A Rapid Test for Accuracy and Bias of Face Recognition Technology

TL;DR

This paper introduces a quick, annotation-free method for evaluating face recognition accuracy and bias, enabling easier and more responsible testing of FR systems using minimal data and revealing demographic disparities.

Contribution

The authors present a novel, embedding-based verification method that benchmarks FR systems rapidly without manual labels, and provide the first public bias benchmark across multiple cloud services.

Findings

The method accurately estimates FR performance with smaller datasets.

Demographic biases, especially lower accuracy for Asian women, are identified.

The approach reduces testing time and costs significantly.

Abstract

Measuring the accuracy of face recognition (FR) systems is essential for improving performance and ensuring responsible use. Accuracy is typically estimated using large annotated datasets, which are costly and difficult to obtain. We propose a novel method for 1:1 face verification that benchmarks FR systems quickly and without manual annotation, starting from approximate labels (e.g., from web search results). Unlike previous methods for training set label cleaning, ours leverages the embedding representation of the models being evaluated, achieving high accuracy in smaller-sized test datasets. Our approach reliably estimates FR accuracy and ranking, significantly reducing the time and cost of manual labeling. We also introduce the first public benchmark of five FR cloud services, revealing demographic biases, particularly lower accuracy for Asian women. Our rapid test method can democratize FR testing, promoting scrutiny and responsible use of the technology. Our method is provided as a publicly accessible tool at https://github.com/caltechvisionlab/frt-rapid-test