Thanks for helping improve the project.
- Never commit real images, labels, model weights, credentials or absolute local paths.
- Keep source labels read-only and write experiment results below a separate output directory.
- Add or update a test when changing IoU, class mapping, thresholding, candidate deduplication or label writing.
- Prefer small, reviewable changes over a complete rewrite of the workflow.
- Keep generated files out of the repository.
python -m py_compile autolabel_with_single_class_models.py
python scripts\validate_docs.py
ruff check autolabel_with_single_class_models.py yolo_label_recovery tests examples scripts platform\tools
python tests\run_smoke_tests.py
pytestBefore submitting a change, run a dry-run on a tiny synthetic dataset and inspect summary.txt, CSV files and at least one review image per class.
Changes under platform/backend must pass mvn test. Changes under platform/frontend must pass npm ci and npm run build. Behavior changes should update the relevant public fixture, documentation entry and CHANGELOG.md.