IBeta 2

A leading company in the electronic payments industry planned to train a facial recognition model to detect advanced attacks and also undergo Level 2 certification from the iBeta laboratory. To achieve this, the use of silicone and latex masks was planned.

However, purchasing the required props independently required a substantial budget. Additionally, preparing the masks in the client’s country would take over 3 months. Since the company had previously collaborated with Training Data, it decided to approach us again

We offered the client the iBeta 2 dataset consisting of 33,000 attacks. The dataset was specifically curated to meet the certification requirements of the iBeta Level 2 laboratory.

The dataset comprised data (50-100 frames, up to 5 seconds) with videos of real people featuring 30 types of clothing, accessories, and facial hair on 40 backgrounds. It also included a set of attacks for each person. Testing at level 2 imposed a cost of up to $300 for each attack. The dataset included attacks such as:

• Individually crafted silicone face masks based on people’s molds
• Latex masks with clothing and facial hair
• Contour-printed photos
• 2D masks with eye cutouts
• 3D cardboard masks of faces

The client received a dataset with 30,000 videos, encompassing all necessary attacks, and was able to save on budget and time