Code was tested using:
python==3.10.12
torch==2.7.1
Clone this repository, create a python environment for the project and activate it. Then install all the dependencies with pip.
cd ShaSpec
python -m venv shaspec_venv
source shaspec_venv/bin/activate
pip install -r requirements.txt
Perform a warm up with all modalities train_SS.py with the following arguments:
python train_SS.py \
--snapshot_dir snapshots/BraTS23_ShaSpec_warmup \
--batch_size 1 \
--num_gpus 1 \
--num_steps 245000 \
--val_pred_every 875 \
--learning_rate 1e-2 \
--num_classes 3 \
--num_workers 4 \
--train_list BraTS23/BraTS23_train.csv \
--val_list BraTS23/BraTS23_val15splits.csv \
--test_list BraTS23/BraTS23_test15splits.csv \
--random_mirror \
--random_scale \
--weight_std \
--warm_up
Run the training script train_SS.py with the following arguments:
python train_SS.py \
--snapshot_dir /snaphots/BraTS23_ShaSpec_training \
--batch_size 1 \
--num_gpus 1 \
--num_steps 406000 \
--val_pred_every 875 \
--learning_rate 1e-2 \
--num_classes 3 \
--num_workers 4 \
--train_list BraTS23/BraTS23_train.csv \
--val_list BraTS23/BraTS23_val15splits.csv \
--test_list BraTS23/BraTS23_test15splits.csv \
--random_mirror \
--random_scale \
--weight_std \
--mode random \
--reload_path ShaSpec/snapshots/BraTS23_ShaSpec_warmup/best.pth \
--reload_from_checkpoint
Run the test script eval.py with the following arguments:
python eval.py \
--num_classes 3 \
--data_list BraTS23/BraTS23_test.csv \
--weight_std \
--restore_from /work/grana_neuro/missing_modalities/ShaSpec/snapshots/BraTS23_ShaSpec_training/final.pth \
--mode <MODE>
where <MODE> indicates the missing modality scenario. In the command above, replace <MODE> with one of the following missing‑modality configurations to evaluate every scenario:
0
0,1
0,1,2
0,1,2,3
0,1,3
0,2
0,2,3
0,3
1
1,2
1,2,3
1,3
2
2,3
3