I'll have to look into the terminology to be confident of the best match for this context, but I figure "Image-Restricted, Label-Free Outside Data Results"[1] m

I'll have to look into the terminology to be confident of the best match for this context, but I figure "Image-Restricted, Label-Free Outside Data Results"[1] might match the scenario where a recognizer provides a yes/no after being trained on e.g. driver's licenses (let's call it the "Super Bowl" model). None of the results appear to be greater than 95%, which I understand to be a hard ceiling for requiring human QC. Many of the other methods get you worse than 50-50 chance of false positives by the time the algo gets to 95% true positives, but not being a statistician I may be reading the results wrong, in which case I'm totally willing to be corrected. Empirically, though, we know that the Super Bowl model has never worked. 1. http://vis-www.cs.umass.edu/lfw/results.html#ImageRestrictedLF http://vis-www.cs.umass.edu/lfw/results.html#ImageRestricted...