PladsElsker/openasl-dwpose

Dataset

This dataset is a derivative of the OpenASL dataset, but contains pose estimations instead of the original RGB videos.
Pose keypoints are predicted using DWPose, and each sample is stored as a pickle file.
All predictions have been made on the original video size.

Field	Type	Description
poses	list	Pose data inferred from the video frames by DWPose
video_frames_per_second	float	Frame rate of the video
video_duration	float	Duration of the video in seconds
video_frame_count	int	Amount of frames in the video
video_width	int	Width of the video
video_height	int	Height of the video
file_stem	str	File name stem (without extension)
raw_text	str	Raw textual description
tokenized_text	str	Tokenized version of the raw text
split	str	Dataset split (e.g., train/val/test)
yid	str	Youtube video ID
start	str	Start timestamp of the Youtube video segment
end	str	End timestamp of the Youtube video segment

References

This dataset is derived from OpenASL repository, but is neither maintained nor endorsed by its original authors.