This dataset was recorded from the standalone r1-mk2.
In each dataset, the subject is steady for the first seconds, in order to allow the skeletonRetriever to observe the skeleton and perform the optimization.
The train folder contains 38 subfolders, each containing 10 repetitions of a single action. Each action, except the reaching, is repeated 5 times and viewed from 2 points of view. The reaching action is performed in order to reach 4 different targets, distributed on the corners/center of a square, centered around the shoulderCenter of the subject.
Points of view are named as following:
- v0: frontal view
- v1: side view
Actions are named as following:
- abduction-right
- external-right
- internal-right
- reaching-right
For the reaching-left action, targets to be reached are distributed in a square, and are specified as following:
- t1: target in the center of the square
- t2: target in the top-right corner
- t3: target in the bottom-right corner
- t4: target in the bottom-left corner
Each folder is named as person name-point of view-target - action name (target is specified only for the reaching action), containing the following subfolders:
- depth: stores depth images
- skeletons-data: stores skeletons bottles
- skeletons-img: stores skeletons images
- opc: stores skeletons from opc
- opc-optimized: stores skeletons from opc after the optimization at three different speeds, obtained by changing the twarp parameter in the skeletonPlayer:
- twarp=0: speed of acquisition;
- twarp=0.3: squeezed original stream (higher speed);
- twarp=1.5: dilated original stream (lower speed).