UAV Macro Tiles (JP2) and Vector Ground Truth for Semantic Segmentation (Buildings/Roads/Water)

This release contains macro-split UAV orthomosaic raster tiles and corresponding vector ground truth prepared for semantic segmentation. Raw UAV orthomosaic data were provided by IIIT-T; the creator prepared the derived macro tiles and ML-ready vector labels as a participant representing IIIT-H. Files: - Rasters.zip: lossless JP2 macro tiles (4x4 macro grid per AOI; 128 tiles total). Tile naming: _rc.jp2 - Vectors.zip: per-tile GeoJSON ground truth aligned to the macro tiles. File naming: _rc.geojson Vector schema (GeoJSON properties): - group, class, class_id, geometry Class mapping for mask generation (background is implicit): - 0: background (implicit) - 1: building_1 - 2: building_2 - 3: building_3 - 4: building_4 - 5: road (merged) - 6: water (merged) Notes: - Roads and waterbody subtypes were merged into single classes for robust segmentation. - Building rooftop subtypes (1..4) are retained. Embargo: - This deposition is under embargo until 2026-05-30. Versioning: - Future updates may publish newer versions of Vectors.zip while keeping the raster macro tiles unchanged.