Researchers have introduced WIDER-FAIR, a new dataset designed to evaluate fairness in face detection models. Built upon the WIDER-FACE benchmark, WIDER-FAIR includes manual annotations for perceived ethnicity and sex across 16,256 images and four ethnic groups: Asian, Black, Indian, and White. Initial experiments using a YOLOv5 model trained on this dataset revealed that face detection performance is significantly lower for Black individuals, and excluding this group from training exacerbates fairness disparities more than excluding any other demographic. AI
IMPACT Highlights the need for diverse datasets to ensure fairness and mitigate bias in AI systems, particularly in computer vision applications.
RANK_REASON The cluster describes a new academic dataset and associated research paper.
- Black
- Indian
- k-nearest neighbors algorithm
- t-Distributed Stochastic Neighbor Embedding
- White
- WIDER FACE
- WIDER-FAIR
- YOLOv5
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
