|── README.md
|── environment_setup.sh
|── conver_testformat.py
|── nnunet
|── nnUNet_raw_data_base (download)
|── nnUNet_preprocessed (download)
|── nnUNet_trained_models (If you train the model, it will be made automatically)
source ./environment_setup.sh
New arguments
- '$save_name': your trained results will be saved here.
- --use_nnblock: if you want to use 3D nnblock, please use this argument
- --use_ws : if you want to use weight standardization, please use this argument
- --use_skip_attention : if you want to use skip attention, please use this argument
CUDA_VISIBLE_DEVICES=1 python nnunet/run/run_training.py 3d_fullres nnUNetTrainerV2 Task000_MYTASK all '$save_name' --use_nnblock --use_ws -w genesis_nnunet_luna16_006.modelIf you run above the script, you can find your training results:
nnUNet_trained_models/nnUNet/3d_fullres/Task000_MYTASK/nnUNetTrainer....v2.1/$save_name/all
-
You have to rename your .model, .pkl files. Please change the name of the model and pickle files named 'model_best.model', 'model_best.model.pkl' by following the below conditions.
model : model_final_checkpoint.model
pickle : model_final_checkpoint.model.pkl
CUDA_VISIBLE_DEVICES=1 python nnunet/inference/predict_simple.py -i nnUNet_raw_data_base/nnUNet_raw_data/Task000_MYTASK/imagesTs -o '$output_path' -t Task000_MYTASK -m 3d_fullres -f all- If you don't want to rename your model, please add '-chk model_best' argument
- $save_name : if you trained with '$save_name' argument, you have to input --name='$save_name' argument in inference phase.
CUDA_VISIBLE_DEVICES=1 python nnunet/inference/predict_simple.py -i nnUNet_raw_data_base/nnUNet_raw_data/Task000_MYTASK/imagesTs -o '$output_path' -t Task000_MYTASK -m 3d_fullres -f all -chk model_best --name='$save_name'If you finished the inference, you can check a new folder named $output_path and $folder_name.
python convert_testformat.py --data_path='$output_path' --save_path '$folder_name'Then, you can find submission folder in your '$output_path'.
please compress it and submit.