Adapt pyveg to work with chips arrays as outputs

urbangrammarai / gee_pipeline

Pipeline for retrieving imagery from Google Earth Engine

MIT License

1 stars 0 forks source link

Adapt pyveg to work with chips arrays as outputs #5

Closed crangelsmith closed 2 years ago

crangelsmith commented 2 years ago

[ ] Make the size of chip also part of the config file.
[x] Save chips as .npy or .npyz files (currently .png).
[ ] Move to eastings-northings + time range to identify the chip.
[ ] ~Ensure that the geo-registration info for the chips is retained (currently, it is lost with the .png files).~ (moved to #22)

andrewphilipsmith commented 2 years ago

Closed by mistake. Reopened now.

darribas commented 2 years ago

Looping @martinfleis here to see if he could give us a sense on how he was working on this front. My sense is that all the chips were written into a .npy file and read back in memory for training. Correct?

martinfleis commented 2 years ago

All the chips per option (say sliced 32x32) are saved as a single npy. Their bounds are saved as a GeoParquet with label and information on a subset (train, validation...). So the npy itself does not have any spatial information but that is stored in the GeoParquet.