Mining GOLD Samples for Conditional GANs

TL;DR

This paper introduces GOLD, a measure for diagnosing and improving conditional GANs, enhancing training, inference, and data selection through re-weighting, rejection sampling, and active learning, with demonstrated superior results.

Contribution

The paper proposes the GOLD measure for cGANs, enabling effective self-diagnosis and three novel applications that improve various aspects of cGAN performance.

Findings

GOLD improves cGAN training and inference quality.

Re-weighting and rejection sampling enhance image generation.

Active learning with GOLD reduces data labeling effort.

Abstract

Conditional generative adversarial networks (cGANs) have gained a considerable attention in recent years due to its class-wise controllability and superior quality for complex generation tasks. We introduce a simple yet effective approach to improving cGANs by measuring the discrepancy between the data distribution and the model distribution on given samples. The proposed measure, coined the gap of log-densities (GOLD), provides an effective self-diagnosis for cGANs while being efficienty computed from the discriminator. We propose three applications of the GOLD: example re-weighting, rejection sampling, and active learning, which improve the training, inference, and data selection of cGANs, respectively. Our experimental results demonstrate that the proposed methods outperform corresponding baselines for all three applications on different image datasets.