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
Sometimes when mapping imagery, certain tiles will not contain any actual imagery. Right now, at least during on-the-fly mapping, the system will use dl.raster.ndarray to pull in the image tile, then test the alpha channel for blankness before proceeding with actual mapping (see util_mapping.map_tile). This means lots of unnecessary calls that slow down the overall mapping process. Instead, we could (theoretically) use the geospatial extent of the dl_tile object and the scene itself to do a very quick overlap check, prior to making any raster calls. This would very substantially speed up automated mapping.
The text was updated successfully, but these errors were encountered:
@eptrszkwcz, you have seemingly implemented something analogous to this for the tree canopy project.
You can see the original issue in the codebase here. As mentioned, the code pulls the actual imagery in and then inspects to see if it is empty, and then concludes if this is the case.
What we would actually want to do is compare the geometry of the tile with the geometry of the relevant scene, and only make a call for actual imagery retrieval if the two coincide. One possibility for accomplishing this would be an optional argument in map_tile for specifying scene geometry. The DL tile, which includes its geometry, is already available.
The final step would be incorporating this into map_scenes_simple, which currently is the only function utilizing map_tile. As it loops through scenes (ie scene_ids), you could do a metadata call to DL, get the shape of the scene, and then pass that shape into each call of map_tile as a keyword argument.
Sometimes when mapping imagery, certain tiles will not contain any actual imagery. Right now, at least during on-the-fly mapping, the system will use
dl.raster.ndarray
to pull in the image tile, then test the alpha channel for blankness before proceeding with actual mapping (seeutil_mapping.map_tile
). This means lots of unnecessary calls that slow down the overall mapping process. Instead, we could (theoretically) use the geospatial extent of the dl_tile object and the scene itself to do a very quick overlap check, prior to making any raster calls. This would very substantially speed up automated mapping.The text was updated successfully, but these errors were encountered: