B · Dataset laboratory
Datasets
Raw data is not redistributed here. Point-cloud views elsewhere on this site are heavily decimated excerpts with attribution.
newer_college_bag9
Newer College 2020, 01_short_experiment, bag 9 (last 27 s)
REAL consecutive
| Sensor | Ouster OS1-64 Gen1 (64 beams, 1024 columns, 10 Hz) |
|---|---|
| Licence | CC BY-NC-SA 4.0 |
| Ground truth | Yes: registered_poses.csv, 10 Hz, pose of the ground-truth body frame (differs from the sensor frame; see C). |
| Format | ROS1 bag → PCD (x y z intensity t reflectivity ring noise range) |
| Provenance | Oxford Robotics Institute, access granted by email; recording rooster_2020-03-10-11-01-34_9 extracted with tools/extract_bag.py |
| Why it is here | Same sensor class as the target (Ouster OS1-64, 10 Hz). Consecutive frames with ground truth. Only 27 s long. |
| Experiments | baseline · laptop-full · frames 0–270 (270) baseline · laptop-4c · frames 0–270 (270) baseline · laptop-1c · frames 0–270 (270) p2plane · laptop-full · frames 0–270 (270) optimized · laptop-full · frames 0–270 (270) optimized · laptop-4c · frames 0–270 (270) optimized · laptop-1c · frames 0–270 (270) baseline · laptop-full · frames 0–270 (270) optimized · laptop-full · frames 0–270 (270) |
newer_college_sparse
Newer College 2020, 01_short_experiment, thinned sample (11.3 % of scans)
REAL-NON-CONSECUTIVE NON-CONSECUTIVE
| Sensor | Ouster OS1-64 Gen1 (64 beams, 1024 columns, 10 Hz) |
|---|---|
| Licence | CC BY-NC-SA 4.0 |
| Ground truth | Not used |
| Format | PCD |
| Provenance | Same sequence, raw PCD release: an 11.3 % thinned sample (median gap 0.6 s, longest consecutive run 4 frames) |
| Why it is here | Not consecutive, so never used for odometry. Used only for single-frame preprocessing timing and point counts. |
| Experiments | baseline · laptop-full · frames 0–200 (200) |
oxford_spires
Oxford Spires, one window
REAL consecutive different sensor
| Sensor | Hesai QT64 (64 beams) - DIFFERENT from OS1-64 (beam layout, FOV, range) |
|---|---|
| Licence | see docs/decision-log.md |
| Ground truth | Yes: centimetre-level poses (LiDAR registered to a survey map), pose of the base frame |
| Format | PCD (x y z intensity timestamp ring, padded) |
| Provenance | Oxford Robotics Institute via Hugging Face (observatory-quarter-01), 300-frame window picked by a fixed rule |
| Why it is here | Long continuous walking sequence. Different sensor (Hesai QT64, ~104° vertical field of view): comparisons must say so. |
| Experiments | baseline · laptop-full · frames 0–300 (300) baseline · laptop-4c · frames 0–300 (300) baseline · laptop-1c · frames 0–300 (300) p2plane · laptop-full · frames 0–300 (300) optimized · laptop-full · frames 0–300 (300) |
Evaluated, not used
- KITTI: adapter kept (Velodyne HDL-64E, automotive). No run published; any use must state the sensor difference.
- Hilti, OpenLORIS, TUM RGB-D: adapters were inspect-only or have no LiDAR, so they were moved to archive/.
- MulRan: not evaluated.