Skip to content

Leaderboard

Calibration methods ranked across sensor configurations and algorithm families. Each entry reports reprojection error, extrinsic accuracy against the Open MVIS consensus baseline, time-offset error, runtime, and convergence rate — with mean ± stddev over N runs where the submitter provided them.

5 entries
Method Family Sensor config Dataset Reproj.
(px)
Rot err
(deg)
Trans err
(cm)
Time offs.
(ms)
Runtime
(s)
Conv.
(%)
Links
Basalt vmaster B-spline 1 IMU + 1 cam
1 IMU · 1 cam
open_mvis_4imu3cam v1.0 0.49 0.12 0.22 12.5 58 90
CamCalib v0.4.2 Factor graph Multi IMU + multi cams
4 IMU · 4 cam
open_mvis_4imu4cam v1.0 0.41 ± 0.03 0.08 ± 0.02 0.12 ± 0.04 0.31 ± 0.05 142 ± 8 100
Kalibr v1.0 B-spline 1 IMU + 1 cam
1 IMU · 1 cam
open_mvis_4imu3cam v1.0 0.52 0.15 0.28 0.04 420 100
MVIS v1.0 Factor graph Multi IMU + multi cams
4 IMU · 3 cam
open_mvis_4imu3cam v1.0 0.38 ± 0.04 0.07 ± 0.02 0.1 ± 0.03 0.28 ± 0.04 178 ± 12 100
OpenVINS v2.6 EKF 1 IMU + 1 cam
1 IMU · 1 cam
open_mvis_4imu3cam v1.0 0.61 0.18 0.34 0.45 38 95

Filter state is preserved in the URL, so a specific comparison can be cited directly — for example /benchmark/leaderboard/?config=multi_imu_multi_cam&sort=timeoffset_err_ms.

  • Lower is better for every error metric; higher is better for convergence rate.
  • means not reported, not zero. Absence is shown rather than filled in.
  • Compare within a sensor configuration, not across. A single-IMU single-camera result and a four-IMU result are different problems.

How the categories are defined, how scores are computed, and what the baseline is: Methodology.

Add your method with a pull request — one YAML file, schema-validated in CI. See Submit results.