Human Skeleton Graphs
- Description:
Kinetics: a large-scale human action dataset with 300000 videos clips in 400 classes. Those video clips are from YouTube with a great variety. The raw Kinetics doesn't contain skeleton data, and [2] uses OpenPose toolbox to generate skeleton with 18 joints on every frame. Kinetics-Skeleton contains 240000 clips of training data and 20000 clips of test data.
NTU-RGB+D: a large and widely used action recognition dataset with 56000 action clips in 60 classes. These clips are performed by 40 volunteers captured in a constrained lab environment, with three camera views recorded simultaneously. The dataset provides 3D joint locations of each frame and 25 joints for each subject. - Statistics:
- Acknowlegement:
Kinetics: Kay, W., Carreira, J., Simonyan, K., Zhang, B., Hillier, C., Vijayanarasimhan, S., ... & Zisserman, A. (2017). The kinetics human action video dataset. arXiv preprint arXiv:1705.06950.
NTU-RGB+D: Shahroudy, A., Liu, J., Ng, T. T., & Wang, G. (2016). Ntu rgb+ d: A large scale dataset for 3d human activity analysis. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1010-1019).
| Name | Type | #Graphs | #Nodes | #Edges | Attributed | Directed | Weighted | Signed | Homogeneous | Spatial | Temporal | Labels |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Human Skeleton Graphs (Kinectics) | Skeleton Graphs | 260,000 | 18 | 17 | NO | NO | NO | NO | YES | 2D | YES | YES |
| Human Skeleton Graphs (NTU-RGB+D) | Skeleton Graphs | 56,000 | 25 | 24 | NO | NO | NO | NO | YES | 3D | YES | YES |