I think they used the correct ROC curve. Actually, it looks like they cherry-picked a few results from the Commercial and the Academic categories, but everythin

I think they used the correct ROC curve. Actually, it looks like they cherry-picked a few results from the Commercial and the Academic categories, but everything is from the unrestricted class. We should wait until Erik Learned-Miller lists this on the LFW results page. He'll know how to interpret their results.