Ended 3 years ago
135 participants
933 submissions

ID RnD Facial Anti-Spoofing

Detect spoofing attack by video frames from face recognition camera

Track description

People use phones and laptop webcams for face identification more and more often, which causes frequent attempts of fooling the ID algorithms. It involves a few different methods: impostors demonstrate printed portrait, or wear it above their face, or simply replay the video from another screen. More information on ways of face ID fraud detection can be found here: "Face Anti-Spoofing, or recognize a fraud of a thousand by its face in a technology-savvy way".

Quality criteria

Participants must build a facial fraud detection algorithm using sequences of frames. Submission must contain a Dockerfile that will be used to build a docker container. Evaluation metrics: P(falsealarm) + 19⋅P(miss), which equals FP/(FP+TN) + 19⋅FN/(FN+TP).


The prize pool is 600 000 ₽! Teams up to 6 participants of any age and citizenship are allowed. The winners get 300 000 ₽, second place receives 150 000 ₽, and a team placed third is granted with 75 000 ₽. There is also an additional Best Solution prize of 75 000 ₽, awarded upon the judges' decision. Moreover, everyone making a submit of 0.15 and better is treated with t-shirts from the upcoming first Data Souls merch collection within next week! 

Best solution nomination

Following the challenge results, the best solution in terms of quality-compactness ratio is awarded a prize of 75 000 ₽, based upon the decision of judges:

Cookies help us deliver our services. By using our services, you agree to our use of cookies.