CG-SSL: Concept-Guided Self-Supervised Learning

Neural Information Processing Systems 

Humans understand visual scenes by first capturing a global impression and then refining this understanding into distinct, object-like components.