OpenTTGames Table Tennis Dataset

A dataset containing 38,752 training samples, 9,502 validation samples, and 7328 test samples. The annotations of the dataset include the position coordinates of the table tennis ball, the segmentation template, and the status information of the game.

OpenTTGames can be used for computer vision analysis related to table tennis: ball detection, semantic segmentation (people, table, scoreboard, as well as ball coordinates and game events).

The dataset consists of annotated full HD videos of table tennis matches recorded at 120 fps by an industrial camera. The annotations are manually added and include the frame number and corresponding target of a specific frame; events in the game, such as ball bounces, net hits, and other targets; as well as ball coordinates and segmentation masks.