Hello i want to know if F1 score in DL4J is equivalent to Dice coefficient.
The dice coefficient is a similarity metric, so in that sense it is not the same as F1.
However, when applied to boolean data, they are equivalent in that they have the same numeric result.
@treo thank you, i use them in semantic segmentation, and i see in so many articles they consider them the same for evaluation of semantic segmentation like here:https://towardsdatascience.com/metrics-to-evaluate-your-semantic-segmentation-model-6bcb99639aa2
For a binary segmentation as it is decribed in the article you’ve linked, it should be the same. But when you have to deal with multiple classes, you might get different results from what you expect.
Take a closer look at the documentaton of evaluation concerning the multi class case: https://deeplearning4j.org/api/latest/org/nd4j/evaluation/classification/Evaluation.html