Comparison of labeled video datasets for action recognition
Compiled by Kevin Murphy and Alex Gorban.
Last updated 29 July 2014.
What do the headers in the table mean?
- Year: year dataset was last update
- Accuracy: State of the art performance of best (whole video) classifier.
Provide a citation in the References column.
- Avg dur: Median duration of each video (s=seconds, m=minutes, h=hours).
- Tot# hours: total duration of all videos in dataset (hours).
#video level: number of types of video-level action labels.
- #frame level: number of types of frame-level action labels.
- #Act/vid: median number of (low or high level) actions per video.
- #Vid/act: median number of videos per action.
- #Ppl/vid: median number of people per video.
- Tot# Ppl: total number of people in the dataset.
- Youtube: Does it contain videos from youtube?
- Lab: Does it contain videos from controlled (laboratory) conditions?
- Sports: Does it contain videos about sports?
- Cooking: Does it contain videos about cooking?
- Movies: does it contain videos from movies or TV shows?
- Segmented: are temporal segmentations of the actions provided?
- Frame tags: are there tags at a frame level? (All videos are assumed to
labels at the whole video label.)
- NLP: are there natural language text annotations?
- Objects: are there annotations (eg. bounding boxes) for objects?
- Skeleton: is there annotation for the human skeleton pose?
- #Cameras: number of cameras uses in recording.
- Static cam: is camera stationary?
- Moving cam: is camera moving?
- Stereo: is camera stereo?
- POV: is camera point of view (head mounted)?
- RGDB: does camera record depth info?
- Resoln: image resolution
- FPS: Frames per second
- Video-level action names
- Frame-level action names (if available)
- References: references to relevant papers using keys
listed in a separate table below.
If you want your dataset to be listed, please fill out
and send me an email (email@example.com)
to notify me that you submitted a form.
(This is because I need to manually
transfer the data from your form to the spreadsheet below, for various reasons.)
If you want to fix any errors in the description of existing datasets, you can edit
the spreadsheet shown below
directly (please be careful!)
Other relevant links
If you want to add new papers to this list,
you can edit
the spreadsheet shown below directly (be careful!)