Turning Waste into Wealth: Leveraging Low-Quality Samples for Enhancing Continuous Conditional Generative Adversarial Networks

TL;DR

This work tackles the data-sparsity challenge in continuous conditional GANs (CcGANs) by introducing Dual-NDA, a specialized negative data augmentation strategy. Dual-NDA uses two negative-sample types—label-inconsistent real images and visually poor fake images filtered by NIQE—and a modified vicinal-discriminator loss to guide CcGANs away from low-quality outputs while preserving accurate conditioning. Across UTKFace and Steering Angle datasets, Dual-NDA improves visual fidelity and label consistency, outperforming vanilla NDA and several state-of-the-art baselines, including diffusion-based methods. The method demonstrates strong performance gains in sparse-data regimes and offers practical benefits for high-quality continuous conditional image synthesis, with code available at the provided GitHub repository.

Abstract

Continuous Conditional Generative Adversarial Networks (CcGANs) enable generative modeling conditional on continuous scalar variables (termed regression labels). However, they can produce subpar fake images due to limited training data. Although Negative Data Augmentation (NDA) effectively enhances unconditional and class-conditional GANs by introducing anomalies into real training images, guiding the GANs away from low-quality outputs, its impact on CcGANs is limited, as it fails to replicate negative samples that may occur during the CcGAN sampling. We present a novel NDA approach called Dual-NDA specifically tailored for CcGANs to address this problem. Dual-NDA employs two types of negative samples: visually unrealistic images generated from a pre-trained CcGAN and label-inconsistent images created by manipulating real images' labels. Leveraging these negative samples, we introduce a novel discriminator objective alongside a modified CcGAN training algorithm. Empirical analysis on UTKFace and Steering Angle reveals that Dual-NDA consistently enhances the visual fidelity and label consistency of fake images generated by CcGANs, exhibiting a substantial performance gain over the vanilla NDA. Moreover, by applying Dual-NDA, CcGANs demonstrate a remarkable advancement beyond the capabilities of state-of-the-art conditional GANs and diffusion models, establishing a new pinnacle of performance. Our codes can be found at https://github.com/UBCDingXin/Dual-NDA.

Figures (16)

• Figure 1: Illustrative workflows for the vanilla NDA sinha2021negative and our proposed Dual-NDA.
• Figure 2: Illustration of the CCGM task and sample images from two regression datasets (UTKFace and Steering Angle).
• Figure 3: Example negative samples from NDA by transforming a realistic training image (from sinha2021negative).
• Figure 4: Some actual low-quality fake images generated from a pre-trained CcGAN at "Age=3" on UTKFace. The term "label-inconsistent" indicates that these fake images do not align with the conditioning label.
• Figure 5: Line graphs of NIQE/Label Score versus Age for the UTKFace (128$\times$128) experiment.