Closed zaie closed 4 years ago
Thanks for the feedback about the quality of temporal annotations. While we have conducted two rounds of cleaning already, we are always happy to incorporate contributions from the community that improve the dataset quality. Please send your feedback, and potential annotation fixes, to the dataset maintainers via email.
The temporal action annotations of Activitynet are so poor that I can not trust them. I believe the data needs to be relabelled. For example, temporal annotations in video 'OiL6Aj0gC14', 'onBAyGhqubg' and 'optJ47P_5Ys'. Is there any reason for that?