The training results are good, but the test results are very poor.

[abner2015]

when i use my dataset train the ocr model , the training results are good, but the test results are very poor. i do not need segment the roi . the train data is number .how can i improve the test result.

[ThomasDelteil]

It looks like you are experiencing overfitting. https://en.wikipedia.org/wiki/Overfitting A few ideas to overcome this and regularization to your model:

• you can stop training when the validation loss start going up for N epochs
• you can add more data augmentation on your training set
• you can increase the size of your training set
• you can add weight decay in your optimizer
• you can add more dropout layer / tune the dropout values
• you can use a different learning rate schedule
• you can use more batch normalization layers

[samar-smida]

@ThomasDelteil I am testing your project but I have an assertion error even though I put my email and password in the credentials.json . In the registration form, they ask for the email and not the username. I found this link in the project to have but it is not functional https://fki.tic.heia-fr.ch/DBs/iamDB/iLogin/index.php please can you help me