se0yeon00/Const_pseudo_dataset

Const Pseudo Dataset

Prepared training, validation and test data for Chamelion: Reliable Change Detection for Long-Term LiDAR Mapping in Transient Environments.

Paper · Project · Pretrained weights

Download

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="se0yeon00/Const_pseudo_dataset",
    repo_type="dataset",
    local_dir="Const_pseudo_dataset",
)

Contents

Split	Submaps	Scans
Train	62	5,877
Validation	12	1,158
Test	2	2,296
Total	76	9,331

The data occupies approximately 3.61 GiB. Training and validation use pseudo labels. The test split contains Const-1F (1,124 scans) and Lab (1,172 scans). Train and validation use separate source sessions. Keep the supplied splits when training and evaluating.

File structure

Const_pseudo_dataset/
├── splits/
│   ├── train.txt
│   ├── val.txt
│   └── test.txt
├── checksums.json
└── sequences/sequence_000/
    ├── poses.txt
    └── submaps/submap_000/
        ├── prior_map.pcd
        ├── scans/000000.pcd
        ├── scan_labels/000000.label
        └── map_labels/static.label

Each split lists submap paths relative to sequences/. Each submap contains one prior map, its fixed label file, and scans with matching label files.

• Clouds use global coordinates. Each pose is a row-major 3×4 local-to-global matrix.
• A scan filename identifies its zero-based row in poses.txt. Preserve frame IDs and gaps.
• Labels are binary int32, with one value per point: 0 static, 1 added, 2 removed and -1 ignored. Test label 251 denotes highly dynamic points excluded from evaluation; it is not a change class.
• sequence_005 is Const-1F; sequence_007 is Lab. Lab starts at frame 001713; use the supplied pose table without renumbering frames.
• checksums.json provides SHA-256 hashes and sizes for 18,821 data files.

Pseudo labels may contain segmentation, tracking and placement errors.

License

The dataset is licensed under CC BY-NC-ND 4.0. Attribution is required. Commercial use and sharing adapted material are not permitted under this license.

Citation

@article{jang2026chamelion,
  title   = {Chamelion: Reliable Change Detection for Long-Term LiDAR Mapping in Transient Environments},
  author  = {Jang, Seoyeon and Lee, Alex Junho and Nahrendra, I Made Aswin and Myung, Hyun},
  journal = {IEEE Robotics and Automation Letters},
  year    = {2026},
  doi     = {10.1109/LRA.2026.3665079}
}