Allow a middlebury dataset directory to be used as a source for a vision node. It loads left, right, groundtruth and calibration. The output is a video feed of the static images.