I would like to ask if anyone has trained a Chinese text recognition model with crnn? The loss is not decreasing anymore, why the accuracy is always 0?

[htaoruan]

I would like to ask if anyone has trained a Chinese text recognition model with crnn? The loss is not decreasing anymore, why the accuracy is always 0?

[htaoruan]

Dataset labels are error free

[gaotongxiao]

Have you taken a look at CRNN's text output? Your dataset size might not be sufficiently large to teach CRNN to use the huge vocabulary base.

[htaoruan]

Thank you very much. That is to say, although the dicts I use is set to common Chinese and English characters, my dataset is too small to saturate the crnn model.