CLI Reference

mapcv init

Writes an annotated example config to a file, or prints it to stdout if no path is given.

mapcv init my_dataset.yaml

Open the file and fill in your region, tile source, and label path. Everything else has sensible defaults.

---

mapcv validate

Parses and validates a config file without downloading any tiles or writing any files. Run this after editing your config to catch errors before a long download.

mapcv validate my_dataset.yaml

---

mapcv generate

Runs the full pipeline: tile download, label rasterization, patch sampling, and dataset writing.

mapcv generate my_dataset.yaml

When it finishes you will find:

output/
  Images/
    patch_0000000.png
    patch_0000001.png
    ...
  Masks/
    patch_0000000.png
    patch_0000001.png
    ...
  manifest.json
  splits/
    train.txt
    val.txt
    test.txt

Mask pixels are 0 for background and 1..255 for class IDs. If no labels section is present in the config, no Masks/ directory is written. If no split section is present, no splits/ directory is written.

---

mapcv split

Re-splits an existing dataset using its manifest.json without re-downloading or re-processing anything. Useful when you want to try different split ratios without rerunning the full pipeline.

mapcv split [OPTIONS] STAGING_DIR

Option	Default	Description
--test-ratio	0.2	Fraction of all patches reserved for test.
--val-ratio	0.1	Fraction of remainder used for validation.
--labeled-ratios	0.10 0.20 0.30	Optional. Pass once per ratio, e.g. --labeled-ratios 0.10 --labeled-ratios 0.20.
--seed	42	Random seed.
--strategy	stratified	random or stratified.
--sample-limit	—	Optional. Cap on total patches sampled.

mapcv split ./output --test-ratio 0.15 --val-ratio 0.10

Split files are written to STAGING_DIR/splits/ (e.g. output/splits/train.txt).