What If We Only Use Real Datasets for Scene Text Recognition? Toward Scene Text Recognition With Fewer Labels | Read Paper on Bytez