List of Datasets

Automotive

SemanticKITTI

Sequential lidar semantic segmentation and semantic scene completion

Point-wise semantic annotation extension of all 22 KITTI Odometry sequences, with 28 native classes, temporal evaluation and future-scan-derived completion targets; test labels are withheld.

Metadata: Complete Metadata curator reviewed Released 2019

Primary places to inspect, download, read about, or reproduce the dataset.

Quick facts

Vehicle type
Automotive
Environment
Outdoor
Data origin
Real-world
Sensor count
1
Ground truth
Available
Calibration
Parameters Only
Annotations
Available

Sensor overview

High-level sensor availability before the detailed sensor records below.

Cameras

Not reported

IMUs

Not reported

GNSS

Not reported

LiDAR

1 reported

Magnetometers

Not reported

Barometers

Not reported

UWB

Not reported

Ground truth

Reference scope, method, rate, coverage, and provenance.

Scope Method Reference system Count Rate Coverage Reference provenance
External
Measured using equipment independent of the platform’s normal onboard sensor pipeline.
Derived
Computed primarily from the dataset’s own recorded sensors.
Hybrid
Combines independent reference equipment with onboard measurements.
Other Manual annotation Not reported Not reported 10 Hz Partial · Public Semantic Labels for Training Sequences 00–10; Test Labels 11–21 Withheld. Reference provenance
External
Measured using equipment independent of the platform’s normal onboard sensor pipeline.
Derived
Computed primarily from the dataset’s own recorded sensors.
Hybrid
Combines independent reference equipment with onboard measurements.
Derived
Accuracy and notes
Accuracy notes
Second-pass label verification, not numerical geometric accuracy.
Reference notes
Point-wise semantic reference on recorded lidar scans. SLAM/loop closure and overlapping spatial tiles support consistent human labeling. Native 28 classes; task mappings use 25 temporal or 19 single-scan classes.
Other Other Not reported Not reported Not reported Partial · Public Completion Targets for Training/Validation; Test Targets and Never-Observed Voxels Excluded. Reference provenance
External
Measured using equipment independent of the platform’s normal onboard sensor pipeline.
Derived
Computed primarily from the dataset’s own recorded sensors.
Hybrid
Combines independent reference equipment with onboard measurements.
Derived
Accuracy and notes
Accuracy notes
Aggregated and voxelized lidar geometry, not externally surveyed scene truth.
Reference notes
Semantic completion voxel references from future labeled scans: 256×256×32 cells at 0.2 m; majority-vote semantics, empty cells and visibility masks. 19130 training/815 validation/3992 test pairs. Future information is a target, not permitted current-scan input.

Calibration and synchronization

Reported calibration level, reproducibility signals, and supporting notes.

Level
Parameters Only
Processed parameters
Available
Raw calibration data
Not reported
Calibration targets
Not reported

Parameters only

Processed calibration parameters are reported, but raw calibration data is not reported.

Notes

Calibration files are obtained with the underlying KITTI Odometry data. SemanticKITTI supplies label/pose/completion products; complete raw calibration is not established.

Known or intentional limitations

Reported constraints and characteristics to check before using the dataset.

Underlying KITTI Odometry scans and calibration must be downloaded separately from the SemanticKITTI label packages.

Public training labels cover 00–10; test labels for 11–21 are withheld for benchmark evaluation.

Class imbalance and sparse distant returns challenge segmentation.

Completion geometry and annotation-support poses are derived from the recorded scans, not independently surveyed scene or trajectory truth.

Cameras

Not reported

IMUs

Not reported

GNSS

Not reported

LiDAR

Velodyne HDL-64E

Model or name
Velodyne HDL-64E
Count
1
Dimensions
3
Rate
10 Hz
Channels
Not reported
Range
Not reported

Additional sensors

Not reported

Annotations

Reported annotation availability and task support.

Availability
Available
Format
Per-scan uint32 labels: lower 16 bits semantic class, upper 16 bits temporally consistent instance ID. Voxel completion includes semantic labels and invalid/occlusion masks.
Class count
28
Semantic segmentation Available
Instance segmentation Available
Object detection Not reported
Optical flow Not reported
Depth ground truth Not reported

Citation

Citation key and BibTeX kept at the end of the page for reference.

Citation key
behley2019semantickittiv3

BibTeX

@misc{behley2019semantickittiv3,
  title={SemanticKITTI: A Dataset for Semantic Scene Understanding of LiDAR Sequences},
  author={Behley, Jens and Garbade, Martin and Milioto, Andres and Quenzel, Jan and Behnke, Sven and Stachniss, Cyrill and Gall, Juergen},
  year={2019},
  eprint={1904.01416},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/1904.01416v3},
  note={Analyzed version 3, dated 16 August 2019}
}

Dataset corrections

Notice a specific issue after reviewing the page?

Report an inaccuracy