Silicone Mask Attack Dataset for Face Anti-Spoofing

Silicone mask attacks are 3D presentation attacks where an attacker wears a professionally manufactured silicone mask designed to replicate a target person's facial features with photorealistic skin texture and depth. Silicone masks are the flagship attack class tested at iBeta Level 2 PAD certification - more sophisticated than 2D print or replay attacks, and harder to detect because real skin texture is impossible to distinguish from high-quality silicone without advanced liveness cues. Testing follows ISO/IEC 30107-3 Level 2 requirements

The dataset contains 12,500+ videos across 18 high-detail silicone masks, approximately 700 videos per mask identity for deep within-mask training. Each mask is recorded with ~40 attribute combinations (hairstyles, glasses, wigs, beards) to maximize within-identity diversity. Videos are captured on 8 iOS and Android device models across 10 distinct locations (4 office spaces, 4 home apartments, 2 outdoor scenes) under three lighting tiers (low, medium, bright), with active liveness sequences (natural head movements, blinking) in every video

Silicone, paper, and replay attacks are three fundamentally different presentation attack vectors, each defeating face recognition and
liveness detection systems through different means.

Paper attacks (cutouts, eyeholes masks, wrapped paper masks) are 2D presentations where a printed photo is held up or shaped to
mimic a face. They are the simplest presentation attack class and are primarily tested in iBeta Level 1 PAD certification.

Replay attacks display a previously captured photo or video on a secondary screen — smartphone, laptop, or monitor. They retain
motion and color but exhibit characteristic moiré patterns, screen edges, and pixelation that anti-spoofing models can detect.

Silicone mask attacks are 3D physical artifacts that reproduce facial geometry, skin texture, and even subtle reflectance properties. Unlike 2D paper or screen-based replay attacks, silicone masks have real depth and can defeat infrared- and motion-based liveness detection. They are tested in iBeta Level 2 PAD certification and are significantly harder to detect than 2D attack vectors.

Silicone masks are the hardest presentation attack vector to detect because they replicate the physical and optical properties of realmhuman faces more faithfully than any other attack type.

First, silicone has reflectance and translucency similar to human skin — defeating texture-based face anti-spoofing models that look for paper grain, screen pixelation, or color artifacts characteristic of paper or replay attacks.

Second, silicone masks have true 3D facial geometry — defeating depth-based and infrared (IR) liveness detection systems that distinguish 2D presentations from genuine 3D faces.

Third, modern high-fidelity silicone masks include movable elements (eyes, mouth) and can be combined with accessories such as wigs, glasses, and beards to mimic active liveness signals like natural blinking and head motion.

This is why iBeta Level 2 PAD certification specifically test against silicone mask attacks — they represent the upper bound of presentation attack realism that production face recognition and biometric authentication systems must defeat under the ISO/IEC 30107-3 standard.

Attacks were captured using front-facing (selfie) cameras on 8 iOS and Android device models: iPhone 14, iPhone 14 Pro, iPhone 13 Pro, Galaxy S23, Xiaomi Redmi Note 12 Pro+, Galaxy A54, Pixel 7, and Honor 70. Each of the 18 silicone masks is recorded across approximately 40 attribute combinations: hairstyles, glasses, wigs, beards, in 10 distinct locations (4 office spaces, 4 home apartments, 2 outdoor scenes) under three lighting tiers (low, medium, bright). Every video includes active liveness sequences: natural head movements and zoom-in/zoom-out

Yes. All data is collected with explicit written informed consent from every participant in compliance with GDPR Article 9, which specifically governs the processing of biometric data. The dataset is licensed for commercial use in AI model training, validation, and iBeta certification preparation. Comprehensive compliance documentation, including consent provenance, collection methodology, and legal basis review, is available upon request for regulatory or audit purposes

Yes. A sample version of this dataset is available on request, you can verify silicone mask quality, the attribute combination diversity, active liveness sequences, and format compatibility with your training pipeline before committing to the full dataset. Submit a request through the form on this page for sample access, typically delivered within 1–2 business days