Open MVIS

v0.1 · public benchmark

Calibrate a whole rig at once

CamCalib — the calibration library behind Open MVIS — solves camera intrinsics and distortion, extrinsics, time offsets, rolling-shutter readout, board poses, camera–IMU transforms, IMU intrinsics, and biases in a single GTSAM factor graph. Pure Python; pip install; Windows, macOS, and Linux.

camcalib calibrate --config config.yaml

Six camera models

radtan, equidistant (KB4), fov, double_sphere, eucm, and experimental omni_radtan — mixable across cameras in one rig, in both camera-only and camera–IMU calibration.

Multi-IMU

Several IMUs in one graph, each with its own extrinsics, time offset, and intrinsics. Per-IMU gyro-only mode for sensors whose accelerometer you can’t trust.

It checks its own work

camcalib roundtrip applies a calibration to its own data and recalibrates. PASS needs IMU intrinsics to collapse to identity and the stereo baseline to hold — far stronger than a low residual.

Reports you can send

One self-contained HTML file: intrinsics tables, 3D extrinsics, error analysis, motion-excitation grades, detection statistics, timing.

No ROS needed

ROS1 and ROS2 bags read through rosbags, plus EuRoC/ASL, image folders, and video — on a machine that has never had ROS installed.

Public leaderboard

Filterable by sensor configuration and algorithm family. Submit by pull request and cite it as Open MVIS leaderboard, vX.Y.

Main features

• Single- and multi-camera calibration — joint optimisation of intrinsics, distortion, extrinsics, time offsets, rolling-shutter readout, and board poses
• Camera–IMU calibration — T_CtoI, td, readout, IMU intrinsics, and biases, for single or multi cam/IMU rigs and all six distortion models
• Batch sweeps — one config, many sequences (input.bags), each into its own result; the basis of the repeatability study
• Three IMU-intrinsics conventions — kalibr, rpng, calyx
• Gyro↔accel offset t_ga — measures a driver’s accel-interpolation artefact; the T265 reads −6.2 ms, stable to ~0.2 ms across bags
• Corrected-dataset export — camcalib apply-calib re-records a raw bag with the calibration applied
• Round-trip validation — camcalib roundtrip, PASS/FAIL, CI-usable
• Boards — chessboard, AprilGrid (Kalibr 2-bit border), ArUco
• Kalibr interop — read and write camchain / camchain-imucam
• Regression harness — manifest-driven runs with dashboards and baselines

Validated

228 runs across 65 dataset families, zero failures. Camera parameters match the published TUM-VI and Intel T265 calibrations to under 0.6% on focals; on TUM-VI raw IMU bags the camera–IMU path reproduces Kalibr’s extrinsics to ≤2 mm lever, ≤0.15 mm stereo baseline, ≤1 ms time offset.

Cite

@misc{openmvis,
  title  = {{Open MVIS}: A Benchmark for Multi-Visual-Inertial Sensor Calibration},
  author = {{Open MVIS Contributors}},
  year   = {2026},
  url    = {https://openmvis.com}
}