A description projection for our JSCAS proposed work
"Towards Better Semi-Supervised Multi-Organ Segmentation from CT Volumes: Random Frequency Masking and Pseudo-Label Refinement".
Run the code for data preprocessing:
python code/data/preprocess_acc.py
Change the data path in
code/utils/config.py
You need to modify the location where the data is stored before the preprocess.
Then, train the model with 5%,10%, and 20% labeled volumes:
bash train3times_acc_5_10_20.sh
You can also test the weights we pre-trained by
tar -xzvf logs.tar.gz
python code/evaluate_Ntimes2.py --task acc_s --exp Task_acc_s_{labeled_ratios}p/{method} --cps AB
The logs.tar.gz can be downloaded here: logs.tar.gz
Have fun!