Valid folder with SegmentationDataLoaders.from_label_func

I need to validate my models againts a concrete folder, using a random porcentage of training data is not possible owing to my requirements.

How could I achieve that with SegmentationDataLoader? Do I need to use the middle API for that objective?

Yes, the factory methods are useful to quickly get started on standard problems, but as soon as you get something different, you need to use the data block API.