Description
We propose a multi-modal video deception detection dataset, named PV3D, designed with a simulated theft experiment scenario. The dataset includes video, audio, heart rate, respiration, blood volume pulse (BVP) signals, and blood oxygen saturation data. A total of 101 students participated, among which 48 lied and 53 told the truth. Video data were captured using Logitech C920 HD Pro cameras at 640×480 resolution and 30 FPS. BVP, heart rate, and respiration signals were recorded synchronously using Contec CMS50E pulse oximeters and HKH-11C respiratory sensors. To our knowledge, this is the first dataset that can be simultaneously used for video deception detection, remote heart rate monitoring, and remote respiration monitoring. For detailed experimental procedures, please click here.
Download
Download the request form, print it, sign it, and send it to rcsong@hfut.edu.cn. Once received, we will share a download link. The dataset size is approximately 639 GB. Each folder contains all data for one subject.
Manual
S*
│── resp
│ └── XXX.csv
│── SPO2
│ ├── XXX.csv
│ └── XXX_wave.csv
│── XXX.avi
└── XXX.m4a
Each subject corresponds to a folder S*, which includes resp, SPO2 folders, video and audio. The XXX.csv file in the resp folder contains respiratory reference data. The XXX.csv file in the SPO2 folder contains heart rate and blood oxygen saturation data, while XXX_wave.csv contains blood volume pulse signals. *∈{1,2,3...101}.
Citation
If you use this dataset in your research, please cite the following publication:
S. Gao, L. Chen, Y. Fang, S. Xiao, H. Li, X. Yang, and R. Song, "Video-Based Deception Detection via Capsule Network With Channel-Wise Attention and Supervised Contrastive Learning," IEEE Open J. Comput. Soc., vol. 5, pp. 660–670, 2024.