MeViS

MeViS is a large-scale dataset for motion expressions guided video segmentation, which focuses on segmenting objects in video content based on a sentence describing the motion of the objects. The dataset contains numerous motion expressions to indicate target objects in complex environments.