This is the repository for the dataset "Multiple Human Interaction Understanding (MHIU)."
MHIU is a dataset for the human interaction understanding task. It includes 40 video clips capturing multiple human interactions performed simultaneously by different human pairs. Each video lasts 1 to 2 minutes. The data is recorded in RGB and depth format with 2 distinct views using ZED 2i cameras. The images are PNG and have the resolution of 1920 x1080. There is a total of 11 interaction classes. The human bounding boxes, individual action labels, pairwise interaction labels, and tracking IDs are provided for all subjects in the middle frames of every three consecutive frames of each video.
| Class ID | Class Name |
|---|---|
| 0 | No action |
| 2 | Hug |
| 3 | Handshake |
| 4 | Exchange things |
| 5 | High-five |
| 6 | Cheer&drink |
| 7 | Carry objects |
| 8 | Help sit down/stand up |
| 9 | Greet |
| 10 | Study |
The MHIU dataset is made available only for academic research purposes. Any form of commercial use is strictly prohibited. Without permission from AIS Lab, redistributing the dataset or creating derivative datasets based on MHIU is considered illegal. Images from the dataset can only be used in academic publications, presentations, or related research and academic documents.
To download the dataset, please contact the emails: lee.aislab@gmail.com and phuong.aislab@gmail.com.