Because future versions of DECALS will have this for object catalogue, and imaginglss does use the same algorithm for reading out per pixel values from the depth maps, it appears we no longer have to query the raw images in select-objects.py. The amount of IO will be a lot smaller. As a consequence, the entire script doesn't need mpi parallelism any more, which could be a plus.
However we then lose the ability to cross-check the depth readout algorithm with tractor.
c.f. https://github.com/legacysurvey/legacypipe/issues/72
Because future versions of DECALS will have this for object catalogue, and imaginglss does use the same algorithm for reading out per pixel values from the depth maps, it appears we no longer have to query the raw images in select-objects.py. The amount of IO will be a lot smaller. As a consequence, the entire script doesn't need mpi parallelism any more, which could be a plus.
However we then lose the ability to cross-check the depth readout algorithm with tractor.
So this is not necessarily a good move.