V13 N1 Paper 5
Annals of the MS in Computer Science and Information Systems at UNC Wilmington
Spring 2019

MetaFace: A System for Benchmarking Face Processing APIs  

Kevin Gay

Committee

Karl Ricanek (chair)
Jeffrey Cummings
Lucas Layman

Abstract

Even though APIs and facial processing are new areas of technology, there are already an abundance of cloud-based APIs that provide facial processing services to companies, researchers, and government agencies. Google, Microsoft, IBM, and Amazon are some of the major companies that offer these services; however, there are dozens of other companies out there, from startups to established multinationals, that offer the same services. Many of the lesser known companies may be cheaper and provide more optimal solutions, i.e. better measured performance, however they lack the name recognition of the software giants. Because of the dominance of these software giants, i.e. Microsoft, Google, IBM, and Amazon, face processing has received negative coverage in recent news due to lack of performance, gender bias, race bias, and privacy concerns [60, 61, 62]. In some cases, the lesser known entities may have developed algorithms that outperform the giant’s, but their collective voices are muted due to the press coverage around the giants. For this reason, I developed a meta-API which calls several face processing APIs, aggregates the outputs, and delivers the outputs in a user configured manner. An end user could use this system to evaluate several solutions against the same set of inputs, performing deep evaluations on against a set of solutions, which currently does not exist. An end user could also aggregate the results based on the solution that performs the best on a given output to mitigate against algorithm bias or, simply, to generate the most accurate result.

download (pdf)

Recommended Citation: Gay, K., Ricanek, K, Cummings, J., Layman, L. (2019) MetaFace: A System for Benchmarking Face Processing APIs. Annals of the Master of Science in Computer Science and Information Systems at UNC Wilmington, 13(1) paper 5. http://csbapp.uncw.edu/data/mscsis/full.aspx.

V13 N1 Paper 5
Annals of the MS in Computer Science and Information Systems at UNC Wilmington
Spring 2019