UT-Interaction

UT-Interaction 数据集包含 6 类人机交互的连续执行视频：握手、指向、拥抱、推、踢和拳。提供了这些交互的真实标签，包括时间间隔和边界框。总共有 20 个长度在 1 分钟左右的视频序列。每个视频每次交互至少包含一次执行，因此平均每个视频执行 8 次人类活动。视频中出现了几名具有超过 15 种不同服装条件的参与者。视频采用720*480分辨率，30fps拍摄，视频中人物身高约200像素。

The UT-Interaction Dataset contains continuous execution videos of 6 types of human-computer interactions: handshakes, pointing, hugs, pushes, kicks, and punches. Ground truth labels for these interactions are provided, including temporal intervals and bounding boxes. There are a total of 20 video sequences, each with a duration of approximately 1 minute. Each video contains at least one execution of each interaction type, resulting in an average of 8 human activity executions per video. Several participants appear in the videos, with over 15 different clothing conditions. The videos are shot at a resolution of 720×480 and 30 frames per second (fps), with the height of the human figures in the videos being approximately 200 pixels.