Multiple Classes

[saskra]

https://github.com/WillBrennan/SemanticSegmentation/blob/7646661aa0e05f5c5b087f2aeda3faff42c34e00/train.py#L83

If I see it correctly, you have also solved semantic segmentations for multiple classes and not only binary decisions like skin vs. background with this source code, for example, in the pizza topping task. Could it be that for this, however, something other than this threshold transformation must be used for the metrics as in the line above? And that e.g. also the sigmoid output transformation in this evaluator would have to be adjusted? Do you happen to have a solution for this?