search

bowang-lab / U-Mamba

U-Mamba: Enhancing Long-range Dependency for Biomedical Image Segmentation
https://arxiv.org/abs/2401.04722
Apache License 2.0
652 stars 60 forks source link

changing nnUNet base directories #24

Closed idinsmore1 closed 6 months ago

idinsmore1 commented 6 months ago

Hello! Thanks for putting this tool together. I just wanted to bring up a bug that I noticed while testing this repo. I've used nnUNet before, so I placed my data in a different location than U-Mamba/data and changed the default nnUNet_raw/preprocessed/results directories using bash's export command. However, your custom script doesn't reflect those changes when you run any of the nnUNet commands, it still points nnUNet_raw to U-Mamba/data/nnUNet_raw (same with preprocessed and results). Just wanted to bring this to attention if other experienced nnUNet users try this out. Great work on the paper and model!

ff98li commented 6 months ago

Hi, Thank you for your interest in trying out U-Mamba! The paths to directories of data in U-Mamba are indeed preset to the codebase's data directory U-Mamba/data. This is for preventing conflicts with environment variables set for pre-existing nnUNet instances. However, feel free to customize these paths to fit your setup by modifying the directory paths in umamba/nnunetv2/path.py according to your preferred nnUNet data directories, as in https://github.com/bowang-lab/U-Mamba/blob/887b3ca35e28f23645708c331b4ca714715e71af/umamba/nnunetv2/paths.py#L44-L48 We hope this addresses your concern.

JunMa11 commented 6 months ago

We also add a remark on readme. Thanks for pointing out this:)