DATA SETSREEDME

Multiview Action DataSet

The color videos are saved as avi movies compressed with XviD codec.
They have resolution of 640 x 520, but actually the 20 first and last lines are redundant.
So the resolution is 640 x 480.

All the actors perform the following actions (coded in the ground truth as numbers):
check watch – 1
cross arms – 2
scratch head – 3
sit down – 4
get up – 5
turn around – 6
walk – 7
wave – 8
punch – 9
kick – 10
point – 11
pick up – 12
throw – 13

The actions are performed continiously – there is not necessarily a pause between them.

The actions are filmed with 8 cameras spread more or less evenly around the actor.
They cover all 360 degrees and placed near the ceiling of the room.

The forground masks can be obtained using background subtraction with the provided backgrounds.
However, the resulting masks are not of good quality.