Seems more like a proof of concept project for that paper than something they are pursuing seriously judging by the GitHub location in some example folder that hasn’t seen any significant updates in over a year. If it is so great I would assume they would pursue it more actively and replace existing models with it two years later.