For Conda users, you can create a new Conda environment using
conda create -n robustseg python=3.10
after activating the environment with
source activate robustseg
try to install all the dependencies with
pip install -r requirements.txt
also install the conda environment for the jupyter notebook kernel.
python -m ipykernel install --user --name=robustseg
Download the ATLAS data, prostate MRI data and Cardiac MRI data, and put extracted the data into ./data.
data/
├── ATLAS_R2.0/
├── ATLAS_2/
├── Training/
├── ...
└── ...
├── Processed_data_nii/
├── BIDMC/
├── ...
└── ...
├── OpenDataset/
├── Training/
├── ...
└── ...
Run datapreprocessing_ATLAS.ipynb, datapreprocessing_MnMCardiac.ipynb and datapreprocessing_Prostate.ipynb in the ./data folder to preprocess the data step by step. After preprocessing, we should have the data format like:
data/
├── Dataset_Cardiac/
├── 1/
├── ...
└── 5/
├── Dataset_Prostate/
├── ISBI/
├── ...
├── HK/
├── Dataset_Brain_lesion/
├── GE Signa Excite/
├── ...
└── Siemens Skyra/
Cardiac
python UNetSegmentationTrain.py --name 3DUNet_vanilla_Cardiac_det --tensorboard --features 30 --deepsupervision --batch-size 30 --patch-size 128 128 8 --epochs 1000 --evalevery 100 --numIteration 100 --sgd0orAdam1orRms2 0 --lr 1e-2 --print-freq 20 --ATLAS0Cardiac1Prostate2 1 --gpu 0 --det
Prostate
python UNetSegmentationTrain.py --name 3DUNet_vanilla_Prostate_det --tensorboard --features 30 --deepsupervision --batch-size 32 --patch-size 64 64 32 --epochs 1000 --evalevery 100 --numIteration 100 --sgd0orAdam1orRms2 0 --lr 1e-2 --print-freq 20 --ATLAS0Cardiac1Prostate2 2 --gpu 0 --det
Brain lesion
python UNetSegmentationTrain.py --name 3DUNet_vanilla_ATLAS_det --tensorboard --features 30 --deepsupervision --batch-size 2 --patch-size 128 128 128 --epochs 1000 --evalevery 100 --numIteration 100 --sgd0orAdam1orRms2 0 --lr 1e-2 --print-freq 20 --ATLAS0Cardiac1Prostate2 0 --gpu 0 --det
Brain lesion asymmetric large margin loss
python UNetSegmentationTrain.py --name 3DUNet_asymargin_2_ATLAS_det --tensorboard --features 30 --deepsupervision --batch-size 2 --patch-size 128 128 128 --epochs 1000 --evalevery 100 --numIteration 100 --sgd0orAdam1orRms2 0 --lr 1e-2 --print-freq 20 --ATLAS0Cardiac1Prostate2 0 --gpu 0 --asy-margin 2 --det
Brain lesion asymmetric focal loss
python UNetSegmentationTrain.py --name 3DUNet_asyfocal_6_ATLAS_det --tensorboard --features 30 --deepsupervision --batch-size 2 --patch-size 128 128 128 --epochs 1000 --evalevery 100 --numIteration 100 --sgd0orAdam1orRms2 0 --lr 1e-2 --print-freq 20 --ATLAS0Cardiac1Prostate2 0 --gpu 0 --asy-focal 6 --det
Brain lesion mixup
python UNetSegmentationTrain.py --name 3DUNet_mixup_ATLAS_det --tensorboard --features 30 --deepsupervision --batch-size 2 --patch-size 128 128 128 --epochs 1000 --evalevery 100 --numIteration 100 --sgd0orAdam1orRms2 0 --lr 1e-2 --print-freq 20 --ATLAS0Cardiac1Prostate2 0 --gpu 0 --mixup --det
Brain lesion assymetric mixup
python UNetSegmentationTrain.py --name 3DUNet_asymixup_m02_ATLAS_det --tensorboard --features 30 --deepsupervision --batch-size 2 --patch-size 128 128 128 --epochs 1000 --evalevery 100 --numIteration 100 --sgd0orAdam1orRms2 0 --lr 1e-2 --print-freq 20 --ATLAS0Cardiac1Prostate2 0 --gpu 0 --mixup --asy --alpha 1 --det
Brain lesion adversarial training
python UNetSegmentationTrain.py --name 3DUNet_adv_ATLAS_det --tensorboard --features 30 --deepsupervision --batch-size 2 --patch-size 128 128 128 --epochs 1000 --evalevery 100 --numIteration 100 --sgd0orAdam1orRms2 0 --lr 1e-2 --print-freq 20 --ATLAS0Cardiac1Prostate2 0 --gpu 0 --adv --det
Brain lesion GIN
python UNetSegmentationTrain.py --name 3DUNet_GIN_ATLAS_det --tensorboard --features 30 --deepsupervision --batch-size 2 --patch-size 128 128 128 --epochs 1000 --evalevery 100 --numIteration 100 --sgd0orAdam1orRms2 0 --lr 1e-2 --print-freq 20 --ATLAS0Cardiac1Prostate2 0 --gpu 0 --GIN --det
We provide pretrained segmentation models here.
UNetSegmentationTest.ipynb