CounterFace: A Synthetic Face Dataset for Fine-Grained Counterfactual Evaluation of Face Recognition Systems
Researchers have introduced CounterFace, a novel synthetic dataset designed for the fine-grained evaluation of face recognition systems. This dataset comprises 11,821 counterfactual face pairs across 20 facial attributes and 8 demographic factors, significantly expanding upon previous synthetic datasets. CounterFace was generated using a fully automated pipeline, removing the need for human verification in the synthesis process. The dataset was used to evaluate six commercial and open-source face recognition systems, revealing performance degradations that vary by attribute and demographic, with occluding factors like masks and facial hair consistently degrading performance. AI
IMPACT Provides a new benchmark for assessing the robustness and potential biases of face recognition AI.