Contents

Evaluation pipeline

tests/regression/evaluator/:

  • trackeval_runner.py: wraps the TrackEval Python module. Builds per-benchmark sequence maps, configures MotChallenge2DBox dataset, runs HOTA, CLEAR, Identity metrics.
  • report_generator.py: print_trackeval_table(results, compare) formats the per-tracker × per-sequence table; export_trackeval_results(results, dir) writes JSON/CSV.
  • run_eval.py: standalone CLI for TrackEvalRunner (independent of the compiled binary).

st eval automatically chains regression_test → TrackEval; pass --eval-only to skip the C++ binary and re-evaluate cached predictions, or --trackeval to invoke the upstream third_party/TrackEval/scripts/run_mot_challenge.py script.