Калибровка камер с метриками L2 и репроекции

★ 7.6 · content

amc-run-sample-calibration is a Claude Code skill that runs end-to-end calibration on the bundled synthetic dataset sdg_08_2_sample_data_010926.zip — four warehouse cameras with ground truth — against a running AutoMagicCalib microservice. It scans ports 8000–8009 to detect the backend, idempotently extracts the archive into a cache directory, executes the bundled Python script (which self-heals by creating a throwaway venv if requests is missing), and reports L2 distance and reprojection error metrics; VGGT refinement is attempted automatically when the project reports vggt_state: READY. If no backend or sample data is found, the skill describes the expected API workflow rather than fabricating outputs. Use it to verify a freshly launched AMC stack before moving on to real video files or live RTSP streams.