Intersectional Fairness-Driven Data Augmentation for Robust Semantic Segmentation

Inspired by the FACET benchmark (Gustafson et al., 2023), which exposes performance disparities across demographic attributes and their intersections (such as skin tone and hair color), this idea proposes a targeted data augmentation and synthesis framework. By analyzing where segmentation models underperform for specific demographic intersections, the system generates or augments training data (using GANs or diffusion models) to fill these gaps. Unlike generic augmentation or fairness-aware evaluation, this approach directly addresses intersectional fairness by iteratively augmenting the dataset to reduce measured disparities, using feedback from benchmarks like FACET. This could lead to segmentation systems that not only perform better on average, but are demonstrably fairer and less biased across complex demographic axes.

References:

1. FACET: Fairness in Computer Vision Evaluation Benchmark. Laura Gustafson, Chloé Rolland, Nikhila Ravi, Quentin Duval, Aaron B. Adcock, Cheng-Yang Fu, Melissa Hall, Candace Ross (2023). IEEE International Conference on Computer Vision.
2. DDP: Diffusion Model for Dense Visual Prediction. Yuanfeng Ji, Zhe Chen, Enze Xie, Lanqing Hong, Xihui Liu, Zhaoqiang Liu, Tong Lu, Zhenguo Li, P. Luo (2023). IEEE International Conference on Computer Vision.
3. Enhanced retinal blood vessel segmentation via loss balancing in dense generative adversarial networks with quick attention mechanisms.. D. Sandeep, K. Baranitharan, A. Padmavathi, Loganathan Guganathan (2025). International ophtalmology.