Closed rengxumiaoshou closed 4 months ago
Hi, it's the validation mode (see README, --mode argument). Code is in main.py, see function validate(...). If you really want to do it properly you of course want to have train/validation/test datasets, but this repo keeps everything as simple as possible, therefore no testset.
I noticed in the readme that it mentioned that 3/4 of the words in the validation set were correctly recognized, giving a character error rate of about 10%. However, I did not find any code related to the performance evaluation of the experimental results. I hope you can tell me where the code to do this is and how to get these data if not. Thank you again for your selfless sharing and advice. You can send a wechat receipt code. Although I am a student, I still want to send red envelopes to express my thanks.