Configure Recognition to return labels with duration object is detected in video instead of individual instances per second of video

[isaacmckinney]

So instead of individual instances for every second of video that the detected object is in frame, we want recognition to return the instances with attributes 'starting timestamp' and 'ending timestamp'