Skip to content

Configuration

BoxMOT uses YAML configs to make benchmark-driven workflows repeatable across both the CLI and Python API.

Config families

Config family Purpose
modes.yaml Shared defaults for track, generate, eval, tune, research, and export
benchmarks/ Thin workflow bundles that choose dataset, detector, and ReID defaults
datasets/ Dataset metadata and evaluation layout
detectors/ Detector profiles and family defaults
reid/ ReID model profiles
trackers/ Tracker runtime defaults and tuning spaces

Resolution flow

When you run:

boxmot eval --benchmark mot17-ablation --tracker boosttrack

BoxMOT resolves the benchmark config first, then loads the associated dataset, detector, ReID, and tracker configs automatically.

Why it matters

This setup lets you:

  • avoid repeating long dataset and model paths
  • reuse caches across generate, eval, tune, and research
  • keep experiment defaults in version-controlled YAML instead of shell history

Detailed config pages