MBTiles Multiclass Mask Builder

Use this tool when you have an MBTiles file that defines the reference grid, one GeoJSON with frame geometries, and one GeoPackage with class polygons. The builder writes a single multi-class GeoTIFF per frame and a dataset.csv manifest.

This is the right tool when the mask resolution must match the highest zoom available in the MBTiles source. The MBTiles file defines the target CRS, transform, width, height, and pixel spacing.

Inputs

Input	Role
reference_mbtiles_path	Reference raster grid used to derive resolution and alignment
frames_path	Frame polygons that define each output mask extent
vector_path	Polygon labels with integer class ids

Output

output_dir/
  masks/
    10.tif
    11.tif
  dataset.csv

The CSV includes the generated mask path, dimensions, CRS, transform, and the paths used to create the manifest.

CLI

pytorch-smt-tools build-mbtiles-multiclass-masks /data/config.yaml

Example YAML

conf/examples/mbtiles_multiclass_mask_builder.yaml

reference_mbtiles_path: /data/tiles.mbtiles
frames_path: /data/locais_mascaras.geojson
vector_path: /data/dsg_masks.gpkg
output_dir: /data/output_masks
frame_layer: locais_mascaras
vector_layer: dsg_masks
frame_id_attribute: rect_id
class_attribute: tipo
background_value: 255

Notes

• background_value: 255 keeps background separate from class ids starting at zero.
• n_workers controls how many frames are processed in parallel. Use 1 to force sequential execution when debugging.
• progress: true enables a tqdm progress bar for long runs.
• Frames and vector polygons must be in a CRS compatible with the MBTiles reference grid.
• Each output mask is written as a single-band uint8 GeoTIFF.