Beyond Appearance: A Semantic Controllable Self-Supervised Learning Framework for Human-Centric Visual Tasks | IEEE Conference Publication | IEEE Xplore