Axon Labs

Axon Labs Datasets offers a catalog of biometric datasets for liveness detection, face recognition, and audio biometrics.

Datasets cover iBeta Level 1–3 attack vectors, including replay, photo print, cutout, paper masks, silicone/latex/3D resin masks, wrapped attacks and other face anti-spoofing scenarios.

Collections include selfies, behavioral videos, NIST-compliant face recognition sets, synthetic children faces, Web IR+RGB captures, and multi-language call center speech corpora for voice biometrics and speech recognition.

Data is delivered in ML-ready formats (MP4/MOV, JPEG/PNG) with CSV metadata and annotations compatible with PyTorch and TensorFlow pipelines.

Balanced train/test splits, demographic diversity, and attack coverage support model training, bias analysis, and pre-certification validation for iBeta and NIST testing.

Custom data collection and supplemental filming address missing attack types or project-specific requirements.

Key Features

• Specialized biometric datasets for liveness detection, face recognition, and audio biometrics
• Comprehensive attack-vector coverage across iBeta L1/L2/L3 (print, replay, cutout, paper masks, 3D/silicone/latex/wrapped masks, etc.)
• ML-ready data formats and metadata (MP4/MOV, JPEG/PNG, CSV annotations) compatible with PyTorch and TensorFlow
• ML-focused dataset organization with proper train/test splits, balanced classes, and detailed annotations aligned to certification tests
• Custom data collection and augmentation services to capture missing attack types and produce tailored datasets

Use Cases

• Train and benchmark robust liveness-detection and face anti-spoofing models using Axon Labs Datasets' ML-ready, annotated samples that cover iBeta Level 1–3 attack types with balanced demographic splits for fair, production-ready performance evaluation
• Develop and validate multi-language voice biometric authentication systems by leveraging the curated speech corpus with precise annotations and custom collection options to optimize speaker recognition and reduce false accepts across languages
• Perform compliance testing, adversarial attack simulations and model hardening using the spoof attack datasets and custom collection features to replicate real-world attacks, measure system resilience, and produce reproducible benchmarks for audits and product demos

Who is it for?

• Machine learning engineers
• Data scientists
• Product managers
• Biometric researchers
• Biometric hobbyists