Closed patrick-tssn closed 2 weeks ago
Good!!! I have checked the data and it's the only one bug data with start >= end
. I will fixed it later. Considering it only affect the results by 1/200=0.005, you can still use the original version.
Thanks for your report again. I have updated the JSON files.
I found the BUG is caused since the video duration is actually 6s
, while the accurate_start=0, accurate_end=8
. To fix it, I simply use start=0, end=8
. In my code, it will change the start
and end
when exceed the boundary.
Thank you for your work. I've discovered an issue within the MVBench, specifically in the object_interaction.json file. It appears that there is a case where the start time is identical to the end timestamp.