Dear organizers,
I have two questions about the OSLWL track.
1. Are we allowed to train unsupervised models on the validation set data for the first phase?
2. Is it correct to assume that there is at most one ground truth annotation per video in the validation set?
Thank you
Posted by: vqaf @ May 13, 2022, 6:57 a.m.Dear participant,
I will comment below.
1) Yes, of course.
2) No. As mentioned in the dataset description "As this Track tries to simulate retrieval after weak labeling, around 10% of the video files will not contain any instance of the sign to be retrieved (simulating a misalignment of subtitles and signing, a failure in the weak labeling pipeline or a word not being signed)."
Best