Published December 6, 2024 | Version 1.0.0

Egocentric video clips of laboratory actions

Description

A collection of video clips in MPEG-4 format related to experimental laboratory actions. Each clip lasts 4 seconds at 4 frames per second and is recorded from an egocentric point of view. The clips contain no personally identifiable information. The collection comprises two datasets that are described below. Further details of the datasets and related Python code are available as resources linked to this data entry.

 

  • lab-actions_annotations_splits.csv
    CSV file containing the filename of each clip along with the corresponding action label and split assignment (testing, training or validation) for the Lab Actions dataset
  • lab-motion_annotations_splits.csv
    CSV file containing the filename of each clip along with the corresponding action label and split assignment (testing, training or validation) for the Lab Motion dataset

 

  • lab-actions
    Folder containing the video clips of the Lab Actions dataset (4529 video clips at 4 s / 4 fps)
    • V
      Folder containing the original video modality of the lab-actions dataset
    • VG
      Folder containing the overlaid gaze coordinate on the original video modality of the lab-actions dataset
    • M
      Folder containing clips of the lab-actions dataset converted to segmentation masks 
    • D
      Folder containing clips of the lab-actions dataset converted to depth signals
  • lab-motion
    Folder containing the video clips of the Lab Motion dataset (2014 video clips at 4 s / 4 fps)

 

Files

lab-actions_annotations_splits.csv

Files (33.6 GB)

Name Size
md5:aaa717d7d3e93e9817bc1fae68c54348
319.8 kB Preview Download
md5:cafe554d7e4eb6cda2bbee11650b4a32
147.7 kB Preview Download
md5:a5b931b33d4b84b9f5e18e85db8d32f9
33.6 GB Preview Download

Additional details

Software

Repository URL
https://github.com/IBM/lab-activity-recognition
Programming language
Python