OVT-B

OVT-B, a large-scale Open-Vocabulary multi-object Tracking Benchmark containing 1,973 videos and 637,608 annotated objects from 1,048 categories, surpassing the diversity of all current MOT datasets. OVT-B also includes some attributes especially for the MOT task, e.g., the scenarios of out-of-view, fast motion, mutual occlusion, and objects with various sizes, shapes, etc.