val_loss can not decrease

qjadud1994 / CRNN-Keras

CRNN (CNN+RNN) for OCR using Keras / License Plate Recognition

MIT License

527 stars 191 forks source link

val_loss can not decrease #58

Open Chung-Yi opened 3 years ago

Chung-Yi commented 3 years ago

I faced some problem when I trained. At the beginning, both the loss and val_loss decreased step by step, but then only loss decreased val_loss frozen between 24 and 25. My trained data is 1922 and my tested data for evaluate is 213, their ratio is 9(9:1), I have no idea if it is enough for training, and I have no idea if it is the reason cause the problem.

Chung-Yi commented 3 years ago

It seems that it was overfitting.

Chung-Yi commented 3 years ago

Does anyone have the same problem?