Harris Census Block Group - Building Instance Population

This dataset contains building-level records for Harris County where each building instance is linked to a Census Block Group (GEOID) and assigned an estimated population label derived from demographic ratios and occupancy assumptions. It combines geometric/location attributes (building polygons or centroids), structural/context features (building type, estimated units, parcel-informed classification), and CBG-level demographic context (ppu_*, occupancy_rate) to create ML-ready supervision targets. The core table for modeling is ml_dataset/labels.csv, which ties each building_uid to image crop/mask paths and a numeric estimated_population target. Companion outputs include building-level estimate files (harris_county_building_population_estimates.csv, harris_county_buildings_with_population.geojson) and a CBG-level reference/validation table (harris_county_cbg_validation.csv) keyed by GEOID. Conceptually, it is a spatially joined, feature-engineered population estimation dataset where many buildings map to the same CBG, enabling both geospatial analysis and supervised learning.