Preparation of GeoPlex dataset

[sfalkena]

Hi,

As far as I have understood, the Geoplex dataset is not available for download in a single place, but rather it it constructed as a multidataset from the 5 individual dataset classes. I am trying to reproduce your pretraining, so I would like to get the datasets in place. I have a few questions on getting this done, and I am happy to contribute back to your repo by forming a guide so others after me can reproduce this too.

Could you confirm where you are getting the original datasets? So far I was trying to use the following, correct me if I am wrong:

• TreeSatAI: https://huggingface.co/datasets/IGNF/TreeSatAI-Time-Series
• PASTIS-HD: https://huggingface.co/datasets/IGNF/PASTIS-HD
• FLAIR: https://huggingface.co/datasets/IGNF/FLAIR-1-2
• Planted: https://storage.googleapis.com/planted-datasets/public
• S2NAIP: https://huggingface.co/datasets/allenai/s2-naip

Do you any additional configuration on the datasets, or should I be able to run all experiments after putting datasets in /data?

Before launching an experiment on GeoPlex, I am trying to launch LP experiments on the individual datasets, starting with Pastis. However, if I look at the config of Pastis (configs/dataset/Pastis.yaml), I see the dataset name is PASTIS-R, not PASTIS-HD. The modalities_dict does suggest that spot, s1 & s2 is being used, so I assume this is just a name thing and you might have renamed the pastis folder from PASTIS-HD to PASTIS-R ?

[gastruc]

Yes this is correct. Indeed, the name of pastis is an artifact in my code but I am using Pastis-HD.

[makovez]

@sfalkena were you able to replicate the results ?