BUFFER-X

BUFFER-X

BUFFER-X is a PyTorch model for zero-shot point cloud registration across indoor, outdoor, homogeneous, and heterogeneous sensor settings.

This Hugging Face repository is intended to host the pretrained BUFFER-X snapshots. The code lives in the official GitHub repository:

https://github.com/MIT-SPARK/BUFFER-X

The repository metadata uses the MIT license to match the included LICENSE file. If you release model weights under different terms, update the YAML metadata before uploading.

Expected Files

Upload pretrained weights under the same layout used by the GitHub code:

snapshot/
  threedmatch/
    Desc/best.pth
    Pose/best.pth
  kitti/
    Desc/best.pth
    Pose/best.pth

The included upload helper preserves this layout automatically when a local snapshot/ directory exists.

Usage

Install BUFFER-X, then download the pretrained snapshots from this model repo:

git clone https://github.com/MIT-SPARK/BUFFER-X
cd BUFFER-X
./scripts/install.sh --cuda cu124 --with-hub
python scripts/download_pretrained_models.py --source hf --repo-id <this-model-repo>

Run evaluation after preparing the datasets:

python test.py --dataset 3DMatch TIERS Oxford MIT --experiment_id threedmatch --verbose

Requirements

BUFFER-X inference uses CUDA-specific dependencies, including pointnet2_ops, KNN_CUDA, custom C++ wrappers, and torch-batch-svd. The GitHub installation script installs these pieces for supported PyTorch/CUDA combinations.

Limitations

• The hosted pretrained snapshots do not include benchmark datasets.
• ScanNet++ preprocessing must be run from the original dataset because modified files cannot be redistributed by this project.
• CPU-only installation is useful for reading utilities and packaging checks, but full BUFFER-X inference requires CUDA extensions.

Citation

@article{Seo_BUFFERX_arXiv_2025,
  title={BUFFER-X: Towards Zero-Shot Point Cloud Registration in Diverse Scenes},
  author={Minkyun Seo and Hyungtae Lim and Kanghee Lee and Luca Carlone and Jaesik Park},
  journal={2503.07940 (arXiv)},
  year={2025}
}