I checked my configs, charset containing Chinese in particular, and ran train.py. The output model can't recognize Chinese characters at all.
I ran another friend's output model. It works in Chinese, but when I used its dataset and retrained, recognization failed again.
Appreciated for any advice.