Pytorch implementation of Carcino-Net.
- Use SICAPV2 prostate data for segmentation of negative (non-cancerous background), low grade (Gleason grade group 3), and high grade (Gleason grade group 4 and 5).
python -m carcino_net.scripts.train_carcino --help
for detail arguments of training scripts.
python -m carcino_net.scripts.validate_carcino --help
for detail arguments of validation and to export showcase output masks.
To DO:
- documentation
- Multi-class focal loss are directly derived from Adeel Hassan's implementation. Alternatively you may use the weighted cross entropy in pytorch combining the (1 - softmax score) as the focal term.