v1.0.0 — First Release

🌀 v1.0.0 — First Release

Initial public release of Stabilo-Optimize: a ground-truth-free benchmarking harness for tuning the Stabilo video/track stabilization library.

---

✨ Features

Ground-truth-free benchmarking: for each scene, a known random homography plus random photometric distortions (brightness, saturation, blur, fog) are injected, the scene is warped, and Stabilo's recovery of that warp is measured — no manually annotated correspondences needed.

Hierarchical grid search: JSON configuration files map each Stabilo hyperparameter (detector, matcher, RANSAC method, thresholds, ...) to a list of values; benchmark.py expands the full Cartesian product into individual runs, encouraging a hierarchical search strategy (fix most parameters, vary one or two at a time) over the full combinatorial space at once.

Metrics: Homography Estimation Accuracy (HEA) and Mean Intersection over Union (MIoU, requires bounding-box masks) are computed per run and aggregated into per-detector performance plots.

Result visualization: optional per-trial stabilization-process video, plus performance plots comparing detectors across the varied hyperparameter.

Sample experiment: experiments/sample_experiment ships with real scenes, vehicle bounding-box masks, and both a minimal (simple_benchmark.json) and comprehensive (comprehensive_benchmark.json) example configuration.

---

📚 Citation

If you use Stabilo-Optimize in your research, please cite it (see CITATION.cff / the README citation block).

Full Changelog: https://github.com/rfonod/stabilo-optimize/commits/v1.0.0