Chamelion pretrained weights

Pretrained Chamelion model for LiDAR change detection with a prior map.

Paper · Project · Dataset

Download

from huggingface_hub import hf_hub_download

checkpoint = hf_hub_download(
    "se0yeon00/Chamelion", "chamelion_pretrained.pt", local_dir="pretrained"
)
settings = hf_hub_download(
    "se0yeon00/Chamelion", "inference.yaml", local_dir="pretrained"
)

chamelion_pretrained.pt is a tensor-only PyTorch state dictionary for inference, without optimizer state. The network voxel size is 0.1 m. Use the supplied inference.yaml with Chamelion's evaluation and inference tools. checksums.json contains file hashes.

Code

We thank MapMOS and TRAVEL for their open-source code.

License

The pretrained weights are licensed under CC BY-NC-ND 4.0. Attribution is required. Commercial use and sharing adapted material are not permitted under this license. The code has a separate 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}
}