search

deephealthproject / eddl

European Distributed Deep Learning (EDDL) library. A general-purpose library initially developed to cover deep learning needs in healthcare use cases within the DeepHealth project.
https://deephealthproject.github.io/eddl/
MIT License
34 stars 10 forks source link

Missing Dice Loss and Focal Loss #153

Closed CostantinoGrana closed 4 years ago

CostantinoGrana commented 4 years ago

In order to work with the pneumothorax challenge, the evaluation is given with the Dice coefficient: https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/overview/evaluation It would be important to have Dice loss

Also having Focal Loss would be useful.

RParedesPalacios commented 4 years ago

I will implement now as a new metric. In any case, this metric requiere discrete activations in the tensors. The simplest case will be binary, in any case we need to binarize, are you aware of that?

I mean that you should provide some threshold in some point. Right now i will put 0.5.

Am I wrong?

Then i will implement the loss

RParedesPalacios commented 4 years ago

Well, it is implemented, bye!