Benchmark protocol

Evaluation

A placeholder for the versioned RoofVIP evaluation specification.

Protocol status

The benchmark dataset RoofVIP v1.0.3.0 is released, but the formal evaluation protocol is not yet released. Metric definitions shown in future drafts should be treated as provisional until a protocol version is assigned.

Planned specification

The released protocol should state, at minimum:

  • benchmark task and admissible model inputs;
  • exact RoofVIP dataset version used for evaluation;
  • train/validation/test split definition;
  • prediction confidence and filtering rules;
  • reference-to-prediction matching procedure;
  • point, line, polygon, and topology metrics where applicable;
  • geometry conversion rules for methods with different native outputs;
  • leaderboard ranking criterion and tie handling;
  • submission format and reproducibility requirements.

Version binding

A benchmark result should identify both a dataset release and a benchmark protocol version. This lets RoofVIP evolve without making historical leaderboard entries ambiguous.