search

HealthML / self-supervised-3d-tasks

Apache License 2.0
185 stars 39 forks source link

Loss calculation #16

Closed ramchandracheke closed 2 years ago

ramchandracheke commented 2 years ago

Hi, Thanks for this repo. I am trying to recreate the result. I found that this loss is negative. Please correct me if I am wrong, it should be 1-dice coefficient?

https://github.com/HealthML/self-supervised-3d-tasks/blob/9f21099867f3ac10fe232bd60899483079fd7c91/self_supervised_3d_tasks/utils/metrics.py#L68

We just have to add 1 in calculation? Ideally it should not harm to other code. Please let me know your inputs.

Thanks, Ram

aihamtaleb commented 2 years ago

Hi Ram,

thanks for your interest in the repo. I think you're right, it's more accurate to minimize (1-dice). The version we had worked fine too, as we were minimizing the negative version of dice. Feel free to use 1-dice