A framework for bias-aware dataset evaluation in soft facial attribute recognition

Download PDF
Open Webpage

Lucia Cascone, Michele Nappi, Chiara Pero, Xinggang Wang

Published: 2026, Last Modified: 13 Nov 2025Pattern Recognit. 2026EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•A unified framework is proposed to assess bias in facial attribute datasets.•Demographic annotations are standardized across datasets via the FairFace model.•Attribute label inconsistencies are addressed using vision-language model outputs.•Bias understanding is advanced through critical analysis of evaluation metrics.