Model Fairness Validation

via “fairface dataset-based demographic fairness”

image-classification model by undefined. 71,05,775 downloads.

Unique: Explicitly trained on FairFace dataset which was designed with demographic fairness as a primary objective, using stratified sampling to ensure balanced representation across age, gender, and ethnicity. This differs from models trained on naturally imbalanced datasets (e.g., IMDB-Face, VGGFace2) which tend to overfit to majority demographics.

vs others: More equitable across demographic groups than generic age classifiers trained on imbalanced datasets; comparable fairness to other FairFace-trained models but with ViT architecture advantages for capturing global facial structure.