AxonData/print-cardboard-mask-face-spoofing

Cardboard Mask Face Anti-Spoofing Dataset

3,000 videos from 50 participants of print cardboard mask attacks with real accessories - wigs, hats, glasses. Recorded on iPhone 12, iPhone 14 Pro, and Samsung S23. Built for training and evaluating face anti-spoofing, liveness detection, and presentation attack detection systems

Full dataset for commercial use — request a license at axonlab.ai

Why Accessory-Enhanced Masks Are a Harder Problem

Standard print attack datasets use flat photos held in front of a camera. Cardboard mask attacks with real accessories are fundamentally different:

• Real accessories (wigs, hats, glasses) add volume and conceal mask edges
• Rigid cardboard holds its shape when worn, unlike paper that bends and warps
• Mixed signals - real hair, clothing, and movement surround a printed face, confusing texture-based detectors

This makes accessory-enhanced cardboard masks a blind spot for PAD systems trained only on flat print or replay attacks

Dataset Specifications

• 3,000 videos from 50 unique participants
• Multi-device: iPhone 12, iPhone 14 Pro, Samsung S23
• Active liveness features: zoom-in/out, natural head movements, blinking
• Varied environments: diverse real-world backgrounds and lighting conditions
• Accessories per attack: wigs (various styles and colors), hats, caps, glasses, sunglasses

Applications

• Face anti-spoofing - train PAD models to detect accessory-enhanced print mask attacks
• Liveness detection - build systems robust to attacks combining printed faces with real-world objects
• iBeta certification preparation - test against advanced 2D presentation attacks before Level 1/2 submission

Need More Data?

This dataset is a ready-made sample. We offer custom data collection for cardboard mask attacks tailored to your requirements, including larger participant pools, additional devices, specific demographic distributions, and custom accessory configurations

Contact us at axonlab.ai to discuss your project.