This toolkit is part of my PhD research about data augmentation (DA) strategies on multi-organ segmentaiton(MOS) (mainly CT). These strategies have been re-implemented for MOS:
- CutMix
- ObjectAug
- CarveMix
- AnatoMix
Some utils are from my another repo mostoolkit.
In cases when inpaint utils are used, then you need to install pytorch.
Use data_preparation.ipynb to preprocess the data for DA strategies and data split configs(maybe needed for DA). prepare_nnunet_dataset.py might be useful to run nnUNet. metrics.py is used to generate the evaluation metrics (micro/macro avaraged dice score)
- CutMix
python cli_cutmix.py -sp split.json -d ./amos128 -s ./amoscutmix -n 200- ObjectAug
python cli_objectaug.py -sp split.json -d ./amos128 -s ./amosobjectaug -n 200 -nc 16 -nw 8- CarveMix
python cli_carvemix.py -sp split.json -d ./amos128 -s ./amoscarvemix -n 200 -nc 16 -nw 8- AnatoMix
python cli_anatomix_v2.py -sp split.json -d ./amos128 -s ./amosanatomix -n 200 -nc 16 -nw 8