A Simple for 2D and 3D U-Net Pytorch Tutorial

The results trained and tested on Liver dataset from Medical Segmentation Decathlon. Checkpoints, loss plots, inference results etc are stored in the 3D UNet Liver and 2D UNet Liver

Usage

• Clone the repository

  git clone https://github.com/mikami520/UNET.git && cd UNET

• Train (modify the YAML as needed)

  Simply run

  python 2dunet_train.py --cfg 2dunet_config.yaml

  or

  python 3dunet_train.py --cfg 3dunet_config.yaml

  for 2D and 3D U-Net, respectively. Add --resume if you need to continue your previous training.

• Test

  Simply run

  python 2dunet_test.py --cfg 2dunet_config.yaml

  or

  python 3dunet_test.py --cfg 3dunet_config.yaml

Metrics

Multiple evaluation metrics are added to this tutorial:

• Mean Dice Score
• Mean Surface Distance
• Mean Hausdorff Distance
• Mean Surface Dice Score
• Mean Chamfer Distance

Star History

---

Developed by Chris Xiao | University of Toronto

© 2025 All Rights Reserved