Official Implementations Preview V1.0
- Python 3.10+
- PyTorch 2.4.0+
- mgamdata package: Gitee Repo
- mmsegmentation redistribution: Gitee Repo
- mmengine redistribution: Gitee Repo
- mmpretrain redistribution: Gitee Repo
- NVIDIA CUDA
This research heavily relies on our mgamdata package, the AutoWindow Module can be found in Gitee Repo.
export mm_workdir="..." # runner workdir
export mm_testdir="..." # runner testdir
export mm_configdir="..." # runner configdir
export supported_models="SegFormer3D,MedNeXt" # supported models/root/data
├── image
│ ├── 00001.nii
│ ├── 00002.nii
│ ├── 00003.nii
│ └── ...
│
└── label
├── 00001.nii
├── 00002.nii
├── 00003.nii
└── ...
itk_resample \
/root/data \
/root/data_resampled \
--mp \
--spacing 2 1 1 # ZYX (You may change to your own spacing)split3d \
/root/data_resampled \
/root/data_patched \
--window-size 16 \
--stride 8 \
--mp/root
├── data
│ ├── image
│ │ ├── 00001.nii
│ │ ├── 00002.nii
│ │ ├── 00003.nii
│ │ └── ...
│ └── label
│ ├── 00001.nii
│ ├── 00002.nii
│ ├── 00003.nii
│ └── ...
│
├── data_resampled
│ ├── image
│ │ ├── 00001.mha
│ │ ├── 00002.mha
│ │ ├── 00003.mha
│ │ └── ...
│ └── label
│ ├── 00001.mha
│ ├── 00002.mha
│ ├── 00003.mha
│ └── ...
│
└── data_patched
├── 00001
│ ├── 0.npz
│ ├── 1.npz
│ ├── 2.npz
│ └── ...
├── 00002
│ ├── 0.npz
│ ├── 1.npz
│ ├── 2.npz
│ └── ...
├── 00003
│ ├── 0.npz
│ ├── 1.npz
│ ├── 2.npz
│ └── ...
└── ...
You may have to modify the config file according to specify your dataset path. All configs are stored in configs/ folder. If you do not want to run the implementation, the configs can provide sufficient information for you to understand the implementation.
mmrun {ConfigVersionPrefix}For any questions, please contact us via email:
Yiqin Zhang (Corresponding Author)