It has detailed training logging. It's more concise than the previous code. It's simpler and easy to use. In this work, we use STISR datasets TextZoom and four STR benchmarks, i.e., ICDAR2015, CUTE80, ...