train problem

[betterhalfwzm]

I'm sorry to bother you again. When i train custom dataset, After several eopch, the score reached the highest level, and then began to decline，how can i solve it? Example of pictures： @zylo117

[zylo117]

it's overfitting

[betterhalfwzm]

it's overfitting

How to solve over fitting in your code? And there are a lot of negative samples in my data set， Does this affect? How should I adjust the code to add negative sample for training? @zylo117

[heitorrapela]

@betterhalfwzm try to add some early stopping and save the best model on validation loss (split the data in train / val / test ).