# Imperfect ImaGANation: Implications of GANs Exacerbating Biases on Facial Data Augmentation and Snapchat Selfie Lenses

arXiv: https://arxiv.org/abs/2001.09528 · HTML: https://mlsanity.com/paper/2001.09528/

## Key facts

- Cited by 27 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 26 Jan 2020.

## Built on by (most cited first, 27 of 27)

- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [GAUDI: A Neural Architect for Immersive 3D Scene Generation](https://mlsanity.com/paper/2207.13751/) (2022)
- [Generative Models as a Data Source for Multiview Representation Learning](https://mlsanity.com/paper/2106.05258/) (2021)
- [Intermediate Layer Optimization for Inverse Problems using Deep Generative Models](https://mlsanity.com/paper/2102.07364/) (2021)
- [A Systematic Study of Bias Amplification](https://mlsanity.com/paper/2201.11706/) (2022)
- [ITI-GEN: Inclusive Text-to-Image Generation](https://mlsanity.com/paper/2309.05569/) (2023)
- [Directional Bias Amplification](https://mlsanity.com/paper/2102.12594/) (2021)
- [Balancing Act: Distribution-Guided Debiasing in Diffusion Models](https://mlsanity.com/paper/2402.18206/) (2024)
- [Semi-Parametric Neural Image Synthesis](https://mlsanity.com/paper/2204.11824/) (2022)
- [Studying Bias in GANs through the Lens of Race](https://mlsanity.com/paper/2209.02836/) (2022)
- [Evidence for Hypodescent in Visual Semantic AI](https://mlsanity.com/paper/2205.10764/) (2022)
- [Counterfactually Measuring and Eliminating Social Bias in Vision-Language Pre-training Models](https://mlsanity.com/paper/2207.01056/) (2022)
- [Diffusion Probabilistic Fields](https://mlsanity.com/paper/2303.00165/) (2023)
- [Covariance Matrix Adaptation MAP-Annealing](https://mlsanity.com/paper/2205.10752/) (2022)
- [A Neural Tangent Kernel Perspective of GANs](https://mlsanity.com/paper/2106.05566/) (2021)
- [Model-Agnostic Gender Debiased Image Captioning](https://mlsanity.com/paper/2304.03693/) (2023)
- [Harm Amplification in Text-to-Image Models](https://mlsanity.com/paper/2402.01787/) (2024)
- [Manifold Diffusion Fields](https://mlsanity.com/paper/2305.15586/) (2023)
- [Analyzing Quality, Bias, and Performance in Text-to-Image Generative Models](https://mlsanity.com/paper/2407.00138/) (2024)
- [Adjusting for Bias with Procedural Data](https://mlsanity.com/paper/2204.01108/) (2022)
- [Interactive Storytelling for Children: A Case-study of Design and Development Considerations for Ethical Conversational AI](https://mlsanity.com/paper/2107.13076/) (2021)
- [Breaking the Spurious Causality of Conditional Generation via Fairness Intervention with Corrective Sampling](https://mlsanity.com/paper/2212.02090/) (2022)
- [Benchmarking the Fairness of Image Upsampling Methods](https://mlsanity.com/paper/2401.13555/) (2024)
- [Balanced Face Dataset: Guiding StyleGAN to Generate Labeled Synthetic Face Image Dataset for Underrepresented Group](https://mlsanity.com/paper/2308.03495/) (2023)
- [Latent Directions: A Simple Pathway to Bias Mitigation in Generative AI](https://mlsanity.com/paper/2406.06352/) (2024)
- [LabellessFace: Fair Metric Learning for Face Recognition without Attribute Labels](https://mlsanity.com/paper/2409.09274/) (2024)
- [Training-Free Debiasing of Diffusion Models via CLIP-Guided Denoising Optimization](https://mlsanity.com/paper/2607.00817/) (2026)

## Abstract (arXiv)

In this paper, we show that popular Generative Adversarial Networks (GANs) exacerbate biases along the axes of gender and skin tone when given a skewed distribution of face-shots. While practitioners celebrate synthetic data generation using GANs as an economical way to augment data for training data-hungry machine learning models, it is unclear whether they recognize the perils of such techniques when applied to real world datasets biased along latent dimensions. Specifically, we show that (1) traditional GANs further skew the distribution of a dataset consisting of engineering faculty headshots, generating minority modes less often and of worse quality and (2) image-to-image translation (conditional) GANs also exacerbate biases by lightening skin color of non-white faces and transforming female facial features to be masculine when generating faces of engineering professors. Thus, our study is meant to serve as a cautionary tale.