https://arxiv.org/abs/2204.07714
Pushing the Performance Limit of Scene Text Recognizer without Human Annotation (Caiyuan Zheng, Hui Li, Seon-Min Rhee, Seungju Han, Jae-Joon Han, Peng Wang)
semi supervised training + domain adaptation 까지 들어갔네요. domain 문제에 대해서는 흥미롭네요. 삼성 쪽에서 참여했군요.
#ocr