Closed kvantricht closed 4 months ago
@GriffinBabe I prepared a first demo dataset as requested. Remember in the future this data will come from requests to the API. But for the time being, please find the file here: /vitodata/worldcereal/tmp/kristof/GFMAP/2021_EUR_DEMO_POLY_110.gpkg
This is coming from EUROCROPS, and hence spans multiple countries which is interesting for testing the spatial splitter. Following attributes are present (can be renamed in the future):
sample_id
: can be considered unique ID in the entire RDM to identify a field or a pointlandcover_label
: worldcereal landcover label to be rasterized in the ground truthcroptype_label
: worldcereal croptype label to be rasterized in the ground truthirrigation_label
: legacy, ignore for nowconfidence
: indiciation of confidence of this ground truth label, needs to be added to metadataextract
: boolean flag indicating if a sample needs to be extracted or notvalid_date
: date for which sample is valid, forms the basis for defining extraction start/end timerange (for now we should take mayba 1,5 year centered around valid_date.ref_id
: id of original dataset the sample belongs to. Needs to be exported to the metadata.So when extract==True
, the field needs to be used for extracting a patch. Remember that for rasterizing ground truth, we want all fields inside the patch regardless if they need to serve as extraction or not. We just need to make sure the fields cover the same (more or less) valid_date to be valid for the actual extraction. In this case, all data comes from same ref_id so we should be safe.
Need a representative dataset from which we can start setting up the extractions workflow.