Source of data: https://github.com/FudanVI/benchmarking-chinese-text-recognition