how fold are assembled

[tzebre]

Hi, I have a question about prediction. How are folds "assembled" when the predict command is done with all 4 folds?

I have found this information in another Q&A. It says that folds are averaged on the softmax prediction. However, I do not understand how the class output (e.g., 1, 2, etc:) can be integer when averaging softmax values. Do you do some sort of rounding?

Thank you

[tzebre]

I believe I have found an answer to my question with the assistance of @naga-karthik .

The model calculates the average probability among predictions from different folds (ensemble.py/average_probabilities).
Argmax is applied to obtain integer results (label-handling.py/convert_probabilities_to_segmentation)

If you need the original probabilities, you should use the argument --save_probabilities when running the prediction.