Dionysus: Recovering Scene Structures by Dividing Into Semantic Pieces | Read Paper on Bytez