0
$\begingroup$

enter image description here

enter image description here

I tried to train CNN with very closely similar images for classification. Batch Normalization and dropout of 0.25 is used to overcome overfitting. This is the result I found.How can I interpret this result? Overfit, underfit or good fit?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

I know its a very late response. But its a classical overfitting example. The point at which your training error decreases but validation error increases is where overfitting starts. I would suggest the following

  1. I would suggest a more stronger dropout(maybe 0.5) and/or batch norm after at every layer to avoid overfitting (or) you should do an early stopping if you see overfitting even after increasing dropout.

  2. Remove FC layers and try FCN network with Network-in-network architecture.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.