The dataset could not be loaded because the splits use different data file formats, which is not supported. Read more about the splits configuration. Click for more details.
Error code: FileFormatMismatchBetweenSplitsError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
ODL-EGO S1 | Stereo Calibration Validation
REAL-WORLD DATA ENGINEERING · TECHNICAL EVIDENCE SERIES
Stereo calibration consistency · Epipolar geometry · Capture-mode sensitivity
Explore GitHub Evidence · Read Technical Field Note · Origin Data Lab
Technical evaluation artifacts · Origin Data Lab · Field Note #04 · October 2026
Scope: This is a compact calibration evaluation artifact collection, not a downloadable video dataset, benchmark leaderboard, or production-certified calibration. The data describes a particular ODL-EGO S1 sample and sensor configuration.
Technical source & inspection tools (GitHub) · Origin Data Lab · Contact
Key results at a glance
| 0.408 px | 1.685 px | 29 pairs | 19,386 matches |
|---|---|---|---|
| Median angular epipolar error | p90 angular epipolar error | Sampled stereo pairs | Unfiltered matches |
Reported metrics for the specified evaluated configuration, not generalized camera accuracy or production certification.
At a glance
| Evaluation item | Reported result | Interpretation |
|---|---|---|
| Capture mode evaluated | 1600 × 1200, 30 fps | Mode-specific results |
| Original Kalibr solution | 800 × 600 | Calibration adapted to the evaluated mode |
| Stereo pairs sampled | 29 | Evaluation sample, not a full deployment survey |
| Unfiltered stereo matches | 19,386 | No RANSAC filtering reported |
| Angular epipolar error, median | 0.408 px | Angular metric expressed in center-equivalent pixels |
| Angular epipolar error, p90 | 1.685 px | Distribution has a nontrivial tail |
| Held-out median, supplied rotation | 0.399 px | As reported by source validation |
| Held-out median, after fitted 3-DoF correction | 0.374 px | Fitted on evaluation setup; not a new certified calibration |
Important: The 0.408 px angular metric is not interchangeable with rectified vertical residual or a metric depth error. A mode-mismatched 16:9 comparison in Field Note #04 reports a different error distribution; do not combine measurements from different capture configurations.
Visual evaluation
1. Epipolar consistency across the image
Figure: spatial error pattern for the tested stereo configuration. Edge and field-angle behavior must be evaluated separately from the overall median.
2. Capture-mode matching matters
Figure: a comparison of matching and nonmatching sensor modes. This is not an estimate of generic camera accuracy.
3. Sensitivity to small rotation changes
Figure: sensitivity of the reported metric to deliberately perturbed stereo rotation.
4. Different error metrics versus field angle
Figure: angular epipolar error and rectified residuals are different quantities.
5. Fisheye inverse-model behavior
Figure: tested OpenCV fisheye unprojection behavior in the specified setup; it is not a universal claim about library correctness.
Downloadable evaluation artifacts
| Folder | Contents |
|---|---|
03_Validation/ |
Original JSON summaries and source PNG evaluation figure |
figures/ |
Five WebP figures for review and citation |
00_Report/ |
Full calibration validation report (PDF) |
data/ |
Compact CSV table derived from the original JSON summaries |
SOURCE_AND_LIMITATIONS.md |
Provenance, metric definitions, cautions |
The original Kalibr YAML outputs, Python inspection scripts, and source-package checksum manifest are maintained in the GitHub repository. This Hugging Face release is intentionally organized for discovery, inspection, and evaluation.
What this release does not validate
- Metric stereo depth accuracy or independently verified baseline accuracy.
- Camera-to-IMU extrinsics or timing to production tolerances.
- Generalization to other lenses, resolutions, crops, devices, or capture modes.
- Universal rectification quality or a universal bug in OpenCV.
- End-to-end independent reproduction of the original calibration run from raw recordings.
The values and figures are published as supplied in the accompanying technical package. Packaging them here is not an independent rerun or certification.
Technical field notes
These article links are intended public URLs; confirm site deployment before relying on them.
Rights and permitted use
Public visibility does not grant an open-data license. See LICENSE.md. For redistribution, incorporation into products, training, or commercial use, obtain written permission from Origin Data Lab: contact@origindatalab.io.
Suggested citation: Origin Data Lab (2026). ODL-EGO S1 Stereo Calibration Validation Artifacts, v1.0. Technical Field Note #04. GitHub: https://github.com/origin-data-lab/stereo-calibration-validation
- Downloads last month
- 45