My object detection model is overfitting , how can i solve this problem?

vision
1

i am using retinanet. when i draw loss graph for training and validation loss, after 5 epoch , validation loss is increasing. (i used augmentation and L2) What do you think about that ?

Thanks,

2

To counter overfitting you could try to collect more data, increase regularization, lower the model capacity etc.
There is no single answer to this problem, so you might need to try out different approaches.

1 Like
3

Maybe this paper can help: https://arxiv.org/abs/1902.04103

1 Like
4

thanks, i will try . Is there any pipeline for ?

5

thanks, i will read this paper.