Using, CAST we develop a new face recognition benchmark, called CAST-Challenging-11. The CC11 has two differentiating factors from previous benchmarks:
- Contains 11 sub-benchmarks to allow comparison between performance on different demographics and attribtues.
- Contains only hard pairs as to not waste compute cycles on trivial verification pairs. Comparatively, current benchmarks contains mostly trivial verification pairs where performance is as high as 99.85%.
The below table shows results of 12 different models on the CAST benchmark.
This work was completed as a collaboration between the VAST Lab @ UCCS and the Computer Vision Research Lab (CVRL) @ Notre Dame. Collaborators on the project are Steven Zhou, Aman Bhatta, Chad Mello, and Vitor Albiero. The project was advised by Dr. Terrance E. Boult and Dr. Kevin W. Bowyer.
This research is based upon work supported in part by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via [2022-21102100003].