GeoIntel release gates / Compile, test, contracts and builds (push) Successful in 1m49s
GeoIntel release gates / Python and npm vulnerability policy (push) Successful in 21s
GeoIntel release gates / Production AI image, SBOM and container scan (push) Successful in 5m39s
GeoIntel release gates / Deploy exact gated revision to Unraid (push) Failing after 58m43s
1.5 KiB
1.5 KiB
Raster Standards
Accepted V1 raster inputs
- GeoTIFF preferred.
- TIFF only if georeferencing metadata exists or sidecar world file is provided.
- JPG/PNG only as non-georeferenced preview unless georeferencing is explicitly supplied.
Metadata required
- width;
- height;
- band count;
- dtype;
- CRS;
- transform;
- bounds in source CRS;
- bounds in EPSG:4326 where possible;
- nodata value;
- resolution;
- file size;
- checksum/hash.
Tile defaults
- tile size: 512x512 pixels;
- overlap: 64 pixels for detection/segmentation;
- padding strategy: reflect or constant nodata, recorded in parameters;
- tile IDs must be stable and reproducible.
Derived raster outputs
Derived rasters must reference:
- parent dataset;
- operation;
- parameters;
- timestamp;
- CRS;
- nodata handling.
Raster statistics
V1 statistics:
- min;
- max;
- mean;
- std where feasible;
- nodata count;
- histogram bins for display.
Reprojection
Reprojection must not overwrite originals. It creates a derived dataset.
Clipping
Raster clipping to Area creates a derived raster and stores the clipping geometry reference.
AI inference preparation
Before inference:
- dataset must be georeferenced;
- CRS must be known;
- transform must be available;
- tiling parameters must be stored;
- band selection must be explicit.
Non-goals for V1
- perfect Sentinel processing pipeline;
- cloud optimized GeoTIFF production;
- large-scale distributed raster processing;
- LiDAR-derived raster generation.