This dataset of hand manipulation depicts an actor performing several steps in a typical assembly tasks using five different hand-tools applied over a wooden plank. The steps of the assembly task are divided into: 1) Move Plank, 2) Draw Line, 3) Saw into two halves, 4) Align at right angles, 5) Push nail and 6) Hammer in the nail. We consider five hand tools: 1) Borer, 2) Hammer, 3) Saw, 4) Screwdriver and 5) Triangle Ruler.
3 different actors following slightly different scripts were recorded from a top-down viewpoint with different amount of clutter and objects on the table. We provide extracted RGB frames for the six steps and a set of hand annotated groundtruths (.groundtruth) indicating the location of the object per frame as [xloc, yloc, width, height].
Questions? Please contact cteo "at" umd dot edu