Leeds Sports Pose Dataset
Sam Johnson and Mark Everingham

Download (ZIP file)

Athletics Badminton Baseball Gymnastics Parkour Soccer Tennis Volleyball

This dataset contains 2000 pose annotated images of mostly sports people gathered from Flickr using the tags shown above. The images have been scaled such that the most prominent person is roughly 150 pixels in length. Each image has been annotated with 14 joint locations. Left and right joints are consistently labelled from a person-centric viewpoint. Attributions and Flickr URLs for the original images can be found in the JPEG comment field of each image file.

Original scale image dataset (ZIP file)

Dataset Format

The ZIP archive contains images in two folders:
images/ - containing the original images
visualized/ - containing the images with poses visualized
The file joints.mat is a MATLAB data file containing the joint annotations in a 3x14x2000 matrix called 'joints' with x and y locations and a binary value indicating the visbility of each joint.
The ordering of the joints is as follows:

  1. Right ankle
  2. Right knee
  3. Right hip
  4. Left hip
  5. Left knee
  6. Left ankle
  7. Right wrist
  8. Right elbow
  9. Right shoulder
  10. Left shoulder
  11. Left elbow
  12. Left wrist
  13. Neck
  14. Head top


If you use this dataset please cite the following work:

Sam Johnson and Mark Everingham
"Clustered Pose and Nonlinear Appearance Models for Human Pose Estimation"
In Proceedings of the 21st British Machine Vision Conference (BMVC2010)
   title = {Clustered Pose and Nonlinear Appearance Models for Human Pose Estimation},
   author = {Johnson, Sam and Everingham, Mark},
   year = {2010},
   booktitle = {Proceedings of the British Machine Vision Conference},
   note = {doi:10.5244/C.24.12}

Experimental Protocol

In our BMVC paper the dataset was split into two parts for training and testing. The first 1000 images (im0001.jpg to im1000.jpg) were used for training and strictly ALL parameter selection. The second 1000 images (im1001.jpg to im2000.jpg) were used for testing. The evaluation method used is the Percentage Correct Parts measure (Estimated part end points must be within 50% of part length from the ground truth part end points. See [Ferrari et al., 2008]). The results obtained using our method are as follows:

Total Torso Upper Leg Lower Leg Upper Arm Forearm Head
Note: columns with two numbers show results for left and right limbs respectively.

Example Images