SteveZeyuZhang/CrossTask_COIN_3rd-party

CrossTask and COIN — Third-party Video Collection

This is an unofficial third-party collection of CrossTask and COIN videos. We do not own the source videos or datasets and do not claim any rights to them. All rights remain with the original authors, video creators, and respective rights holders. This collection is not affiliated with or endorsed by the original dataset authors.

CrossTask

Coverage is measured against the 2,750 annotated primary-task videos in the official CrossTask release. It does not describe coverage of the full collection of primary and related tasks.

Official CrossTask release

MaLoW subset

• Directory: CrossTask/MaLoW.
• Coverage: 1,649 / 2,750 videos (59.96%), spanning all 18 primary tasks.
• Files: three TAR archives, approximately 14.62 GB in total.
• These are prepared videos. One inspected video is 640 × 360 at 1 FPS; the subset should not be assumed to preserve the original frame rate. Encoding has not been checked for every video.

MaLoW-Instruct-Videos

WorldPrediction subset

• Directory: CrossTask/WorldPrediction.
• Coverage: 202 / 2,750 videos (7.35%), spanning 17 primary tasks.
• All 202 videos belong to the official localization validation set: 202 / 360 videos (56.11%) of that set.
• Files: 202 MP4 videos, approximately 16.39 GB in total.
• One inspected video is 1920 × 1080 at 24 FPS. Encoding has not been checked for every video.

WorldPrediction

The two CrossTask subsets share 132 video IDs. Together they cover 1,719 / 2,750 unique videos (62.51%). Their encodings differ, so files sharing a source ID are retained separately.

COIN

• Directory: COIN.
• Coverage: 9,876 / 11,827 official video IDs (83.50%).
• Files: 9,877 MP4 files, approximately 343.78 GB in total. Coverage counts distinct filenames matching official video IDs, rather than all MP4 files.

COIN video source

Official COIN annotations

Rights and original authors

Original dataset terms and the rights of the respective video creators remain applicable. This mirror does not grant additional rights or permissions. For permission to use the data, licensing questions, access to original videos, or missing videos, please contact the original authors of the CrossTask and COIN papers through their official project pages.

CrossTask paper

CrossTask project and author contact

COIN paper

COIN project and author contact