Research map / Image generation / GANs
Studying Bias in GANs through the Lens of Race
Vongani H. Maluleke, Neerja Thakkar, Tim Brooks, Ethan Weber, Trevor Darrell, Alexei A. Efros, Angjoo Kanazawa, Devin Guillory
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply truncation (1 to 0.5) when generating from FFHQ-trained StyleGAN2-ADA | StyleGAN2-ADA on FFHQ · FFHQ racial distribution | share of Black-labeled generated images: 4% to 0% | Hurt |
| apply truncation (1 to 0.5) when generating from FFHQ-trained StyleGAN2-ADA | StyleGAN2-ADA on FFHQ · FFHQ pairwise quality | % preferred: 58.7 ± 1.02 | Improved |
| increase fraction of white images in training (80B-20W to 20B-80W) | StyleGAN2-ADA on FairFace · FairFace pairwise quality | % Gen A preferred: 53.9 ± 1.02 | Improved |
| change training racial ratio on FairFace subsets | StyleGAN2-ADA · FairFace | FID: 5.60, 5.63, 5.68 | No clear effect |
| imbalanced training data (80B-20W, 50B-50W, 20B-80W) vs generated output | StyleGAN2-ADA · FairFace / FFHQ racial distribution | racial ratio preservation | No clear effect |
| increase share of a race in training data | StyleGAN2-ADA · Top-K image quality ranking | share of top-K from split: white K=10: 0.00 -> 0.71 (80B-20W vs 20B-80W) | Improved |
| label as white instead of Black (real FairFace images) | AMT pairwise evaluation · Real FairFace pairwise quality | % preferred: 55.2 ± 2.3 | Improved |
Closest papers by meaning
- Inclusive GAN: Improving Data and Minority Coverage in Generative Models (ECCV 2020 · similarity 0.89)
- Characterizing Bias in Classifiers using Generative Models (NeurIPS 2019 · similarity 0.88)
- Would Deep Generative Models Amplify Bias in Future Models? (CVPR 2024 · similarity 0.87)
- A Classification-Based Study of Covariate Shift in GAN Distributions (ICML 2018 · similarity 0.87)
- Debiasing Pretrained Generative Models by Uniformly Sampling Semantic Attributes (NeurIPS 2023 · similarity 0.87)
- On the "steerability" of generative adversarial networks (ICLR 2020 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
