Contribution to the RescueNet Challenge 2023.
-
Announcement: https://www.hadr.ai/2023-rescuenet-challenge
-
Data: https://drive.google.com/drive/folders/1Gv7fr5YiIkNrEdVKMiIIKugTxuB6dHQH
-
Image Size: 704 (needs to be divisible by 32)
-
Augmentation methods and probabilities can be changed in
utils/augmentation.py
Run training with python fine_tune.py. Will use config/att_unet.json. Hyperparameters can be changed by writing a new config in the same style. For more information on possible values see config/README.md. Logs will be stored in metrics/scores/fine_tune. Adding --use_wandb option to train call will additionally use wandb for monitoring.
Hyperparameter tuning and prediction scripts are not tested yet.