Eventually it should be easy to work with rslearn datasets that sit in S3.
For now we should make it seamless to train data from S3. Not sure if that means going through dataset like we did for multisat and putting each batch into a separate object, or just put each window as separate object, or something else.
Eventually it should be easy to work with rslearn datasets that sit in S3.
For now we should make it seamless to train data from S3. Not sure if that means going through dataset like we did for multisat and putting each batch into a separate object, or just put each window as separate object, or something else.