You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is your feature request related to a problem? Please describe.
Not all datasets in torchgeo are GeoDatasets -- some are NonGeoDatasets that do not use geospatial information. I am having issues with CRSs when using a global, pre-patched dataset like SSL4EO-S12. torchgeo already supports this dataset with a torchgeo.datasets.SSL4EO-S12 but is a NonGeoDataset. However, a lot of minerva is designed to work with GeoDataset, particularly the paired sampling.
Describe the solution you'd like
Generalise minerva to work with NonGeoDataset. In particular, there will need to be versions of PairedDataset and PairedUnionDataset that work with NonGeoDataset and the associated paired sampling.
Describe alternatives you've considered
The issue encountered with the CRS in SSL4EO-S12 relates to the projection system resulting in patches sampled with missing data. This could alternatively be avoided by rejecting any samples containing zeros -- but this would be an inefficient and messy implementation.
The text was updated successfully, but these errors were encountered:
Is your feature request related to a problem? Please describe.
Not all datasets in
torchgeo
areGeoDatasets
-- some areNonGeoDatasets
that do not use geospatial information. I am having issues with CRSs when using a global, pre-patched dataset like SSL4EO-S12.torchgeo
already supports this dataset with atorchgeo.datasets.SSL4EO-S12
but is aNonGeoDataset
. However, a lot ofminerva
is designed to work withGeoDataset
, particularly the paired sampling.Describe the solution you'd like
Generalise
minerva
to work withNonGeoDataset
. In particular, there will need to be versions ofPairedDataset
andPairedUnionDataset
that work withNonGeoDataset
and the associated paired sampling.Describe alternatives you've considered
The issue encountered with the CRS in SSL4EO-S12 relates to the projection system resulting in patches sampled with missing data. This could alternatively be avoided by rejecting any samples containing zeros -- but this would be an inefficient and messy implementation.
The text was updated successfully, but these errors were encountered: