Evaluating on the train sets

[isaaccorley]

First I want to thank the authors for the hard work on generating the dataset!

I noticed in the benchmark_dataset_info.py script that it appears the evaluation for 4/8 datasets is being done on the train sets and not the test sets. Any particular reason why?

For the datasets with a test set like AID and EuroSAT could you also explain the splitting process used?