HumanBench: Towards General Human-centric Perception with Projector Assisted Pretraining SCPNet: Semantic Scene Completion on Point Cloud