Low loss but low accuracy

[ThaiLe189]

Hi @Pay20Y
I was training model with 200000 data like this:
Img:

label:
2,3,48,3,48,24,3,25,quyền
51,5,81,4,81,22,51,23,hạn
84,5,110,4,111,19,84,20,của

Loss:

but while i test, results like:
img:

recog:
-2,6,45,5,45,24,-1,24,u)
47,7,81,7,80,21,46,21,1.
73,7,123,7,122,22,72,21,ni

What should I do to improve my model?

Thank you very much

[Pay20Y]

Hi, the recognition branch may need more data, you can try to pretrain the model on Synthtetic datasets first.

[ThaiLe189]

I have train it with 200000 images, loss is very low but while i tested it with the trained image, it not correct.
Is my input image too small?
Thank you!

[vanqm]

Pre-Trained model is trained with Synthtetic datasets, an English dataset.
If you want to apply to Vietnamese language, I think you should:

• Not use pre-trained model. You must train the model from beginning with your own dataset (Vietnamese)
• Or you can apply transfer-learning technique with your dataset

Hope it helpful with you.