Prepare map patches for U-Net prediction — unet_prep

Tiles the full ortho extent (or a clipped region) into overlapping patches for wall-to-wall prediction. Patches are saved as numpy arrays alongside a CSV of spatial origins for later reassembly.

Usage

unet_prep_map(
  fitid = NULL,
  model = NULL,
  clip = NULL,
  resources = NULL,
  local = FALSE,
  trap = TRUE,
  comment = NULL
)

Arguments

fitid: Fit id in the fits database. The model name is resolved from the database. Provide either fitid or model, not both.
model: The model name (base name of the prep .yml in <pars>/unet/). Use this to prep map patches for a model that may have multiple training runs, or before any training has been registered in the fits database.
clip: Optional clip extent, vector of xmin, xmax, ymin, ymax. If NULL, tiles the full ortho extent.
resources: Slurm launch resources. See launch.
local: If TRUE, run locally; otherwise, spawn a batch run on Unity
trap: If TRUE, trap errors in local mode
comment: Optional slurmcollie comment