##About the dataset##
This dataset was developed at the University of Barcelona (Human Pose Recovery and Behaviour Analysis group). 
We designed a dataset of 90 images consisting of 9 people who have been recorded in video by a digital camera in different positions and performing different gestures. The process for treating these videos has been done by dividing each video onto a sequence of frames which are processed for detecting limbs. Then, for each frame different limbs are manually labelled using masks as follows:

0->Background
1-> head
2-> trunk
3-> left hand
4-> right hand
5-> left forearm
6-> right forearm
7-> left arm
8-> right arm
9-> left foot
10-> right foot
11-> left leg
12-> right leg
13-> left thigh
14-> right thigh

##Directories##
images: Contains 90 source images.
masks_ground-truth: Contains the masks for the limbs using previous annotation. 

##Data set designers##
Juan Carlos Ortega
Daniel Sanchez Abril
Miguel Angel Bautista
Sergio Escalera
