This dataset was made for experimental testing and no ground truth labeling is available. The dataset contains 4 videos, two indoor and two outdoor. Two different persons appear in the videos and only one person is visible at a time. Each video contains a sequence for training a background model. Additionally, 25 training images of each person's head is available. Five images for each head orientation ordered as:

images 1-5:	left 90
images 6-10:	left 45
images 11-15:	front 0
images 16-20:	right 45
images 21-25:	right 90

All images are hand segmented and cropped to fit the head.

The videos are recorded with different cameras and resolution. Videos are in encoded in Xvid format, and the images are given in uncompressed .tifs.