The VisaBorder takes pairs from Visa and Wild datasets. The Visa photos are enrolled, and the algorithm needs to verify the images from the wild dataset. The visa border and wild datasets are more complex to verify, considering the photos in the wild are unconstrained, implying that the image's lighting conditions, pitch, yaw, and roll may vary.
Testing Method:
The comparisons are of visa-like frontals against border crossing webcam photos.
The number of genuine comparisons is on the order of 1,000,000.
The number of impostor comparisons is on the order of 100,000,000.
The impostors are paired by sex but not by age or other covariates.
The number of persons is on the order of 1,000,000.
The number of images used to make one template is 1.
The number of templates used to make each comparison score is two, corresponding to simple one-to-one Verification.