Face recognition and verification performance under up to 1 million distractors. Performance is measured using probe and gallery images from FaceScrub, a labeled data set. FG-Net is also used to further stress the age invariance properties of algorithms. Interested to examine the results further? Download the JSON results from Challenge 1 here.
Training on 672K identities, and then testing recognition and verification performance under 1 million distractors. Probe and gallery images are used from FaceScrub (celebrity photos) and FG-Net (to test age invariance). Interested to examine the results further? Download the JSON results from Challenge 2 here.