AxonData/replay-attack-dataset

Commercial collection of 10,000+ replay attack videos covering both display-based and mobile-based replay spoofing. Captured from 2,500+ unique participants with balanced gender and ethnicity representation. Suitable for training face anti-spoofing models and preparing for iBeta PAD Level 1 certification

Dataset Summary

Replay attacks are one of the most common presentation attack types in face biometric systems. Attackers replay a video or image of a genuine user on a screen (phone, tablet, monitor) to bypass liveness detection. A robust liveness model must see diverse replay scenarios during training

This dataset combines two attack streams into a single training resource:

• Display replay attacks - captured from a variety of screen devices (monitors, laptops, tablets) with diverse lighting conditions
• Mobile replay attacks - captured across 15+ different phone models spanning low-end, mid-range, and high-end segments, including Samsung Galaxy A54, Honor 70, Google Pixel 7

Both attack streams share the same capture protocol: slow camera movement, multi-angle coverage

Dataset Characteristics

Feature	Value
Total attack videos	10,000+
Unique participants	2,500+
Attack sub-types	Display replay + Mobile replay
Mobile device coverage	15+ phone models (incl. Galaxy A54, Honor 70, Pixel 7)
Camera behavior	Slow movement, multi-angle
Gender distribution	Balanced
Ethnicity distribution	Balanced
iBeta compliance	Level 1 (PAD)

For commercial use, production deployment, or access to the full 10,000+ video dataset, contact Axon Labs at axonlab.ai

Use Cases

• Face anti-spoofing model training: expand replay attack coverage beyond narrow academic datasets
• iBeta PAD Level 1 preparation: replay attacks are a required category for iBeta Level 1 certification
• Cross-device generalization: evaluate model performance across display types and phone models
• Benchmarking: compare production PAD models on diverse replay scenarios
• Fraud prevention R&D: detect screen replay attempts in KYC and remote onboarding systems

What Makes This Dataset Different

• Both attack streams in one: most public replay datasets cover either display OR mobile, not both
• Realistic mobile coverage: 15+ real phone models
• Commercial license available: suitable for production deployment, not research-only
• Consent-based: all subjects provided explicit consent for AI training use
• Scale: 10,000+ videos is larger than most publicly available replay collections

Related Datasets from Axon Labs

• iBeta Level 1 Dataset - full Level 1 certification package (paper + replay attacks)
• iBeta Level 2 Dataset - 3D mask attack coverage (silicone, latex, wrapped paper)
• iBeta Level 3 Dataset - high-fidelity mask attacks
• Photo Print Attacks Dataset - paper photo spoofing, 3K+ individuals
• Silicone Mask Dataset - 10K+ videos from 18 silicone masks

Full catalog: axonlab.ai/datasets

About Axon Labs

Axon Labs creates high-quality datasets for AI training, with a focus on biometric security, face recognition, and liveness detection

• Website: axonlab.ai
• Contact: sales@axonlabs.pro
• Full catalog: axonlab.ai/datasets

Other Axon Labs datasets on Hugging Face: huggingface.co/AxonData