AxonData/paper-attack-dataset

Liveness Detection Dataset: iBeta level 2 advanced mask attacks (5 K videos)

This advanced paper mask attack dataset focuses on complex paper-based presentation attacks for face anti-spoofing, liveness detection, and biometric face recognition systems. The dataset contains 5,000 videos across 7 distinct attack scenarios - including printed-attribute photos, cut-out photo masks, photo masks worn by actors with real accessories (wigs, hats, glasses)

Recorded from 25 participants across iOS and Android devices with balanced gender mix and multi-ethnic representation (Caucasian, Black, Asian, Latinx). Active-liveness phases (fixed, zoom-in, zoom-out) are included for robust presentation attack detection (PAD) model training. Aligned with the ISO/IEC 30107-3 standard and designed for iBeta Level 2 certification preparation

Full version of dataset is availible for commercial usage - leave a request on our website Axonlabs to purchase the dataset 💰

Types of Presentation Attacks (paper masks)

• 1. Printed attributes on photo – a flat facial photo with accessories (e.g., glasses, hat) printed together with the face.
• 2. Cut-out attributes in photo – a flat facial photo cut to the shape of the face.
• 3. External attributes on top of photo – a flat facial photo with real accessories (glasses, cap, etc.) attached on top.
• 4. Photo mask on actor + external attributes – a full-size photo fixed to an actor’s face; real items such as a hood or wig are added.
• 5. Photo mask on actor, printed attributes – a fixed photo that already contains additional printed attributes.
• 6. Photo mask on actor with eye holes + external attributes – eye openings are cut in the photo; the actor blinks through them while wearing real wig/clothing.
• 7. Photo mask with printed attributes and eye holes – combines printed accessories on the photo with the actor’s live eyes visible through cut-outs.

Potential Use Cases

• Liveness detection R&D: train / benchmark algorithms that separate selfies from 3D mask spoofs with high accuracy.
• iBeta level 2 pre-certification: stress-test PAD models against high-realism 3D mask scenarios before formal audits.
• Cross-material studies: analyse generalisation gaps between silicone, latex, paper and textile attacks for robust deployment.

Related Datasets

• 3D Paper Mask Attacks for Liveness — volumetric 3D paper mask attacks
• Display Replay Attacks — screen replay attacks
• Print Attack Dataset — photo print attacks
• iBeta Level 1 Certification Dataset — iBeta L1 attack set

Keywords: iBeta certification, PAD attacks, Presentation Attack Detection, Antispoofing, Facial Biometrics, Biometric Authentication, Security Systems, Machine Learning Dataset