The ViTT dataset consists of human produced segment-level annotations for 8,169 videos. Of these, 5,840 videos have been annotated once, and the rest of the videos have been annotated twice or more.
11 PAPERS • 2 BENCHMARKS