Stanford 40 Actions
                                                           ---- A dataset for understanding human actions in still images

Introduction
The Stanford 40 Action Dataset contains images of humans performing 40 actions. In each image, we provide a bounding box of the person who is performing the action indicated by the filename of the image. There are 9532 images in total with 180-300 images per action class.

        

Download
Please download the dataset using the links below:

• Images: 297.6MB;
• Annotation in Matlab: Bounding boxes of humans, one box per image;
• Annotation in XML: Bounding boxes of humans, one box per image. You can use the parser provided by PASCAL VOC to parse the XML files;
• Image sets: suggested train/test split of the images;
• All: you can download all above files here.

Citation
Please cite this paper if you want to use this dataset in your research:

• B. Yao, X. Jiang, A. Khosla, A.L. Lin, L.J. Guibas, and L. Fei-Fei. Human Action Recognition by Learning Bases of Action Attributes and Parts. Internation Conference on Computer Vision (ICCV), Barcelona, Spain. November 6-13, 2011.   [PDF]   [BibTex]

Related Dataset
Other datasets of human actions in still images:

• People Interacting with Musical Instruments (PPMI): Images of people interacting with 12 different musical instruments. For each instrument, there are images of human playing the instrument or holding the instrument but not playing.
• PASCAL VOC Action Images: Images of humans performing 10 different actions.
• Sports Dataset: Images of six different sports with 30 images per class.

Contact
Please feel free to email to bangpeng@cs.stanford.edu if you have any question or suggestion about this dataset.