Cross-Domain Deception Dataset
- Citation Author(s):
-
Sayde King
(University of South Florida)
Tempestt Neal
(University of South Florida)
- Submitted by:
- Sayde King
- Last updated:
- DOI:
- 10.21227/b7hf-de49
- Data Format:
16 views
- Categories:
- Keywords:
Abstract
The Cross-Domain Deception Dataset (CD3) contains frame-level features extracted from video data using OpenFace and OpenPose to support research in deception detection through facial expressions, facial action units, body and hand gestures, and gaze coordinates. Using a commercial off-the-shelf laptop and Microsoft Teams, we collected video data of 45 participants completing mock interviews where they answered questions related to biographical information, academic success, and well-being across two sessions. The dataset includes features from 1270 truthful and 587 deceptive clips of participant responses to interview questions. CD3 enables future research exploration to inform the development of robust deception models for accurate deception detection in novel domains such as well-being.
Instructions:
1. Dataset Contents
Each sample includes the following 983 frame-level features:
- 8 Gaze Features: Gaze direction (left & right eyes: 2 vectors x 3D = 6), gaze angle (pitch, yaw = 2)
- 136 2D Facial Landmarks: 68 points x 2D (x,y)
- 204 3D Facial Landmarks: 68 points x 3D (x,y,z)
- 112 2D Eye Landmarks: 56 points x 2D (x,y,)
- 168 3D Eye Landmarks: 56 points x 3D (x,y,z)
- 6 Head Pose Features: Translation (x,y,z = 3), Rotation (pitch, yaw, roll = 3)
- 40 Face Shape Parameters: 40 PCA-based shape parameters
- 35 Facial Action Units: Presence (18 AUs), Intensity (17 AUs)
- 50 Body Keypoints: 25 keypoints x 2D (x,y)
- 84 Hand Keypoints: 21 keypoints x 2D (x,y) x 2 hands
- 140 Face Keypoints: 70 keypoints xx 2D (x,y)
2. File Format
- The data file is provided in .csv format
- Each row in the CSV represents a frame from a participant's response.
3. Educational and Research Use
This dataset is made available for educational and research purposes. It can support coursework, capstone projects, graduate theses, or experimentation with deception detection systems.
