The dataset was created with the goal of evaluating models for liveness detection on the fly, based on photoplethysmography (rPPG) estimated pulses. It simulates a setting where a liveness verification step is conducted during an user authentication phase in a mobile phone. To keep the application responsive, liveness needs to be detected on an atmost 5s video, in a non-intrusive manner.
For more details and dataset statistics, please refer to our paper: Temporal Convolutional Networks for Robust Face Liveness Detection , R. Padnevych, D. Carmo, D. Semedo, J. Magalhães, Proceedings of the 10th Iberian Conference on Pattern Recognition and Image Analysis (ibPRIA), 2022. (Full paper)
The dataset can be downloaded here.
The dataset zip file is structured in two folders, each corresponding to a class:
- Real - Samples belonging to the "Real" class, i.e. genuine pulses.
- Fake - Samples belonging to the "Fake" class, i.e. pulses corresponding to presentation attacks.
Each sample csv is structured as output_<First Frame Instant>-<Last Frame Instant>_000.mp4 (<Video id>).csv. For example, the file output_004-009_000.mp4 (32).csv denotes a pulse sample extracted from video 33, starting at second 4 and ending at second 9.
Each sample (.csv file) line (corresponding to an instant t) has the following properties:
- Timestamp - Relative sample timestamp t;
- BPM - Estimated Beats Per Minute;
- Signal stable? - Flag that indicates if the signal is stable
- Valid pulse? - Flag that indicates if the pulse is valid;
- Number of peaks - Number of peaks in the signal;
- Nr. peaks > 2 - Flag that indicates if the Number of Peaks is > 2;
- Peaks between 40-240 bpm? - Flag that indicates if the peaks correspond to a bpm in the inveral 40-240;
- avg(peak amplitudes) - Average peak amplitudes;
- s(peak amplitudes) - Standard deviation of peak amplitues;
- s(peak amplitudes) < 0.5? - Flag that indicates if standard deviation of peak amplitues is lower than 0.5;
- All peak amplitudes > 60? - Flag that indicates if all peak amplitues are greater than 60;
- s(peak distances) - Standard deviation of peak distances;
- s(peak distances) < 0.5?- Flag that indicates if the standard deviation of peak distances is lower than 0.5;
- Raw signal - Raw value of the signal;
- Pulse signal - Pulse signal value.
If you found this dataset useful, please cite the following paper:
@InProceedings{10.1007/978-3-031-04881-4_21,
author="Padnevych, Ruslan
and Carmo, David
and Semedo, David
and Magalh{\~a}es, Jo{\~a}o",
editor="Pinho, Armando J.
and Georgieva, Petia
and Teixeira, Lu{\'i}s F.
and S{\'a}nchez, Joan Andreu",
title="Temporal Convolutional Networks for Robust Face Liveness Detection",
booktitle="Pattern Recognition and Image Analysis",
year="2022",
publisher="Springer International Publishing",
address="Cham",
pages="255--267",
isbn="978-3-031-04881-4"
}