First in class in NIST's facial recognition evaluation.
Across seven independent trials. In three of them, the algorithm set a record in the entire history of testing.
NIST
First in class across seven independent trials, with records in three
IARPA
1st in verification accuracy and identification speed
ActEV · CVPR
2nd in activity detection in extended video
WIDER Challenge
3rd in pedestrian and cyclist detection on public VMS footage
Results since 2015.
- 2015University of WashingtonFace recognition benchmarkFirst in class facial recognition accuracy, ahead of Google and other competitors.
- 2017IARPAFace recognition challengeFirst in class verification accuracy, and best in class identification speed.
- 2017EmotioNet ChallengeEmotion recognitionBest in class emotion recognition score.
- 2018NISTFace recognition evaluationBest in class results for matching footage captured in uncontrolled conditions against database records.
- 2018WIDER ChallengePedestrian detectionThird place for detecting pedestrians and cyclists in public VMS footage.
- 2019ActEV · CVPRActivities in extended videoSecond place for movement and action recognition in video footage.
- 2020KaggleForgery detectionThird place in the forgery recognition contest.
- 2021NISTFace recognition evaluationThe first in class facial recognition algorithm according to seven independent trials. In three, it set a record in the entire history of testing.
Fast enough for a city's cameras.
- 0.3 s
- to search 1.5 billion facial images
- 24×7
- on an unlimited number of video streams
What these benchmarks measure.
- NISTThe US National Institute of Standards and Technology
- NIST runs the longest-standing independent evaluation of face recognition algorithms. Developers submit algorithms; NIST tests them on large sequestered datasets the developers never see.
- IARPAIntelligence Advanced Research Projects Activity
- IARPA's face recognition challenge tested how accurately and how quickly algorithms verify and identify faces in unconstrained images.
- ActEV · CVPRActivities in Extended Video
- ActEV measures how well systems detect human activities, such as people entering vehicles or carrying objects, in long surveillance videos.
- WIDERWIDER Face and Pedestrian Challenge
- WIDER benchmarks detection of faces and pedestrians in crowded, real-world scenes, including footage from surveillance cameras.