Robust Conditional GAN from Uncertainty-Aware Pairwise Comparisons

TL;DR

This paper introduces PC-GAN, a robust conditional GAN that uses pairwise comparisons and Bayesian uncertainty to perform image attribute editing with less supervision and noise resistance.

Contribution

It presents a novel weakly supervised GAN framework leveraging pairwise comparisons and Bayesian methods for robust image attribute editing.

Findings

Performs comparably to fully-supervised methods

Outperforms unsupervised baselines

Demonstrates noise resistance in attribute estimation

Abstract

Conditional generative adversarial networks have shown exceptional generation performance over the past few years. However, they require large numbers of annotations. To address this problem, we propose a novel generative adversarial network utilizing weak supervision in the form of pairwise comparisons (PC-GAN) for image attribute editing. In the light of Bayesian uncertainty estimation and noise-tolerant adversarial training, PC-GAN can estimate attribute rating efficiently and demonstrate robust performance in noise resistance. Through extensive experiments, we show both qualitatively and quantitatively that PC-GAN performs comparably with fully-supervised methods and outperforms unsupervised baselines.