From June to September 2017, NIST evaluated 41 face recognition algorithms from 16 developers. The algorithms were applied to datasets of 2D still photographs in two ways: verification of “wild” photojournalism and social media images, and identification of faces from surveillance videos against portrait galleries of size up to 691 thousand. The Face Recognition Prize Challenge (FRPC) is documented in NIST Interagency Report 8197.
In conjunction with IARPA, NIST is running its first Face Recognition Prize Challenge (FRPC) to assess capability of the latest algorithms operating on unconstrained images. IARPA will award cash prizes to the most accurate identification and verification algorithms. Details appear on the IARPA challenge website. Participants must follow the IARPA rules, and then implement NIST’s API, below. The test proceeds by NIST executing participants’ algorithms on images sequestered at NIST.
The FRPC is aimed at measurement of automated face recognition accuracy when sub-optimal images are identified and verified against either other such images, or against standards-compliant portraits. Sub-optimal images are characterized by non-frontal head pose, low resolution, poor and uneven illumination, non-neutral facial expression, and occlusion. In almost all cases, the input image will contain one face only. NIST's performance reports will include measurements of accuracy, speed, storage and memory consumption. NIST will report the dependence of performance on the properties of the images and the subjects.
Participation Agreement [last update: 2017-04-11]
FRPC is conducted by NIST, an agency of the United States Government. Participation is free of charge. FRPC is open to a global audience of face recognition developers. All organizations who seek to participate in FRPC must sign and submit all pages of this Participation Agreement.
Validation [last update: 2017-04-11]
The validation package is available here. All participants must run their software on the images provided in the validation package prior to submission. The purpose of validation is to ensure reproducibility. Our execution of your algorithm must produce identical outputs to your execution.
Important Dates [last update: 2017-04-12]
Submission Deadline: All participation agreements and algorithms shall be submitted to NIST by 2017-06-15.
Prospective FRPC participants may find value in submitting their algorithm into NIST’s existing FRVT evaluation. That test, which currently only evaluates 1:1 verification, returns results to developers within approximately 10 days. FRVT does not award cash prizes. The latest FRVT report is posted on the FRVT website.
Inquiries and comments may be submitted to FRPC@nist.gov.