Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks (Updated 11 months, 3 weeks ago)