Towards Human-Embodied Visual Intelligence