Open KennSmithDS opened 2 years ago
Furthermore, clarification would be helpful on the implementation of the name
property of the ClassObject
, which currently states:
REQUIRED The property key within the asset's each Feature corresponding to class labels. If labels are raster-formatted, use null.
If the label extension is intended to be used with vector assets only, then for raster assets with contain labels, shouldn't the line "If labels are raster-formatted, use null" be removed from the specification, and the classification:classes
property take precedence over label:classes
?
Pending @kbgg review on #13
Background:
At Radiant Earth, we've been using the
label
extension for both raster and vector label assets in our catalogs/datasets. This means that any item which represents a label asset from an ML dataset will employ thelabel
extension. These labels will most commonly be either GeoJSON or GeoTIFF, with some exceptions of Numpy array files, etc. However, reading closer into the specification today, I noticed a contradiction.Issue
In the first section of the specification, there is the verbiage:
These labels can take several forms, though all are expected to be contained with a GeoJSON FeatureCollection
- emphasis on expected to be contained with a GeoJSON FeatureCollectionYet then, the
label:type
field stores an ENUM of eitherraster
orvector
.If the former is true, that label assets must be contained within a GeoJSON, then raster images with classification labels do not apply to this extension.
Proposal/request
vector
value forlabel:type
could also imply other vector file formats, e.g. shapefile, not just GeoJSON.