HM2-Net

This repository provides the official implementation of HM2-Net, a hybrid U-Net architecture that integrates multi-scale state space modeling with uncertainty-guided attention for robust and efficient medical image segmentation.

Installation

• Python version: 3.10
• Install the required Python packages using: pip install -r hm2net/requirements.txt

Pretrained Models (Optional for ViT-based backbones)

You can download ViT-based pretrained models from the following link: https://storage.googleapis.com/vit_models/imagenet21k/{MODEL_NAME}.npz Example models:

• R50-ViT-B_16
• ViT-B_16
• ViT-L_16

Run the code

Run segmentation on your dataset using: python hm2net/test.py --dataset {dataset_name} --vit_name R50-ViT-B_16