Toyota Smarthome Trimmed is a dataset designed for activity classification tasks, involving 31 activities. The videos are based on a single activity clip, resulting in a total of 16,115 RGB+D short video clips. The activity behaviors are natural and not controlled by the experiment.

Therefore, this dataset poses great challenges: the differences between categories are large, the categories are very unbalanced, the activities have similar movements, and the durations vary greatly. The labels include both coarse and fine labels.