KTH Action Recognition Dataset

The KTH dataset is one of the most standard action recognition datasets, containing six actions: walking, jogging, running, boxing, waving, and clapping.

To showcase the nuances of performance, each action was performed by 25 different actors, and the background environment for each actor and each action was systematically varied. The environment variations included: outdoors, outdoors of different sizes, outdoors with different clothing, and indoors. These variations tested each algorithm's ability to recognize actions independent of the background, the actor's appearance, and the actor's proportions.