HAA500: Human-Centric Atomic Action Dataset with Curated Videos

Jihoon Chung, Cheng-hsin Wuu, Hsuan-ru Yang, Yu-Wing Tai, Chi-Keung Tang
Abstract

We contribute HAA500, a manually annotated human-centric atomic action dataset for action recognition on 500 classes with over 591k labeled frames. Unlike existing atomic action datasets, where coarse-grained atomic actions were labeled with action-verbs, e.g., "Throw", HAA500 contains fine-grained atomic actions where only consistent actions fall under the same label, e.g., "Baseball Pitching" vs "Free Throw in Basketball", to minimize ambiguities in action classification. HAA500 has been carefully curated to capture the movement of human figures with less spatio-temporal label noises to greatly enhance the training of deep neural networks. The advantages of HAA500 include: 1) human-centric actions with a high average of 69.7% detectable joints for the relevant human poses; 2) each video captures the essential elements of an atomic action without irrelevant frames; 3) fine-grained atomic action classes. Our extensive experiments validate the benefits of human-centric and atomic characteristics of HAA, which enables the trained model to improve prediction by attending to atomic human poses. We detail the HAA500 dataset statistics and collection methodology, and compare quantitatively with existing action recognition datasets.

Sample Video : Push Up
Dataset
HAA500 v1.1   Download

Zip file contains 500 txt files for each class of HAA500 and corresponding videos.

pushup.txt

https://www.youtube.com/watch?v=IODxDxX7oi4,3.37,7.08,1,1
https://www.youtube.com/watch?v=AhdtowFDKT0,182.71,185.11,0,1
https://www.youtube.com/watch?v=cABuP6ZdUno,7.18,8.58,0,1
https://www.youtube.com/watch?v=FaIpD_zfrJI,50.51,52.53,0,1
https://www.youtube.com/watch?v=TlQ8txalLYg,125.48,127.08,0,1
https://www.youtube.com/watch?v=DpLIDPzxZ8o,219.42,220.53,0,1
https://www.youtube.com/watch?v=4hn3XzCEHyE,201.24,203.32,0,1
https://www.youtube.com/watch?v=Eh00_rniF8E,54.09,57.91,0,2
https://www.youtube.com/watch?v=XIu1A2MtTWs,121.67,122.75,0,1
https://www.youtube.com/watch?v=eUBc5uKg1Ik,19.82,21.24,0,1
https://www.youtube.com/watch?v=AM5AjrfhlEw,43.6,44.76,0,1
https://www.youtube.com/watch?v=rUcvA5351kk,201.9,203.26,0,1
...
First 16 lines are training set, 17th is validation set, rest are test set.
txt files are formated as:
youtube_url,start_time,end_time,is_camera_moving,num_of_dominant_figure


HAA500 v1.0   Download
While HAA500 v1.0 was used in our experiments, it has been updated to v1.1. Please consider using the latest version.

Citation
@article{haa500,
    title={HAA500: Human-Centric Atomic Action Dataset with Curated Videos},
    author={Jihoon Chung and Cheng-hsin Wuu and Hsuan-ru Yang and Yu-Wing Tai and Chi-Keung Tang},
    year={2020},
    eprint={2009.05224},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}