Edited analyze, preprocess, train, evaluate functions. More needed: train/val losses, dose metrics, full preprocessing of input data for dose prediction. I made sure that both segmentation mask and dose prediction still work with these changes.
The assumption is that the input data for now are already preprocessed. I can always integrate the preprocessing steps later as the main task is to see if we can train and get reasonable results with these changes.
analyze.py and preprocess.py are edited to create config file and numpy input and target data for dose prediction.
training function is edited to include train/val loss such as MSE, MAE, .... For now when training, it seems like the val loss doesn't really do anything, so I am investigating that.
Inference code is edited to ensure that we can predict dose data.
Evaluation code still needs major overahuls. For now, I am using dice/haus95, just because I wanted to add basic changes that will allow the code to work with dose data. Pretty soon I will also include correct dose metrics such as dose score, dvh, ...
I still have many questions regarding the code and we can schedule a meeting for Friday to discuss them.
New commit: I just edited the loss and loss_utils code to correctly compute training and loss function. I wasn't indexing the y_true and y_pred correctly.
Edited analyze, preprocess, train, evaluate functions. More needed: train/val losses, dose metrics, full preprocessing of input data for dose prediction. I made sure that both segmentation mask and dose prediction still work with these changes.
The assumption is that the input data for now are already preprocessed. I can always integrate the preprocessing steps later as the main task is to see if we can train and get reasonable results with these changes.
I still have many questions regarding the code and we can schedule a meeting for Friday to discuss them.