Change3D (SLPCCD)

Notes on the Dataset

  • The corresponding paper has been retracted due to errors in the evaluation. A later paper from 2023 again describes the dataset, enriched by synthetic data. However, this second version of the dataset can not be found anywhere.
  • The paper speaks of "over 78 [...] point cloud pairs", but the dataset actually contains exactly 78 and for 2020 one additional scan with prefix "15_" that is missing in 2016.
  • For computing the label distribution, we extracted cylindrical environments around the POIs and removed the groundplane, as described in the paper. Wang et al. (2023) already provide the data with annotated change labels, which they refer to as the new dataset SLPCCD.
  • The acquisition is unclear. The paper states that "The 3D data from CycloMedia are generated from depth maps instead of original LiDAR scans", but the website and the second paper speak of "vehicle mounted LiDAR sensors".

Detailed Information

General Information
NameChange3D (SLPCCD)
Release Year2021
Terms of UseUnclear
Access RequirementsNone
Dataset Size7.6 GB | Partial download is possible, i.e. the data is split into several files (e.g., epochs or data types)
DocumentationMulti-paragraph description of the dataset, e.g., in the form of a paper section or a comprehensive readme file
CodeYes
ApplicationsBuilt environment change detection and classification
Detailed ApplicationsPer-object point cloud change detection
Acquisition
Number of Scenes78
Number of Epochs per Scene (minimum)2
Number of Epochs per Scene (median)2
Number of Epochs per Scene (maximum)2
General Scene TypeUrban
Specific Scene TypeStreet segment
LocationSchiedam (Netherlands)
Acquisition TypeMobile laser scanning | Educated guess by the authors
Acquisition DeviceUnclear
Acquisition PlatformCar
Scan IntervalYears
Acquisition Months
Representation
Data RepresentationUnstructured, globally aligned (e.g., point cloud or ray cloud)
Specific Data RepresentationPoint cloud
File Format/EncodingLAZ
Raw Data-
Additional Data-
Coordinate System28992
Quality and Usability
RegistrationFinely registered
Number of Partial Epochs21.80%
Unusable Data Reason-
SplitsYes
Per-Point Attributes
Intensity/ReflectivityNo
ColorYes
Semantic Labels-
Instance LabelsNo
Change Labels5 | The labels can be computed from other data/mapped to the point cloud, i.e., it can not trivially be done during point cloud construction (at least not for all epochs)
Statistics
Number of Points per Epoch (minimum)566K
Number of Points per Epoch (median)710K
Number of Points per Epoch (maximum)1M
Avg. Point Spacing (minimum)4.8mm
Avg. Point Spacing (median)9.8mm
Avg. Point Spacing (maximum)1.7cm
Avg. Change Points per Epoch1.43%

Paper Reference 1

@article{ku2021datasetshrec,
    author = {Tao Ku and
                  Sam Galanakis and
                  Bas Boom and
                  Remco C. Veltkamp and
                  Darshan Bangera and
                  Shankar Gangisetty and
                  Nikolaos Stagakis and
                  Gerasimos Arvanitis and
                  Konstantinos Moustakas},
    title = {SHREC 2021: 3D point cloud change detection for street scenes},
    journal = {Comput. Graph.},
    volume = {99},
    pages = {192--200},
    year = {2021},
    doi = {10.1016/J.CAG.2021.07.004},
}

Paper Reference 2

@article{wang2023datasetslpccd,
    author = {Zhixue Wang and
                  Yu Zhang and
                  Lin Luo and
                  Kai Yang and
                  Liming Xie},
    title = {An End-to-End Point-Based Method and a New Dataset for Street-Level
                  Point Cloud Change Detection},
    journal = {IEEE Trans. Geosci. Remote. Sens.},
    volume = {61},
    pages = {1--15},
    year = {2023},
    doi = {10.1109/TGRS.2023.3295386},
}