STAIR Actions

A Large-Scale Video Dataset of Everyday Human Actions

January 30, 2019: STAIR Actions v1.1 is released!


STAIR Actions is a video dataset consisting of 100 everyday human action categories.   Each category contains around 900 to 1800 trimmed video clips. Each clip lasts 5 to 6 seconds. Clips are taken from YouTube video or made by crowdsource workers.

spec

SPEC (v1.1)

Train Validation Total
# of video clips 99,478 10,000 109,478
Avg. # of video clips
per category		 1055.44 100 1,094.78
S.D. over categories 146.87 0 146.87
Min-Max 802-1,745 100-100 902-1,845

Comparison with other datasets (v1.1)

# of
categories		 # of
videos		 Avg. # of
videos per category		 # of everyday
home actions		 Overlapping w.r.t. STAIR Actions
STAIR Actions 100 109,478 1,094.78 100 ---
ActivityNet 203 849hrs 137 105 49 categories match 28 SA categories
Kinetics 400 300,000 750 215 122 categories match 54 SA categories
AVA 80 57,600 min=2,
max=45,790		 80 40 categories match 43 SA categories

List of 100 Actions

hi
download

Terms of Use

By downloading “STAIR Actions” (the Dataset), you agree to the following terms.

  • You will use the Dataset only for the purpose of AI research.
  • You will NOT distribute the Dataset or any parts thereof.
  • You will treat people appearing in the Dataset with respect and dignity.
  • The Dataset comes with no warranty or guarantee of any kind, and you accept full liability.

License

Assuming you agree the terms of use,

  • STAIR Actions metadata file (.json) is licensed under Creative Commons Attribution 4.0 International license (CC BY 4.0)  - (https://creativecommons.org/licenses/by/4.0/legalcode)
  • The type field of STAIR Actions medata file (.json) indicates license condition of individual movie file by one character, where C, X, Y indicate Creative Commons CC0 Public Domain Dedication, ordinary YouTube license, CC BY, respectively.  - (https://creativecommons.org/publicdomain/zero/1.0/deed.en)

Download

Download from GitHub

If you don’t agree the terms of use, please do not download.

publications

Our Publication list

  • Yuya Yoshikawa, Jiaqing Lin, Akikazu Takeuchi, “STAIR Actions: A Video Dataset of Everyday Home Actions,” arXiv:1804.04326, Apr. 2018. [PDF]
  • 吉川友也, 竹内彰一, “家庭やオフィス内の動作認識用大規模動画データセットの構築,” 平成29年度人工知能学会全国大会 (JSAI2017), 2017.

Bibliography

If you use STAIR Actions dataset, please cite the following paper.

@article{stairactions2018,
    author = {Yoshikawa, Yuya and Lin, Jiaqing and Takeuchi, Akikazu},
    title = {STAIR Actions: A Video Dataset of Everyday Home Actions},
    journal={arXiv preprint arXiv:1804.04326},
    url = {http://arxiv.org/abs/1804.04326},
    year = {2018}
}
About us

Owner



Chiba Institute of Technology
Software Technology and Artificial Intelligence Research Laboratory (STAIR Lab)

Contributors

  • Akikazu Takeuchi, STAIR Lab
  • Yuya Yoshikawa, STAIR Lab
  • Teruhisa Kamachi, Mokha Inc.
  • Hiroshi Ogino, Scaleout Inc.
  • Takahiro Tsutsumi, PlayNext Japan Inc.
  • Student volunteers in Chiba Institute of Technology, and many crowd workers

Acknowledgment

This work is supported by NEDO, Japan.

Contact

Akikazu Takeuchi
takeuchi at stair.center

Software Technology and Artificial Intelligence Research Laboratory (STAIR Lab), Chiba Institute of Technology, Japan.