News

  • The 7th Large-scale Video Object Challenge in conjunction with ICCV 2025 is ongoing! Checkout
  • Due to the numerous requests from the community, we now have released the ground truth labels for validation sets of VOS2019, VIS[2019, 2021, 2022] in the corresponding codalab download links!
  • Due to maintainance issues of the old Codalab website, we have migrated the VOS and VIS evaluation servers of the 2019 challenge to the new codalab site. Details

What is YouTube-VOS

YouTube-VOS is the first large-scale benchmark that supports multiple video object segmentation tasks.

  • Semi-supervised Video Object Segmentation
  • Video Instance Segmentation
  • Referring Video Object Segmentation

It also has the following features.

  • 5000+ high-resolution YouTube videos
  • 90+ semantic categories
  • 7800+ unique objects
  • 190k+ high-quality manual annotations
  • 340+ minutes duration

Research papers

Please cite the following papers if you find our dataset is useful.

Semi-supervised video object segmentation

Video instance segmentation

Referring Video Object Segmentation

Dataset examples