Cutout Print Attack Dataset
Check samples on Kaggle
Dataset summary
|
Parameter
|
Value
|
|---|---|
|
Volume
|
4,000+ cutout mask attacks from 2,000+ participants
|
|
Coverage
|
Cutout print attacks for iBeta Level 1 PAD certification
|
|
Demographics
|
Balanced mix of genders and ethnicities, adult participants
|
|
Devices
|
iOS and Android phones
|
|
Conditions
|
Indoor and outdoor, varied lighting, multiple print quality levels
|
Introduction
The Cutout 2D Attacks Dataset is a specialized resource aimed at enhancing Presentation Attack Detection (PAD) systems by focusing on cutout photo print attacks. This dataset is designed to support AI developers in training liveness detection models capable of identifying 2D cutout print attacks. This dataset could be utilized by both iBeta and NIST FATE to evaluate and advance anti-spoofing measures
Source and collection methodology
Collected through the participation of more than 2000 individuals, the dataset includes high-quality cutout photos used in attacks. Each attack is captured in a 10-15 second video, with photos presented flat and directly facing the camera to maintain consistency. These controlled conditions support model training by providing a stable view of each attack scenario
Use cases and applications
Ideal for researchers and developers in the field of liveness detection, this dataset enables robust training for models to accurately differentiate between real faces and cutout 2D photo prints. This resource is particularly valuable for those working on facial recognition systems, biometric authentication, and PAD technology
Dataset features
- 2,000+ participant diversity: Wide identity coverage for cross-face model generalization
- High-fidelity print quality: Realistic skin tones and color reproduction
- Standardized flat presentation: Cutout photos held flat, straight-on to camera
Photo print attack description
- Each attack comprises of 10-15 sec. video
- High-quality cutout photos with realistic colors
- Paper attacks conducted on flat photos with a straight view on the camera (not bent or skewed)
See Also: Broader Print Attack Coverage
This Cutout Print Attack Dataset focuses specifically on 2D cutout photo attacks – flat photos cut to face contours. For broader print-based presentation attack coverage including standard photo prints, zoom-in effects, and NIST-FATE-compliant scenarios, see our Photo Print Attack – 7,000+ attacks from 3,000+ participants, the comprehensive print attack collection
Other Datasets for iBeta 1:
Download information
A sample version of this dataset is available on Kaggle. Leave a request for additional samples in the form below
Have a question?
A cutout print attack is a 2D presentation attack vector against face recognition and liveness detection systems, where a printed photo of a real person is cut to the shape of a face contour and presented to a camera to deceive biometric authentication. Cutout attacks are a subtype of paper print attacks distinguished by the precise face-shaped cutting that reduces visible image borders during liveness checks. They are tested as part of iBeta Level 1 PAD certification under the ISO/IEC 30107-3 standard and represent one of the foundational spoofing techniques in face anti-spoofing research
A regular photo print attack uses a flat printed photo with full rectangular borders held in front of the camera. A cutout print attack uses the same flat photo but cut precisely to face contours, removing the rectangular border that makes basic prints easier to detect. This makes cutout attacks harder for face anti-spoofing models that rely on detecting paper edges or rectangular shapes around the face. Cutouts therefore serve as a more challenging variant of 2D print attacks and are especially relevant for evaluating presentation attack detection (PAD) algorithm robustness in production face recognition systems
The dataset contains 4,000+ cutout print attacks recorded from 2,000+ unique participants. Each attack is captured as a 10–15 second video featuring high-quality cutout photos with realistic colors, presented flat and directly facing the camera under controlled conditions. The participant pool includes a balanced gender mix and broad ethnicity coverage to enable face anti-spoofing models to generalize across diverse demographics. The dataset is designed for liveness detection model training, biometric authentication system testing, and PAD algorithm benchmarking aligned with the ISO/IEC 30107-3 standard
Yes. Cutout 2D print attacks are explicitly tested in iBeta Level 1 PAD certification, which evaluates whether biometric systems can defeat basic 2D presentation attacks before being approved for production deployment. This dataset provides 4,000+ cutout attack samples that can be used to train and evaluate face anti-spoofing classifiers ahead of formal iBeta testing. The dataset is also compatible with NIST FATE benchmark methodology, supporting algorithm evaluation under both iBeta and NIST testing frameworks
The Photo Print Attack Dataset is the broader collection of all print-based presentation attacks, including standard rectangular printed photos, zoom-in scenarios, and 7,000+ attacks from 3,000+ participants. This Cutout Print Attack Dataset focuses specifically on the cutout subset — flat photos cut to face contours without visible borders, captured under controlled conditions for consistent model training. The two datasets are complementary: use the Photo Print Attack Dataset for general PAD coverage, and the Cutout Print Attack Dataset for targeted training against contour-cut spoofing variants. Both align with iBeta Level 1 PAD certification requirements
Contact us
Tell us about yourself, and get access to free samples of the dataset
