Add more segmentation models

[qubvel]

Hi nnUNet team,

I'm Pavel, the creator of segmentation_models.pytorch (or SMP for short)! SMP is a Python library offering a collection of pre-trained neural networks for 2D image semantic segmentation tasks.

The library features:

• 12 model architectures, including Unet variants.
• Each mode is compatible with almost any SOTA image encoder from the timm library.
• Models are tested with torch.jit.script, torch.jit.trace, torch.compile, and export functionalities.

In addition, we are focusing on minimizing dependencies for the library to be installed.

I was wondering if you’d be interested in collaborating to integrate SMP into nnUNet. This could provide users with an option to build networks using SMP, unlocking additional flexibility and pre-trained options for nnU-Net’s pipelines. Thanks in advance for considering this idea! I’d be happy to assist in exploring how this integration could work.

Best regards,
Pavel

cc @FabianIsensee