README.md

---
license: cc-by-nc-4.0
tags:
- ibeta-level-3
- ibeta-certification
- face-anti-spoofing
- face-liveness-detection
- liveness-detection
- presentation-attack-detection
- pad
- biometrics
- face-recognition
- spoofing
- anti-spoofing
- 3d-mask-attack
- rubber-mask
- 3d-resin-mask
- silicone-mask-alternative
- high-fidelity-mask
- iso-30107-3
- active-liveness
- ekyc
- identity-verification
- biometric-authentication
- incode-ibeta-l3
- veridas-ibeta-l3
task_categories:
- video-classification
language:
- en
size_categories:
- 1K<n<10K
pretty_name: iBeta Level 3 Dataset
---
# iBeta Level 3 Dataset: 10,000+ Resin Mask Attack Videos

First-to-market anti-spoofing dataset for **iBeta Level 3** certification testing. 10,000+ attack videos with 12+ premium resin masks — the most advanced and realistic attack type in the PAD taxonomy.

Resin masks are custom-manufactured to replicate specific individuals' facial features with high precision, making them significantly harder to detect than silicone or latex masks.

![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F20109613%2Fba9ac07738cb252a7e0417da8d47327d%2FFrame%20131.png?generation=1758814004219160&alt=media)

## Dataset Details

- *10,000+ videos**, multiframe
- **12+ unique resin masks** — balanced mix of genders
- Captured on **iOS and Android** phones
- **Active liveness prompts:** zoom in/out, right/left head rotation, head tilt
- Indoor locations with varied lighting conditions
- Partially paired - some masks include genuine face videos of the same subject

![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F20109613%2F64a3e85ed7c620d8d785b7e99397123a%2FFrame%20182.png?generation=1771854753130447&alt=media)

## Use Cases

1. **iBeta Level 3 pre-certification** — evaluate liveness detection models against resin mask attacks before submitting to iBeta testing
2. **PAD system hardening** — stress-test anti-spoofing pipelines with the most realistic attack type available
3. **Active liveness R&D** — train models that rely on motion-based cues (zoom, rotation, tilt) to detect mask presentations

## Full version of dataset is available for commercial usage — leave a request on [Axon Labs](https://axonlab.ai/?utm_source=hugging-face&utm_medium=cpc&utm_campaign=profile&utm_content=profile_link) website.
## For feedback and additional sample requests, please [contact us](https://axonlab.ai/dataset/ibeta-level-3-dataset/?utm_source=hugging-face&utm_medium=cpc&utm_campaign=profile&utm_content=profile_link)!