some conflicts in the code and the results in the paper

[ricupa]

In the code you have shared, i.e., the unet_mobilenetv2.ipynb (or .py), it doesn't seem like the entire code reproduces the results in the paper. The paper gives binary segmentation results (cloud/no cloud, valid/invalid/ etc.), but your code is for multi-class segmentation. Also, the metrics classes are not for the results you have presented.
So, am I looking at the wrong code? Have I found the wrong code, and is the correct one somewhere else in the repo?

I have one more concern: in the Unet model that you have initialized, you have not applied any activation at the output, and during the validation, while calculating the metrics, you take the arg max (without the softmax across the classes; see line 359 in the py file), in such the case the metrics calculation can be incorrect. Please let me know if I am wrong here or if I am missing something.