This set of 305 images is scaled to contain people of roughly 150
pixels in height. There is a .mat file of manually-labeled joint
positions for each image. The first 100 images were used for training, 
while the last 205 were used for testing.

Many thanks to Navneet Dalal, Mun Wai Lee, Greg Mori, and Jiayong 
Zhang for generously making their datasets available.

-Deva Ramanan
 ramanan@tti-c.org


