KTH (KTH Action dataset)

2004 年，KTH 皇家理工学院开始努力创建一个重要且公开可用的动作识别数据集。KTH 数据集是最标准的数据集之一，它包含六个动作：步行、慢跑、跑步、拳击、挥手和拍手。考虑到表演的细微差别，每个动作由 25 个不同的人执行，并且针对每个演员的每个动作系统地更改设置。设置变化包括：室外（s1）、室外有尺度变化（s2）、室外不同衣服（s3）、室内（s4）。这些变化测试了每个算法识别独立于背景、演员外观和演员规模的动作的能力。

In 2004, KTH Royal Institute of Technology launched an initiative to develop a prominent, publicly accessible action recognition dataset. The KTH Actions Dataset is among the most widely recognized standard benchmark datasets for action recognition, containing six core actions: walking, jogging, running, boxing, hand waving, and clapping. To account for subtle variations in performance, each action is performed by 25 distinct subjects, and the recording settings are systematically modified for every action of each actor. The setting variations include: outdoor (s1), outdoor with scale changes (s2), outdoor with different attire (s3), and indoor (s4). These variations are designed to evaluate an algorithm’s capacity to recognize actions independently of background, actor appearance, and actor scale.