Mean dice coming out to be 0 about medicalnet

[DecentMakeover]

Hi, thanks for the response

Its a 2 class segmentation task.

[cshwhale]

Hi, there are 3 suggestions:

1. If your custom data is small amount, it may be better to use batchnorm parameters from the pre-trained model during training.
2. If the task is unbalanced segmentation (small-sized target), patch-based segmentation method may be a good choice.
3. Use shallow neural network or skip-connection.
   Hope it helps.

[DecentMakeover]

Thanks for your suggestions, Ill check .