UniDataPro/2d-masks-pad-attacks

2D Masks with Eyeholes Attacks

The dataset comprises 11,200+ videos of people wearing of holding 2D printed masks with eyeholes captured using 5 different devices. This extensive collection is designed for research in presentation attacks, focusing on various detection methods, primarily aimed at meeting the requirements for iBeta Level 1 & 2 certification. Specifically engineered to challenge facial recognition and enhance spoofing detection techniques.

By utilizing this dataset, researchers and developers can advance their understanding and capabilities in biometric security and liveness detection technologies. - Get the data

Attacks in the dataset

The attacks were recorded in various settings, showcasing individuals with different attributes. Each photograph features human faces adorned with 2D masks, simulating potential spoofing attempts in facial recognition systems.

Variants of backgrounds and attributes in the dataset:

💵 Buy the Dataset: This is a limited preview of the data. To access the full dataset, please contact us at https://unidata.pro to discuss your requirements and pricing options.

Researchers can utilize this dataset to explore detection technology and recognition algorithms that aim to prevent impostor attacks and improve authentication processes.

Metadata for the dataset

Variables in .csv files:

• name: filename of the printed 2D mask
• path: link-path for the original video
• type: type(wearing or holding) of printed mask

The dataset provides a robust foundation for achieving higher detection accuracy and advancing liveness detection methods, which are essential for preventing identity fraud and ensuring reliable biometric verification.

Frequently Asked Questions

What video resolution, format, and duration does the full 2D Masks with Eyeholes Attacks Dataset provide?

The full anti-spoofing mask dataset is delivered in MP4 and MOV formats, with resolutions ranging from 1920 × 1080 up to 3840 × 2160. Each clip runs approximately 4 seconds, providing enough footage to capture a presentation attack attempt while keeping storage and preprocessing requirements manageable for real-time liveness detection pipelines.

Was the 2D mask attack data collected from real-world recordings?

Yes. The collection was gathered through crowdsourcing platforms rather than generated entirely through synthetic video. The recordings therefore represent actual capture conditions involving people and printed 2D mask attacks.

How much background variation is included in the dataset?

The videos were recorded against nine different backgrounds, introducing environmental variation beyond the mask attack itself.

🌐 UniData provides high-quality datasets, content moderation, data collection and annotation for your AI/ML projects