High-Fidelity Vector-Tile Reduction for Interactive Map Exploration

HiFIVE is a visualization-aware vector-tile reduction framework built on top of BEAST / DaVinci.
It generates size-bounded Mapbox Vector Tiles (MVT) while preserving spatial structure, visual density, and attribute diversity, enabling scalable and faithful interactive map exploration of large geospatial datasets.

• Tile-level reduction with strict size constraints
• Visualization-aware feature scoring (geometry, pixels, attributes)
• Multiple reduction strategies (LP, greedy, sampling)
• Designed for large-scale Spark-based pipelines
• Compatible with BEAST vmplot

• Java 8+
• Apache Spark
• BEAST / DaVinci
• (Optional) Gurobi Optimizer (for LP-based reduction)

Below is a complete example command using LP-based reduction on the eBird dataset.

bin/beast vmplot \
  /path/to/input/ebird/features.geojson \
  /path/to/output/ebird_LP_vmplot_zoom20 \
  levels:0..20 \
  iformat:geojson \
  compact:false \
  reductionThreshold:256000 \
  reductionMethod:lp \
  'sampler:mode=sort;priority=vertices:largest;capacity=cells:1500000'