AxonData/liveness-detection-dataset

Face Liveness Detection Dataset for Anti-Spoofing & PAD Certification

100,000+ spoofing videos for liveness detection

A comprehensive face liveness detection dataset for face anti-spoofing, biometric face recognition, and presentation attack detection (PAD) systems. Unlike narrow public benchmarks that cover only one or two attack types, this dataset combines all major presentation attack categories in a single resource: paper attacks, replay attacks, and 3D mask attacks (silicone, latex, paper-wrapped, resin, cloth), delivering 100,000+ videos across 11+ labeled attack types mapped to iBeta Level 1, Level 2, and Level 3 PAD certification

What Is Face Liveness Detection?

Face liveness detection is the biometric verification step that determines whether a captured face belongs to a live person rather than a spoofed presentation (printed photo, video replay, 3D mask). It is the core defense against presentation attacks in face recognition systems used for eKYC, fintech onboarding, banking authentication, and government identity verification

Robust liveness detection requires training on diverse attack vectors, which is why this aggregated dataset combines multiple attack categories rather than focusing on a single one

Why Use This Dataset

• All major attack vectors in one resource - reduces dataset assembly overhead vs combining CASIA-FASD + Replay-Attack + OULU-NPU + others
• Direct iBeta mapping - each attack labeled with its iBeta level (L1/L2/L3) for certification-ready training
• Modern capture quality - current-generation smartphones (iPhone 14/13 Pro, Galaxy S23, Pixel 7, Redmi, Honor 70), not academic 2015-era setups
• Production-aligned conditions - indoor and outdoor environments, varied lighting, balanced demographics

Quick Stats

• ~100,000 videos
• 11+ attack types (expandable on request)
• Capture devices: iPhone 14, iPhone 13 Pro, Samsung Galaxy S23, Google Pixel 7, Xiaomi Redmi, Honor 70, and others
• Indoor and outdoor environments
• Balanced gender mix and multi-ethnic representation (Caucasian, Black, Asian, Latinx)
• Active liveness phases: fixed, zoom-in, zoom-out

Face anti-spoofing dataset — examples of presentation attacks

Full version of the dataset is available for commercial usage. Leave a request on our website Axonlabs to purchase the dataset 💰

For feedback and additional sample requests, please contact us!

Attack Types in This Dataset

Each video sequence is labeled with one of the following classes:

Label	Type	iBeta Level
live	Bona fide (genuine face)	—
photo_print	Printed photo	iBeta L1 (2D)
cutout_2d_mask	Cut-out 2D mask	iBeta L1
on_actor_print	Worn paper attack	iBeta L1
cylinder_3d_mask	Cylinder paper mask	iBeta L1
mobile_replay	Phone screen replay	iBeta L1
display_replay	Monitor/tablet replay	iBeta L1
3d_paper_mask	3D paper mask	iBeta L1 / L2
wrapped_3d_print	Paper-wrapped 3D	iBeta L2
silicone_mask	Silicone 3D mask	iBeta L2
latex_mask	Latex mask	iBeta L2
cloth_3d_mask	Fabric 3D mask	iBeta L2
resin_mask	High-fidelity resin mask	iBeta L3

Example Use Cases

• Train binary anti-spoofing classifier - bona fide vs all attack types combined
• Multi-class attack-type classifier - useful for explainable AI ("this is a silicone mask attack" vs "this is a replay") and model debugging (which attack types fail?)
• iBeta certification preparation - filter dataset to L1 attacks → train → benchmark; same for L2 and L3
• Cross-attack generalization research - analyze performance gaps between paper, replay, and 3D mask attacks

Academic Reference

This commercial dataset complements canonical academic benchmarks in face anti-spoofing research:

• Idiap Replay-Attack — replay attack baseline
• Idiap CSMAD / 3DMAD — silicone and 3D mask baselines
• OULU-NPU — mobile face liveness benchmark
• MSU-MFSD — mobile spoofing detection benchmark
• CASIA-FASD / CASIA-SURF — 2D and multi-modal benchmarks

This dataset extends those research lines with significantly more participants, modern smartphone capture conditions, broader demographic diversity, and direct iBeta certification mapping, designed for production face recognition systems rather than research benchmarks alone

Related Datasets by Axon Labs

• Silicone Mask Dataset - high-realism silicone masks for iBeta Level 2
• 3D Paper Mask Attack Dataset - volumetric paper attacks
• Advanced Paper Attacks Dataset - 7 advanced paper attack scenarios
• Photo Print Attack Dataset - basic 2D photo print attacks
• Display Replay Attacks Dataset - display-based replay attacks
• Replay Attack Dataset - broader replay coverage
• iBeta Level 1 Certification Dataset
• iBeta Level 2 Certification Dataset
• iBeta Level 3 Dataset

About Axon Labs

Axon Labs builds biometric AI training datasets. We specialize in face liveness detection, face recognition, and voice anti-spoofing data for production identity verification, eKYC, fintech, and government applications

Commercial Access

Sample subset publicly available for evaluation. For full commercial dataset access, pricing, and licensing terms, contact sales@axonlabs.pro or visit axonlab.ai