The implementation of Weakly Supervised Volumetric Segmentation via Self-taught Shape Denoising Model
- Python 3.6
- Pytorch 1.4
- Torchvision 0.4
- Cuda version 10.1
Please refer to data_preparation.md
python data_process/process_pipeline2.py
# 1. train segmentation model with weak labels
python main.py --cfg exp/weak_trachea/1012_tra_r1_01.yaml --id 1012_tra_r1_01 --parallel
# 2. train shape denoising network
python main.py --cfg exp/ae/1013_tra_aelo_13.yaml --id 1013_tra_aelo_13 --parallel
# 3. iterative learning
python main.py --cfg exp/iterative/1016_trar1_emitw_24.yaml --id 1016_trar1_emitw_24 --parallel
python main.py --cfg exp/iterative/1016_trar1_emitw_24.yaml --id 1016_trar1_emitw_24 --parallel \
--demo 'val' --weight_path '/*/*/best_model.pth' [--ae_weight_path '/*/*/best_model.pth']
TODO
- Introduction of our shape-aware segmentation model
- Data preparation detail
- Trained model and final results