Human Pose Estimation


The database was collected for our ICCV'11 (workshop) paper Putting the pieces together: Connected Poselets for Human Pose Estimation. The dataset contains 345 training images and 347 test images of 3 people. For each frame, a depth image of 640x480 pixels and the annotation is provided. Ground truth is in the form of the (x,y) coordinates (i.e.2D image plane locations) of each of the 10 tracked joints.

The database is made available for research purposes only. Please cite our work whenever publishing anything directly or indirectly using the data:


Putting the pieces together: Connected Poselets for Human Pose Estimation

@InProceedings{Holt2011Putting,
  author = {B. Holt and E-J. Ong and H. Cooper and R. Bowden},
  title = {Putting the pieces together: Connected Poselets for Human Pose Estimation},
  booktitle = {1st IEEE Workshop on Consumer Depth Cameras for Computer Vision (ICCV'11)},
  year = {2011},
  month = nov
}

Files:

Data (9.3 MB, .zip compressed)

Source code (.zip compressed)