proposes a novel Mahalanobis distance-based approach within a VAEGAN framework to reduce projection bias in generalized zero-shot learning, significantly improving recognition accuracy for unseen classes.

Contribution

It introduces a parameterized Mahalanobis distance metric and a dual-branch VAEGAN architecture to mitigate projection bias in GZSL, enhancing robustness during inference.

Findings

Outperforms state-of-the-art GZSL methods by up to 3.5% on the harmonic mean metric.

Effectively reduces projection bias through a new loss function and distance learning.

Demonstrates robustness across four benchmark datasets.

Abstract

Generalized zero-shot learning (GZSL) aims to recognize samples from both seen and unseen classes using only seen class samples for training. However, GZSL methods are prone to bias towards seen classes during inference due to the projection function being learned from seen classes. Most methods focus on learning an accurate projection, but bias in the projection is inevitable. We address this projection bias by proposing to learn a parameterized Mahalanobis distance metric for robust inference. Our key insight is that the distance computation during inference is critical, even with a biased projection. We make two main contributions - (1) We extend the VAEGAN (Variational Autoencoder \& Generative Adversarial Networks) architecture with two branches to separately output the projection of samples from seen and unseen classes, enabling more robust distance learning. (2) We introduce a novel loss function to optimize the Mahalanobis distance representation and reduce projection bias. Extensive experiments on four datasets show that our approach outperforms state-of-the-art GZSL techniques with improvements of up to 3.5 \% on the harmonic mean metric.