Tight Coupling of Character, Word, and Place Recognition for End-to-End Text Recognition in Maps