Datasets:
Potential duplicate samples
#4
by
geomlyd
- opened
Some checks in my code seem to have come across some samples that look a lot like duplicates. For example, if one searches for the video "NYHD7" (e.g. in HF's Viewer) in the action_sequence subset, two results come up, and they both contain the exact same question, video, as well and start and end (presumably seconds). The only thing that sets them apart is the actual question id (Sequence_T3_448 vs Sequence_T4_363). This is also not the only such case in the dataset (my code seems to have spotted about 24 of them). Is this intentional?
Thank you for the feedback. We've revised the annotations to eliminate duplicate samples in the Action Sequence and Unexpected Action tasks, making additional improvements to some annotations.