schmitt-muc/SEN12MS

Missing validation split files

isaaccorley opened this issue · 2 comments

isaaccorley commented

@schmitt-muc Just want to start off with saying I appreciate the effort made to create a multimodal dataset like this.

I am trying to use the dataset from the SEN12MS/classification/dataloader.py script, however it seems that a val_list.pkl and val_list.txt file is missing from the SEN12MS/splits directory. Any chance that this could be uploaded? Thanks!

schmitt-muc commented

@isaaccorley And I appreciate your interest in the dataset and repository. :)

You are correct, we have not provided a validation file list, as there are different strategies to define a good val set. Thus, we decided to leave it to the users.

I would suggest the following procedure:

  • Decide how large you want your val set to be, i.e. how many patches it should contain
  • Sample (randomly) those patches from the "training" split, i.e. from those patches that are not part of the hold-out split provided by us.
  • Remove the selected validation patches from the training set list.

Hope that helps!

isaaccorley commented

@schmitt-muc This definitely helps. Thanks for the insight.