Can I use own nnunet model weight?

[mumulin122333]

Hello,

Thank you for your great work!
I would like to ask if it's possible to use my own nnUNet model weights for segmentation.
If so, that would be extremely helpful for our project.

Thank you very much!

[DanielNobbe]

As I understand, this model is a version of nnUnet that is trained on inputs of image + interaction (the image and interactions are concatenated before they're fed into the model). The model is then trained to output the segmentation mask that it thinks is meant by the interaction input. You can use the repo to train your own version of nnInteractive, but your dataset size may be limiting.

[FabianIsensee]

The training code is not available, so you cannot train your own nnInteractive model. If you know public datasets that we should include in our training, please let us know!