The dataset includes selfies from over 1,000 people. Later, a team of over 200 people made 5,000+ replay display attacks based on these selfies. The attacks provide diversity of lighting, devices, and screens.
Introduction
The Liveness Detection: Display Replay Attacks Dataset is a comprehensive resource designed to improve anti-spoofing technology, specifically in identifying display replay attacks. This dataset features authentic selfies from over 1,000 participants, followed by 5,000+ replay attacks performed by a team of over 200 people. By offering a variety of lighting, devices, and display scenarios, this dataset supports the development of robust liveness detection models
Dataset summary
With a balanced representation of gender and ethnicity, the dataset provides over 5,000 replay attack scenarios based on authentic selfies from diverse participants. Each attack involves dynamic camera angles, enabling models to accurately distinguish between live and replayed display images in real-world conditions
Source and collection methodology
The dataset was collected through voluntary contributions of selfies from 1,000+ individuals, each image of high quality (720p or greater) and free of filters. Replay display attacks were then performed on various screens and recorded from multiple angles, with each video lasting at least 12 seconds to simulate realistic attack scenarios
Use cases and applications
Ideal for developers and researchers focusing on liveness detection, this dataset is particularly useful for training models to identify display replay attacks. Its applications include enhancing facial recognition systems and biometric authentication, as well as improving general anti-spoofing measures within security solutions
Dataset features
Over 1,000 individuals shared selfies
Balanced mix of genders and ethnicities
More than 5,000 display attacks crafted from these selfies
Real life selfies description
Each person provided one selfie
Selfies are at least 720p quality
Faces are clear with no filters
Replay display attacks description
Videos last at least 12 seconds
Cameras move slowly, showing attacks from various angles
Download information
A sample version of this dataset is available on Kaggle. Leave a request for additional samples in the form below
Replay attack description:
Over 1,000 individuals shared selfies. Balanced ethnicities: Caucasian, Latin American, Black, Asian
15 seconds length of Screen attacks. Camera moves slowly, showing attacks from various angles
Variety of capturing monitors and devices
Devices used are: laptop, phone, tablet
Real life selfie description:
Each person provided one selfie
Selfies are at least 720p quality
Faces are clear with no filters
Best used for
Liveness detection
Antispoofing attack detection
Contact us
Tell us about yourself, and get access to free samples of the dataset