Data credits and licences
Whose data, under which terms.
Our results stand on public datasets made by other people. This page credits them, says what we publish from each, and lists the software this site uses.
Maps on this site
The 3-D maps are published under CC BY 4.0. We publish maps only from data we may share.
The Rosario Dataset
CIFASIS-CONICET, Argentina. robot.cifasis-conicet.gov.ar. Licence: CC BY 4.0.
Pire, T., Mujica, M., Civera, J. and Kofman, E. "The Rosario dataset: Multisensor data for localization and mapping in agricultural environments." The International Journal of Robotics Research, 2019.
Credit on every Rosario map: "Source images: The Rosario Dataset, CIFASIS-CONICET, CC BY 4.0 (Pire et al., IJRR 2019). Changes: 3-D Gaussian-splat map reconstructed by Necessity Robotics (xclean) from the images and xclean's own live camera poses." The dataset's authors do not endorse our maps.
Our OAK-D-Lite recordings
Recorded by Necessity Robotics with its own OAK-D-Lite camera (five routes, all mapped here). Maps released under CC BY 4.0: "3-D map by Necessity Robotics (xclean)."
Benchmark datasets (numbers only)
We report our accuracy numbers on these datasets. Except for Rosario (CC BY 4.0, above), we publish no images, maps or films made from them.
| Dataset | Reference | Licence |
|---|---|---|
| EuRoC MAV | Burri, M. et al. "The EuRoC micro aerial vehicle datasets." IJRR, 2016. ETH Zurich. | In Copyright – Non-Commercial Use Permitted (ETH Research Collection, checked 2026-10-30) |
| KITTI odometry | Geiger, A., Lenz, P., Stiller, C. and Urtasun, R. "Vision meets robotics: The KITTI dataset." IJRR, 2013. | CC BY-NC-SA 3.0 (non-commercial) |
| 4Seasons | Wenzel, P. et al. "4Seasons: A cross-season dataset for multi-weather SLAM in autonomous driving." GCPR, 2020. | CC BY-NC-SA 4.0 (non-commercial) |
| MADMAX | Meyer, L. et al. "The MADMAX data set for visual-inertial rover navigation on Mars." Journal of Field Robotics, 2021. DLR. | No licence named by its authors. |
| Rosario | Pire, T. et al., IJRR 2019 (above). | CC BY 4.0 |
| Rosario v2 | CIFASIS-CONICET (one route on the board). | See the dataset's official page. |
Inside the xclean package
TartanAir (place-recognition vocabulary)
xclean's ORB vocabulary was built from 1,318 TartanAir images; the file holds only cluster centres, no image data.
Wang, W. et al. "TartanAir: A dataset to push the limits of visual SLAM." IROS, 2020. Carnegie Mellon University, AirLab. theairlab.org/tartanair-dataset. Licence: CC BY 4.0. Its authors do not endorse xclean.
xclean's runtime dependencies (NumPy, SciPy, Numba, CuPy, OpenCV, depthai, the NVIDIA CUDA libraries and others) are installed separately and keep their own licences; THIRD_PARTY.md in the repository lists each one.
Baseline systems
We ran these systems separately to compare against. None of their code is part of xclean.
- ORB-SLAM3 V1.0 (UZ-SLAMLab/ORB_SLAM3, commit 4452a3c, 2022-02-10), built on Windows with portability-only patches (headless, MSVC); tracking, mapping, loop-closing and optimiser code unmodified: Campos, C. et al. "ORB-SLAM3: An accurate open-source library for visual, visual-inertial and multi-map SLAM." IEEE T-RO, 2021. GPLv3.
- OKVIS2-X (ethz-mrl/OKVIS2-X, commit 38043e4, 2026-03-17), built on Windows with Windows-port patches; per our build notes, no patch changes its estimator, front end or dataset logic: Boche, S. et al. "OKVIS2-X: Open Keyframe-based Visual-Inertial SLAM Configurable with Dense Depth or LiDAR, and GNSS." Accepted to IEEE T-RO (arXiv:2510.04612, 2025). BSD-3-Clause.
Software on this site
| Library | Version | Licence | Used for |
|---|---|---|---|
| three.js | 0.186.1 | MIT, © 2010-2026 three.js authors | 3-D rendering |
| Spark (@sparkjsdev/spark) | 2.3.1 | MIT, © 2025 World Labs Technologies, Inc. | Gaussian-splat rendering |
| fflate (bundled in Spark) | 0.8 | MIT, © Arjun Barrett | decompression |
The libraries are served from this site, unmodified, with their licence files in vendor/. Text uses your device's system fonts; no font is downloaded. The map files are hosted on Hugging Face.
Privacy
This site sets no cookies, runs no analytics and loads no third-party trackers. Map files (the 3-D renders, films and point clouds) are fetched from Hugging Face, where they are hosted: by the landing page once it is idle, and by the 3-D viewer when you open a map.