Artifacts
All files are written to the slamlab_core.artifacts.dirpath directory:
| File | Description |
|---|---|
slamlab-config.yaml | Copy of the input configuration |
slamlab-log.txt | Per-point-cloud processing log |
slamlab-update-slam-results.csv | Per-point-cloud metrics (runtime, iterations, correspondences, etc.) |
slamlab-step-timings.csv | Per-point-cloud per-step runtimes in milliseconds |
slamlab-loop-closure-attempts.csv | Loop closure attempts with their status and numbers of inliers and correspondences (if loop_closure_detection is enabled) |
trajectory.poly | Trajectory in POLY format (can be opened with CloudCompare) |
trajectory_kitti.txt | Trajectory in KITTI format |
trajectory_tum.txt | Trajectory in TUM format |
trajectory_minimal.csv | Minimal trajectory CSV with poses only |
trajectory_complete.csv | Trajectory CSV with poses and, if estimate_pose_covariances is enabled, pose covariances (ordered [rx, ry, rz, tx, ty, tz] in the local pose frame) |
trajectory_relative_poses.csv | Relative poses between consecutive states |
trajectory.geojson | Trajectory as GeoJSON |
clouds_map_frame/ | Deskewed point clouds in map frame (optionally tiled and filtered), as LAZ files |
prior_map.voxelpca, prior_map.laz | Prior map generated from the saved point clouds (if save_clouds.generate_prior_map is enabled) |
slamlab-pipeline-metadata.yaml | Pipeline metadata |
The original (raw, non-deskewed) sensor-frame clouds are written to the separate slamlab_core.clouds.dirpath directory; the map-frame clouds are derived from them by deskewing with the final estimated trajectory (densified with IMU measurements when available).
Point cloud file formats
Section titled “Point cloud file formats”| Files | Format |
|---|---|
| Map-frame clouds, tiles, prior map | LAZ (.laz) |
Original clouds in slamlab_core.clouds.dirpath, debug clouds | PLY (.ply) |
- LAZ: compressed LAS 1.4, approximately 4 times smaller than PLY. The coordinates have a resolution of 0.1 mm; attributes and point stamps are lossless. CloudCompare, PDAL, LAStools and laspy can open the files.
- PLY: binary and lossless. The SLAM thread writes these files and the pipeline reads them again, so they use the fastest format.
- Attributes in LAZ files: an attribute uses the standard LAS field of the same name (for example
intensity) if it has one row, an integer type and values in the range of the field. All other attributes are LAS Extra Bytes with the attribute name (scalar fields in CloudCompare). The point stamps are the Extra Bytessecandnsec. - Tiles: the tiles are written as PLY while the clouds are saved and converted to LAZ at the end.