vision-calibration-py

Python bindings for calibration-rs.

This crate exposes high-level calibration workflows from vision-calibration:

• planar intrinsics
• single-camera hand-eye
• rig extrinsics
• rig hand-eye
• laserline device
• scheimpflug intrinsics

Build locally

maturin develop -m crates/vision-calibration-py/Cargo.toml

Python package

The Python package name is vision_calibration.

import vision_calibration as vc

print(vc.__version__)

# Build Python-native dataset/config objects with docstrings:
obs = vc.Observation(
    points_3d=[(0.0, 0.0, 0.0), (0.1, 0.0, 0.0), (0.1, 0.1, 0.0), (0.0, 0.1, 0.0)],
    points_2d=[(100.0, 100.0), (200.0, 100.0), (200.0, 200.0), (100.0, 200.0)],
)
dataset = vc.PlanarDataset(views=[vc.PlanarView(observation=obs)] * 3)
config = vc.PlanarCalibrationConfig(
    max_iters=80,
    robust_loss=vc.robust_huber(1.0),
)

result = vc.run_planar_intrinsics(dataset, config)
print(result.mean_reproj_error)

Scheimpflug workflow:

import vision_calibration as vc

obs = vc.Observation(
    points_3d=[(0.0, 0.0, 0.0), (0.1, 0.0, 0.0), (0.1, 0.1, 0.0), (0.0, 0.1, 0.0)],
    points_2d=[(100.0, 100.0), (200.0, 100.0), (200.0, 200.0), (100.0, 200.0)],
)
dataset = vc.PlanarDataset(views=[vc.PlanarView(observation=obs)] * 3)
config = vc.ScheimpflugIntrinsicsCalibrationConfig(
    fix_scheimpflug={"tilt_x": False, "tilt_y": False}
)
result = vc.run_scheimpflug_intrinsics(dataset, config)
print(result.camera.sensor)

Migration: hard break to typed high-level API

High-level runner functions now require typed dataset/config objects. Raw mapping/list payloads are no longer accepted by:

• run_planar_intrinsics
• run_scheimpflug_intrinsics
• run_single_cam_handeye
• run_rig_extrinsics
• run_rig_handeye
• run_laserline_device

Before (no longer supported):

import vision_calibration as vc

result = vc.run_scheimpflug_intrinsics(
    {"views": [...]},
    {"max_iters": 80},
)

After (typed high-level API):

import vision_calibration as vc

dataset = vc.PlanarDataset(views=[...])
config = vc.ScheimpflugIntrinsicsCalibrationConfig(max_iters=80)
result = vc.run_scheimpflug_intrinsics(dataset, config)

If you still need raw serde payload control for migration/interop, use low-level helpers from vision_calibration._api (for example _run_scheimpflug_intrinsics_raw).

Runnable Python examples

Python workflow examples live in crates/vision-calibration-py/examples/ and mirror the Rust examples from crates/vision-calibration/examples/.

Install detector dependencies for real-image examples:

./.venv/bin/python -m pip install "vision-calibration[examples]"

Run all:

for f in crates/vision-calibration-py/examples/*.py; do ./.venv/bin/python "$f"; done

Run individual examples:

./.venv/bin/python crates/vision-calibration-py/examples/planar_synthetic.py
./.venv/bin/python crates/vision-calibration-py/examples/planar_real.py
./.venv/bin/python crates/vision-calibration-py/examples/stereo_session.py
./.venv/bin/python crates/vision-calibration-py/examples/stereo_charuco_session.py
./.venv/bin/python crates/vision-calibration-py/examples/handeye_synthetic.py
./.venv/bin/python crates/vision-calibration-py/examples/handeye_session.py
./.venv/bin/python crates/vision-calibration-py/examples/rig_handeye_synthetic.py
./.venv/bin/python crates/vision-calibration-py/examples/laserline_device_session.py

Notes:

• planar_real.py, stereo_session.py, stereo_charuco_session.py, and handeye_session.py run detector-based corner extraction from real images using calib-targets.
• vision_calibration.types is low-level compatibility surface for advanced interop only; prefer typed dataclasses/models for new code.