-
Notifications
You must be signed in to change notification settings - Fork 28
Open
Description
Hi,
Description:
I've encountered inconsistencies in the number of extracted frames when comparing different extraction methods with the pre-extracted frames provided by the dataset.
For example P01_04.MP4:
- Pre-extracted frames (provided): 6320 frames
- Using torchvision.io.read_video: 6308 frames
- Using ffmpeg with 60fps (as suggested in EPIC-KITCHENS 55 repo): 6314 frames
ffmpeg -i "P01_04.MP4" -vf "scale=-2:256" -q:v 4 -r 60 "P01_04/frame_%010d.jpg"
Could you share the exact method used to extract the pre-extracted frames provided in the dataset?
Note: I have already checked the md5 hash of tar file with md5.csv and it's matched.
Thank you very much!
Metadata
Metadata
Assignees
Labels
No labels