In binary image segmentation, for given a set of images, it's true mask and predicted mask. How do you compute dice score? Should I compute the dice score for each image separately and then find mean across all images? Or compute the dice score for all images at once by flattening tensor? Which is the correct way?

  • 1
    $\begingroup$ Does an event happen? If so in each image or over the set of images? $\endgroup$ – Teabelly Apr 13 at 17:34
  • $\begingroup$ Event happens over a subset of images. There can be multiple events, but each image can only belong to a single event $\endgroup$ – spb Apr 14 at 1:37

The correct way is to compute the DICE score per image and then find the mean, median and STD across all test images. It is good practice to report all three metrics to provide a clear intuition to the reader.

For more details, please refer to this answer.


Your Answer

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

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