Time issue vector_samples_merge
When we gather the extraction files by chunk into a single netcdf file for deep learning workflow, the build_raw_db function
opens and reads all the chunk files to be concatenated in order to get the list of all the field names (here).
This process it can take a very long time when there are many features, tiles and chunks.
If I'm right, the fields must be the same for all the chunks of a same tile.
To save time, we'd like to read only one file per tile.
Do you see any problem?
Edited by Hélène Touchais