MultiSports is a spatiotemporal behavior detection dataset. The dataset has the following features:

• Multiplayer scene:In the same scene, different people perform different fine-grained actions, reducing the information provided by the background.
• Action boundaries are well defined.
• Relatively fine-grained action categories.

By selecting 4 sports categories, the dataset collects 3,200 video clips to construct the MultiSports v1.0 dataset. The dataset also uses 902k bounding boxes to annotate 37,701 action instances. The dataset has the important characteristics of high diversity, dense annotation and high quality.