I know it's basically impossible to pinpoint the exact answer to this question without data, code, architecture, etc, but I'm curious as to if anyone has any general ideas on why this might be happening:

enter image description here

You can see the general trend towards convergence, but for both the test and training, at certain times, the dice score can be either 0.0 and 1.0. You can see this in the unsmoothed scores (lighter shades), where blue is train and red is test.

What are some common sources of variance in a convolutional neural network?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.