Dataset

For the competition there will be no training database supplied other than a few samples per spoof category to learn the basic properties of images/videos. Competitors may use any publicly available, or proprietary face PAD databases. Click here to see a list of suggested training datasets.

Test datasets will have different attack types that are shown below in the table. The datasets will be the same between the two competitions with one being a single image and the other being a five second video.

ClassAttack typesSample countSensorData Supplier
LiveN/A400DSLR, Samsung Galaxy S9, Google Pixel, Iphone XClarkson University
SpoofFace printed out on inkjet or lower quality printer200DSLR, Samsung Galaxy S9, Google Pixel, Iphone XClarkson University
SpoofFace printed out on photography paper200DSLR, Samsung Galaxy S9, Google Pixel, Iphone XClarkson University
SpoofFace displayed on a phone or laptop200DSLR, Samsung Galaxy S9, Google Pixel, Iphone XClarkson University
SpoofFace printed out and made into paper mask200DSLR, Samsung Galaxy S9, Google Pixel, Iphone XClarkson University
Spoof3D printed mask160DSLR, Samsung Galaxy S9, Google Pixel, Iphone XClarkson University
SpoofSilicon mask30RGB GQ-WMCA subsetIdiap Research Institute
SpoofSilicon mask80DSLR, Samsung Galaxy S9, Google Pixel, Iphone XClarkson University
Total1470

All samples in the table have not yet been published, and will be offered to the public through BEAT platform after the competition has concluded.